×
Evaluate Your Team's AI Readiness

Take our official assessment and explore training paths to build secure, compliant AI workflows.

Access Portal

AI at Work Podcast

Ep #25 When Your Team Says Claude But Means GPT

AI at Work, AI in Business

Audio episode

Listen to this episode

The AI industry loves throwing around the word ‘agents,’ but most teams are still stuck in basic prompting mode. Kevin and Eli cut through the semantic noise to reveal what actually matters: sophisticated automation is now accessible through natural language, not technical configuration.

In this episode, they explore the practical reality of moving from one-off prompts to systematic workflows, why the ‘agent’ versus ‘automation’ debate misses the point, and how natural language interfaces are removing technical barriers that used to require specialized workflow knowledge.

Key topics covered:

✅ Why most people are still just prompting instead of building workflows

✅ How natural language makes complex automation accessible

✅ The practical difference between projects, automations, and agents

✅ Real examples of workflow automation without technical expertise

✅ Why focusing on results beats debating terminology

✅ Moving from ChatGPT tabs to systematic AI integration

This isn’t about the latest AI hype – it’s about practical transformation that works Monday morning.

TIMESTAMPS:

00:00 — Future of AI and robotics discussion

08:16 — Current state of enterprise AI adoption

16:30 — Job displacement and economic impact

25:40 — Moving beyond basic prompting

35:20 — Context and platform lock-in

42:30 — Agents vs automations semantics

52:00 — OpenAI agents vs Claude workflows

58:30 — Real-world automation examples

Explore practical AI implementation: https://assessment.ascendlabs.ai/

Book a conversation with Kevin: tidycal.com/kevinwilliams

Show Notes & Links: https://www.spark6.com/podcast

Submit listener questions:

[email protected]

[email protected]

Check out Kevin’s stuff:

AscendAI: https://ascendlabs.ai/

Follow Kevin on LinkedIn: https://www.linkedin.com/in/kevinguywilliams/

Check out Eli’s Stuff:

SPARK6 Agency: https://www.spark6.com/

Sign up for FREE AI Framework Friday Newsletter: https://www.spark6.com/newsletter

Follow Elijah on LinkedIn: https://www.linkedin.com/in/elijahszasz/

Transcript and citation details

Transcript

00:00 Play from here Kevin, what's up, homie?

00:13 Play from here Hey, man.

00:14 Play from here Good morning.

00:15 Play from here Good morning.

00:16 Play from here Kevin, I've got a problem.

00:17 Play from here Let me tell you what my problem is.

00:18 Play from here I'm old, but I have young children.

00:20 Play from here When you're old, you just really need your sleep.

00:23 Play from here Man, when I was young, I didn't care if I had a bad night's sleep.

00:26 Play from here And when I have a bad night's sleep now, I feel like it just ruins my entire life.

00:30 Play from here And I had this very short-lived fantasy about this AI super being just taking care of my

00:37 Play from here small child in the middle of the night.

00:39 Play from here And I made a passing comment to my wife of that show Westworld, where there's people,

00:44 Play from here you can't even tell if they're humans or not.

00:46 Play from here Isn't that weird that someday we're going to have these humanoid robots helping us take

00:51 Play from here care of our kids?

00:53 Play from here And she looked at me completely serious.

00:55 Play from here I have so much more faith in humanity that we would never let that happen.

00:58 Play from here I don't think that'll ever happen.

01:00 Play from here Do you think that's going to happen?

01:01 Play from here Oh, that's 100% going to happen.

01:02 Play from here That's 100% going to happen, right?

01:03 Play from here 100% I think that's going to happen.

01:05 Play from here Because everything's happening so much faster than we initially thought it was going to

01:08 Play from here with artificial intelligence, when I saw a show like Westworld, or even her, or any of

01:14 Play from here these sci-fi movies, I'm not going to say Terminator because I do have a little bit

01:18 Play from here of faith in humanity, but I would think like, oh, wow, yeah, yeah, yeah, I get how all the

01:23 Play from here pieces come together.

01:25 Play from here This is plausible.

01:26 Play from here Like, wow, maybe my, you know, something my great grandkids will see.

01:30 Play from here And now we're kind of living in the middle of the on-ramp to superintelligence.

01:34 Play from here Like how long, if you had a guess, how long is that going to be until we have much better

01:39 Play from here models inside the bodies of figures that you can't distinguish from humans?

01:43 Play from here Oh, boy, that's a big question.

01:46 Play from here I mean, right now, you already have.

01:48 Play from here But it's like, but when you think about it, right, all of these technologies are concurrently

01:53 Play from here being worked on in tandem, and at some point they're going to merge.

01:57 Play from here And we're already seeing it with models, with, you know, these research models, with biotech

02:03 Play from here models, with things that are specific to medicine now, and then all of the things happening

02:08 Play from here in robotics.

02:09 Play from here And so we have all these things happening in tandem.

02:12 Play from here And then they're just going to start merging until I wake up at one in the morning and

02:18 Play from here hand my kindergartner over to my robot, right?

02:21 Play from here Like, what is that going to be?

02:23 Play from here So I'm actually spiraling a little bit on the fact that you already have major toy makers

02:27 Play from here who are building chatbots into, like, plush toys.

02:31 Play from here So when your two-year-old wakes up in the middle of the night and says, sing me the

02:35 Play from here song, it will sing them the song, which, well, cute, is also like the stuff of nightmares,

02:41 Play from here for sure.

02:42 Play from here Right.

02:43 Play from here It could be a real nightmare if, like, Grok is powering that toy or, I don't know, there's

02:46 Play from here some really disturbing songs coming out of that.

02:48 Play from here Yeah, you might want to pay attention to what that bear is saying.

02:52 Play from here I would say, yeah, but I think the question is valid, but there's several roads here.

03:00 Play from here First, you said superintelligence, and superintelligence is different from general intelligence.

03:06 Play from here What we have today.

03:07 Play from here It's also a loaded word, to be fair, right?

03:09 Play from here Like, it's a little subjective, depending on who you're talking to.

03:13 Play from here But the AGI definition, this is the artificial general intelligence idea, where the intelligence

03:19 Play from here is generally as equipped as your median human to interact and do things.

03:27 Play from here That's sort of different from the robotic intelligence, and those two coming together

03:32 Play from here probably have a little bit longer to go.

03:34 Play from here But the last time I looked at it, this is probably on Kelshi, but I'm not like a predictive

03:40 Play from here market guy.

03:41 Play from here But I bet if you looked on Kelshi, there'd be a predictive market out there as far as

03:46 Play from here when AGI is going to happen.

03:48 Play from here And it sounds like luminaries are still talking 27, which is not very long from now, like

03:56 Play from here 2027 to 2029.

03:57 Play from here Like how you throw out the year, and you could have just said, yeah, in roughly eight months.

04:06 Play from here Oh my gosh.

04:08 Play from here The robotics have a little ways to go, but you're already seeing a lot of, there's a

04:14 Play from here lot of language out there around autonomous driving and trucking.

04:19 Play from here And it's sort of a resignation that this is going to happen sooner than later.

04:26 Play from here And it does feel like a lot of those pieces are going to start clicking into place.

04:30 Play from here And there is a compounding effect between, let's say, miles on the road with an autonomous

04:35 Play from here driving vehicle and the processing and the techniques that are needed for that, that

04:40 Play from here then translate into robotics that could be in your home or whatever it may be.

04:47 Play from here I do think that's farther.

04:49 Play from here Again, I of course don't know, but I think you're going to have AGI in, gosh, this is

04:55 Play from here wild that we're even talking about this.

04:57 Play from here But you're going to have some semblance of AGI that is not going to necessarily be human,

05:02 Play from here but it's going to be able to take on all of these tasks and do things quite a while before

05:08 Play from here that's integrated in a, in a tactile sense into robotics that can be a, what was it in

05:14 Play from here the Jensen's?

05:15 Play from here Rosie?

05:16 Play from here Rosie the maid.

05:17 Play from here Yeah.

05:18 Play from here And now, now we're both sounding old.

05:20 Play from here I know.

05:21 Play from here I know.

05:22 Play from here And she looks more like a tea kettle than she did a human, but perhaps that's a good

05:25 Play from here idea though.

05:26 Play from here I don't think that we should necessarily make our humanoids indistinguishable from the rest

05:30 Play from here of us.

05:31 Play from here Yeah.

05:32 Play from here That's Westworld.

05:33 Play from here Okay.

05:34 Play from here Here's, here's a part that I think a lot of people miss in all of this.

05:40 Play from here When you look at all of these concurrent technologies bubbling up at the same time is, and by the

05:46 Play from here way, I want to go to these prediction markets now and see what their definition of AGI is

05:49 Play from here because they've got to nail one, right?

05:51 Play from here If you're putting a bunch of money on it, what's, what's specifically their definition?

05:55 Play from here They're going to pull it up right now, aren't you?

05:57 Play from here You're going to go to your account or your prediction market account.

05:59 Play from here As long as you're not betting on wars, Kevin, I'm good with you.

06:05 Play from here If you're not betting on war, you bet on whatever you want.

06:08 Play from here 54% before 2030 is what Kalshi is saying.

06:12 Play from here And do they have a clear definition of AGI?

06:15 Play from here Because that's important.

06:16 Play from here That's a good question.

06:17 Play from here You know, it's close to what I just said.

06:20 Play from here Artificial general intelligence is a hypothetical for the moment type of AI with advanced reasoning

06:24 Play from here skills on par or more advanced than humans.

06:29 Play from here In other words, AI that's our intellectual equal or superior.

06:32 Play from here What about like, you know, Elon's tweet over Christmas break last year was that we've entered

06:37 Play from here the singularity, right?

06:40 Play from here That's a little predictive of like, we're, we're, we're now in the weeds of AGI.

06:45 Play from here Like we're, you know, or, or people, or people who say that it isn't a moment.

06:49 Play from here It's really, it's a, it's a sequence.

06:52 Play from here It's a set of steps and there's going to be an entry point, the thick of it, and then

06:57 Play from here coming out where it's unpredictable.

07:00 Play from here In other words, not unpredictable.

07:01 Play from here Oh, this is going to be totally crazy, but we can't, we can't predict exactly what it's

07:05 Play from here going to be capable of.

07:07 Play from here So I don't know, maybe I'll get in on that prediction market and then make an argument

07:10 Play from here and just take the pot.

07:13 Play from here I think it's really important though.

07:16 Play from here We've said this before, not that AGI is important, but it's this, this ascension curve that the

07:21 Play from here technology is on.

07:23 Play from here And if AGI is it, we're actually relatively close on like a task basis.

07:30 Play from here As far as the types of things that, that AI can take on, you mean as well or better than

07:36 Play from here a human?

07:37 Play from here As well or better than a human.

07:38 Play from here What are the tasks that it can already do?

07:42 Play from here But preparing for the future is really important that there's a, there, there is, there's a

07:46 Play from here lot of preparing for today and you know, I'll meet with companies and they're really excited

07:52 Play from here because they have this great AI strategy and it's basically addressing just this moment

07:56 Play from here in time as opposed to, okay, you're just now after three and a half years grappling with

08:03 Play from here all of the ramifications for your business with AI.

08:06 Play from here You're doing the right things.

08:07 Play from here You're making the investments, you're building literacy, you're getting the right tools,

08:10 Play from here you're getting some reps.

08:11 Play from here That is all great, but you're, you're not necessarily projecting forward in terms of

08:18 Play from here strategic comparatives what the ground is going to look like, be it 2028, 2030, when

08:26 Play from here you have semi ubiquitous intelligences that are all over the place.

08:31 Play from here Like it's, it's really hard to grapple with on like every layer.

08:35 Play from here And I'm not saying that you're remiss and not doing it.

08:38 Play from here You're absolutely bog standard if you're not doing it because it's, it's, it's really hard

08:43 Play from here to predict what it looks like.

08:44 Play from here Well, there's just more and more reports coming out about talk about prediction markets, predicting

08:50 Play from here like what's, what's going to happen to the job market.

08:52 Play from here What's going to happen to the economy.

08:53 Play from here Right.

08:54 Play from here And you, you brought up the autonomous driving stuff, which I think is super interesting

08:56 Play from here when my wife says that she's horrified of the idea of a humanoid robot taking care of

09:01 Play from here a baby.

09:02 Play from here I know that at its core, it's because it doesn't have a soul.

09:06 Play from here Speaking of these various subjective definitions, right?

09:10 Play from here Like if you can't tell the difference, does it have a soul?

09:12 Play from here Big question.

09:14 Play from here You know, to outsource something like that, it's like, okay, is this going to do a better

09:20 Play from here job than a parent at caring for this child?

09:24 Play from here Little bit of, little bit of future prediction there, who knows what's going to happen.

09:27 Play from here But what you said about the autonomous driving, that's super interesting because it's going

09:31 Play from here to be undoubtedly very, very economically disruptive globally when that really, really

09:38 Play from here takes root.

09:39 Play from here And you and I have both spent a lot of time in California where there are the Waymo cars

09:44 Play from here everywhere.

09:45 Play from here And last year in the US, I believe 44,000 people died of auto accidents.

09:51 Play from here And these are all human errors, right?

09:53 Play from here And I think Waymo has now driven something like 25 million miles without a single fatality,

09:58 Play from here which is crazy.

09:59 Play from here So you, you, you have this AI technology that's already objectively better than the task that

10:05 Play from here people can do at the same time.

10:07 Play from here Yet it's also going to be very disruptive, right?

10:10 Play from here This isn't, this isn't, this isn't an AI assisting technology for humans.

10:15 Play from here This is an AI replacing technology for humans.

10:18 Play from here And you know, hey, this is AI work and a lot of people's work happen to be in this world

10:23 Play from here of logistics.

10:24 Play from here And when you're looking on the horizon and identifying things that AI can do better than

10:30 Play from here people today, you know, that's, that's one of them.

10:32 Play from here It's just not completely widespread yet.

10:35 Play from here How do you think about that?

10:36 Play from here Like the logistics and the pros and the cons and where that's going to go?

10:40 Play from here I heard a crazy stat once.

10:41 Play from here It was something like 74% of global GDP is somehow tied to somebody driving something,

10:47 Play from here whether it's an airplane, a forklift in a warehouse or last mile or a train.

10:54 Play from here Like it's, it's, it's, it's a lot, right?

10:56 Play from here And so if you solve that use case in one place, let's call it San Francisco at rush hour from

11:01 Play from here end to end of the city, like one of the hardest use cases ever to operate flawlessly on.

11:06 Play from here You can then project out of, okay, wow, this, yeah, this is going to be everywhere probably

11:11 Play from here really soon.

11:12 Play from here It'll be jagged, of course.

11:15 Play from here It's going to take a little while and humans aren't particularly good at calibrating on

11:21 Play from here autonomous risks versus human risks.

11:24 Play from here You know, as you said, every time you get in that into a, into a vehicle, like you're

11:27 Play from here risking your life, whereas you get into a Waymo or you get into an airliner you're,

11:34 Play from here you're virtually perfectly safe at, at, at the end statistically.

11:39 Play from here But all it takes is one incident and all of a sudden the path of progress gets derailed.

11:46 Play from here Even though to be fair, even though we had 44,000 deaths from human error, it's crazy.

11:52 Play from here Did you, did you hear the story?

11:53 Play from here I think it was maybe early last year that somebody made a kind of Wile E. Coyote painting

12:00 Play from here of the road ahead and put it on the road and then drove a Tesla at it and it couldn't tell

12:03 Play from here the difference between the painting and it just crashed through it like, like a Bugs

12:07 Play from here Bunny cartoon.

12:08 Play from here And that, that got global headlines, right?

12:11 Play from here Like it's full self driving is a sham.

12:13 Play from here It's too dangerous.

12:14 Play from here We can't, you know, we can't, we can't roll this out.

12:16 Play from here So yeah, when something bad happens, it's really amplified right now.

12:21 Play from here So there was a, I believe it may have been Ezra Klein who had an essay this week in the,

12:26 Play from here in the times about job displacement, maybe not being as much of a thing.

12:31 Play from here And there's a, there's an economist named Alex Imus, I think is how you pronounce his

12:36 Play from here name who has definitely been doing the rounds about this.

12:40 Play from here And one of the observations is that if it is not, and this is not Imus, I think this

12:46 Play from here is, this is Ezra Klein.

12:48 Play from here If it's not really fast, there is, the society isn't likely to snap forward and make adjustments.

12:56 Play from here If it's more like a boiling frog and you're not displacing, you know, every trucker at

13:03 Play from here once or every marketing person that wants, and instead it's sort of an air quotes, merely

13:10 Play from here a tripling of unemployment in a particular segment.

13:13 Play from here So if you don't dislodge 10 million marketers, you instead just increase the unemployment

13:18 Play from here among marketers.

13:19 Play from here It's not going to get the societal attention and it's not going to, going to get the policy

13:24 Play from here attention that maybe it needs, or maybe that it warrants because natural adaptation will

13:30 Play from here happen in the economy as, as people spread out and they find more opportunities because

13:35 Play from here humans are really good at adapting over time.

13:38 Play from here I don't fault that.

13:39 Play from here But also very difficult to predict as well, right?

13:42 Play from here Like the, the, the speed of potential replacement.

13:45 Play from here When you think about Uber and taxi cabs, very different of course, because autonomous vehicles

13:50 Play from here would replace people, whereas Uber replaced cab companies, really.

13:54 Play from here But when you remember there was all that lobbying with cab companies, with cities, like we got

14:00 Play from here to get this out of our city.

14:02 Play from here This is disrupting this economy.

14:04 Play from here When you think back to that, and I'd love to see a study on this, would in your mind

14:08 Play from here did that happen slowly or did that happen really fast?

14:10 Play from here Oh, it happened.

14:11 Play from here It happened pretty slowly over about five years.

14:13 Play from here I mean, that's five years slow, five years.

14:16 Play from here Well, how long did it take for the cab industry to develop?

14:19 Play from here Like it's horses and buggies going to, you know, hired cabs, like it took 100, 150 years

14:26 Play from here to develop and five years to completely disrupt.

14:29 Play from here Yeah.

14:30 Play from here So yeah, it depends on your, it depends on your time scale.

14:33 Play from here So five years in today's AI talk seems like an eternity, but on an economic scale.

14:39 Play from here That's really short.

14:41 Play from here It's really, really short.

14:43 Play from here Hey, Kevin, I wanna back out

14:46 Play from here and talk to you about something here.

14:48 Play from here We've been told that maybe we get

14:50 Play from here a little too technical sometimes.

14:53 Play from here And I don't know if it's open claw or agentic use cases

14:58 Play from here or just all the weird crap that we end up building

15:01 Play from here for ourselves to scratch our own itch,

15:03 Play from here which I've continued to do this week.

15:04 Play from here I've got a lot of stories.

15:06 Play from here But I was thinking about this from,

15:10 Play from here your average person's daily AI use

15:13 Play from here as they're sitting in their office.

15:15 Play from here And the stuff that makes the news

15:17 Play from here are like these big AI transformations.

15:20 Play from here Maybe half the company headcount got reduced

15:24 Play from here so they can invest in AI infrastructure.

15:26 Play from here And these are all the grand plans or,

15:29 Play from here hey, here's this startup with a single founder

15:32 Play from here and 20 AI agents burning $20,000 a month in tokens

15:38 Play from here to run the whole company, right?

15:39 Play from here But those are things that get attention

15:42 Play from here because they sound outlandish.

15:45 Play from here But the reality is a lot of people,

15:47 Play from here and I wanna talk about this report

15:49 Play from here that you just sent me this morning.

15:50 Play from here So that's what you love to do

15:51 Play from here is just send me like something really long

15:53 Play from here to read 10 minutes before we record.

15:56 Play from here But before we do, I was thinking,

15:58 Play from here okay, if we had a breakout,

16:01 Play from here the different ways that people use

16:04 Play from here the main models that everybody has,

16:06 Play from here we've got Gemini, GPT, and we've got Clawed,

16:09 Play from here and they all have similar capabilities.

16:12 Play from here So there's not a one-for-one overlap.

16:13 Play from here So you can't really go wrong with any of these.

16:16 Play from here And I was thinking, okay, in my day,

16:20 Play from here I've got prompts, like I just open one of these up

16:23 Play from here and I say that I need help with something, right?

16:27 Play from here And most people, to be fair,

16:29 Play from here that prompt is something more likely

16:31 Play from here to type into Google as opposed to a specific case

16:34 Play from here that AI would really be good at solving.

16:36 Play from here And then you've got projects,

16:37 Play from here which would be the reusable aspects of doing that.

16:40 Play from here So I've got this problem,

16:42 Play from here and it's a problem I have all the time.

16:44 Play from here Maybe it's once a week, maybe it's every day,

16:46 Play from here and I don't want to feed it the context of this problem

16:49 Play from here and the attachments as references and all the instructions.

16:52 Play from here So I'm gonna save it as this project

16:54 Play from here or this custom GPT or this gem.

16:56 Play from here And then you have automations, right?

16:58 Play from here And you and I went deep down into the rabbit hole

17:01 Play from here of these third-party automation tools

17:03 Play from here that would use an intelligence like Gemini or Cloud

17:07 Play from here to power them as part of a workflow.

17:09 Play from here But it was really a, if this, then that type of thing.

17:12 Play from here These automations, they're fixed.

17:15 Play from here They're looking for a trigger

17:16 Play from here or they're running on a cron job.

17:18 Play from here In other words, like at 10 o'clock every morning,

17:19 Play from here do this thing,

17:20 Play from here or when this email comes in from this person,

17:22 Play from here run the sequence.

17:23 Play from here And then you have agents.

17:24 Play from here And I think agents are probably the thing

17:26 Play from here that people have the least experience with.

17:28 Play from here They're being used the least in enterprise today,

17:30 Play from here but when they are used, we all hear about them

17:32 Play from here because it just sounds super magical.

17:35 Play from here Am I missing something?

17:36 Play from here Like when you zoom out,

17:38 Play from here are those kind of the four categories

17:40 Play from here that you would think about somebody using this at work today?

17:44 Play from here Yeah, I think that's totally fair is, you know,

17:47 Play from here to be honest, from what I'm seeing in the market,

17:51 Play from here people are still just sort of in the prompting phase

17:54 Play from here where they're-

17:55 Play from here I was gonna ask you.

17:55 Play from here I was gonna ask you that.

17:56 Play from here Nice.

17:57 Play from here That's a predictive answer right there.

17:59 Play from here And that's surprising to me.

18:01 Play from here But again, we spend a lot of time doing this.

18:03 Play from here We're gonna be pushing the capabilities.

18:06 Play from here So I was gonna follow it up with saying like of all those,

18:09 Play from here that's what I do the least these days.

18:12 Play from here And I was trying to think of a use case

18:14 Play from here just in the last week,

18:15 Play from here where I just jumped in with a prompt

18:18 Play from here that wasn't part of a project

18:19 Play from here or an agent that I had or something,

18:21 Play from here and just did that.

18:21 Play from here Are you still using this?

18:23 Play from here And if you're not,

18:24 Play from here how are the people that you work with

18:25 Play from here still using that specifically?

18:27 Play from here Okay.

18:28 Play from here So the first is it's so easy to get FOMO about this stuff.

18:31 Play from here And-

18:32 Play from here For sure.

18:33 Play from here Wow, everybody's rampaging forward

18:34 Play from here and they're doing all of these things.

18:35 Play from here And the calibration is,

18:37 Play from here most people are still opening up

18:38 Play from here probably a free LLM, a chatbot of some kind,

18:42 Play from here and they're using it like Google to answer a question.

18:45 Play from here The more corporate approach

18:49 Play from here is consolidation around a single platform.

18:52 Play from here And that's a really good idea.

18:53 Play from here If you haven't done that,

18:54 Play from here you need to do that.

18:55 Play from here I would,

18:56 Play from here the three that you're going to choose from

18:59 Play from here are likely going to be

19:00 Play from here Anthropx, Clod, OpenAIs, GPT,

19:04 Play from here or Google Gemini.

19:05 Play from here For different reasons,

19:07 Play from here they have different powers,

19:08 Play from here but they're all kind of converging

19:09 Play from here in terms of capabilities.

19:11 Play from here So as long as you lean into them

19:13 Play from here from a team perspective

19:15 Play from here and lever them to a high degree,

19:17 Play from here like you're in a pretty good place.

19:19 Play from here And I think that

19:21 Play from here a lot of organizations are there

19:23 Play from here and they're starting to get grappled then

19:25 Play from here with like shared knowledge

19:27 Play from here and projects and custom GPTs or gems,

19:31 Play from here which are consolidating information.

19:34 Play from here And we talk about context.

19:36 Play from here Context is really important

19:38 Play from here into certain verticals.

19:40 Play from here And then they sort of tail off

19:43 Play from here as far as how they're applying these things.

19:47 Play from here To answer your question directly,

19:49 Play from here I actually,

19:50 Play from here this is a little funny,

19:51 Play from here but I keep a GPT tab up during the day.

19:54 Play from here This time last year,

19:55 Play from here I would have had 10, 15 GPT tabs up in Chrome,

19:59 Play from here maybe one Clod.

20:01 Play from here Which has different threads that you started.

20:03 Play from here Different threads,

20:03 Play from here different things,

20:04 Play from here different projects through the day that I'm doing.

20:07 Play from here And now I probably have five or six Clod

20:09 Play from here because I've shifted most of my user,

20:11 Play from here my personal behavior in that direction.

20:13 Play from here But I keep a GPT up

20:15 Play from here and I'm not shifting

20:16 Play from here my GPT up.

20:17 Play from here And I'm not sure quite why I do this,

20:18 Play from here except that it might be visual.

20:20 Play from here It's not really nostalgia,

20:23 Play from here but like I am paying for it.

20:25 Play from here There's a lot of context in there that it's useful,

20:28 Play from here but I actually use that one tab

20:31 Play from here for like my information dumping ground.

20:33 Play from here If I do want something

20:34 Play from here that is a more exotic version of Google,

20:37 Play from here by the way,

20:38 Play from here I do use Google still for like something that I know

20:40 Play from here is going to be like a Google type result,

20:44 Play from here as opposed to more of a guidance type result.

20:47 Play from here If you're looking for,

20:48 Play from here if you're looking for a set of ideas,

20:51 Play from here Google's a good place to go.

20:53 Play from here If you're looking to solve a specific problem,

20:57 Play from here the LLMs are the better place to go for that.

21:00 Play from here Not to dive into a rabbit hole from another rabbit hole,

21:02 Play from here but I'm just going to do it anyways.

21:03 Play from here When it comes to Google,

21:05 Play from here have you found any parallel with any other model

21:08 Play from here when it comes to deep research?

21:09 Play from here Do you think it's the best?

21:10 Play from here You know,

21:11 Play from here I actually think that GPT's deep research is very good.

21:14 Play from here I think that Perplexity's deep research

21:16 Play from here is very good as well.

21:19 Play from here I don't think that people need advanced,

21:22 Play from here advanced deep research

21:24 Play from here based on most of the things that they're after,

21:27 Play from here but Google has the edge,

21:28 Play from here but it's user behaviors are different.

21:31 Play from here I pay for Gemini as well,

21:33 Play from here because of course I do.

21:34 Play from here But if I were to do deep research right now,

21:36 Play from here I would probably go to OpenAI,

21:38 Play from here not for a particularly good reason

21:40 Play from here other than just user behavior.

21:42 Play from here Familiarity.

21:43 Play from here Yeah, familiarity and context.

21:45 Play from here So context is important.

21:48 Play from here If I want to spend less time contextualizing a model,

21:51 Play from here I will go to GPT

21:53 Play from here because I haven't really bothered

21:55 Play from here to build up the context of Gemini.

21:59 Play from here But getting back to the point

22:01 Play from here of where people should be here,

22:04 Play from here that's not a bad place to be.

22:05 Play from here Having some policies,

22:07 Play from here having a central like platform

22:09 Play from here and hopefully teaching people

22:11 Play from here kind of how to use it responsibly,

22:13 Play from here how to bake projects

22:15 Play from here and have that context and et cetera.

22:17 Play from here That's pretty powerful.

22:19 Play from here And then you get kind of where we are right now.

22:22 Play from here You know, the term agent gets tossed all over the place,

22:26 Play from here but essentially it's the idea

22:28 Play from here of it being a more autonomous action

22:30 Play from here that accomplishes something

22:32 Play from here that then can come back into the organization.

22:35 Play from here And it's been super fiddly

22:37 Play from here to make all of that stuff happen.

22:39 Play from here I know, it's been like my favorite term.

22:40 Play from here I love that word, fiddly.

22:42 Play from here But it is, it's true.

22:43 Play from here I told you Claude's favorite word for it, brittle.

22:46 Play from here It's brittle.

22:47 Play from here It's brittle.

22:47 Play from here It's brittle.

22:48 Play from here It's like a sand dollar you find at the beach.

22:50 Play from here You're like, this thing's amazing.

22:52 Play from here It's beautiful.

22:52 Play from here And you put it in your pocket

22:53 Play from here and it turns to dust.

22:55 Play from here But the term that is just burning a hole

22:57 Play from here in my brain is agentic harnesses.

23:00 Play from here And it's this, it's kind of,

23:02 Play from here I don't love the term.

23:04 Play from here It's descriptive in a way

23:07 Play from here that I think that people like us understand.

23:09 Play from here Dirty.

23:12 Play from here But the idea is you're moving things

23:15 Play from here to a point where it's not just the chatbot

23:19 Play from here and not just the information that the chatbot has.

23:22 Play from here Previously, if you wanted to have really good context

23:25 Play from here or you wanted to build a proposal or a response,

23:28 Play from here you could have connected it to documents

23:30 Play from here and you could have pulled those documents in

23:32 Play from here and you could have kind of worked with them.

23:34 Play from here But now that landscape is becoming so much clearer

23:37 Play from here as far as the mono threads

23:40 Play from here that we've talked about in previous weeks.

23:42 Play from here But the idea that you can connect

23:44 Play from here all of these different data sources

23:47 Play from here and then actually start interacting

23:49 Play from here across those data sources,

23:51 Play from here which I'm finding very, very powerful.

23:53 Play from here That now that I have,

23:56 Play from here so cloud enabled in Microsoft, for example,

24:00 Play from here I have continuity across my productivity

24:05 Play from here plus my data storage, plus my chats,

24:08 Play from here plus the context that I have in cloud.

24:12 Play from here And that's super useful.

24:13 Play from here There's a lot less like upload here,

24:16 Play from here download here, move around, copy, et cetera.

24:19 Play from here And I'm actually no Microsoft fan,

24:21 Play from here but all of a sudden this stuff is starting to work

24:25 Play from here in a way that Copilot couldn't figure out,

24:28 Play from here which is Microsoft's AI.

24:29 Play from here And now that it has a cloud layer on it,

24:32 Play from here things are just working.

24:33 Play from here And probably some new GPT layers on it too

24:36 Play from here because they've renewed their partnership.

24:39 Play from here Microsoft has a massive interest in open AI

24:42 Play from here and they put out a very brief press release

24:46 Play from here that wasn't very, very descriptive

24:50 Play from here as to exactly what this was gonna mean.

24:52 Play from here But for the next years to come,

24:55 Play from here they're in bed together

24:56 Play from here and I bet we see a bunch of the capabilities

24:58 Play from here trickle over into the Microsoft ecosystem as well.

25:02 Play from here And you said something that I wanna circle back to

25:06 Play from here about just the context history.

25:08 Play from here And when people are betting on one of these horses,

25:12 Play from here if they don't have the bandwidth or the finances

25:15 Play from here to be rolling all of them simultaneously,

25:20 Play from here or if it's not what they do for a living like you,

25:22 Play from here this context, this history, this wasn't always a thing

25:26 Play from here and it keeps getting better and better.

25:28 Play from here And so what we're talking about is a model's ability

25:33 Play from here to have this rolling memory of all your conversations.

25:37 Play from here Sometimes they get siloed into the projects

25:39 Play from here or the custom GPTs, but it's ever evolving.

25:43 Play from here And now there's this seeping through

25:46 Play from here with a lot of the models of the more time you spend

25:50 Play from here working with them on these different problems,

25:52 Play from here on these different projects, on these different workflows,

25:55 Play from here the more collective context they have about you,

25:58 Play from here how you work and the problems that you're trying to solve.

26:01 Play from here In essence, they become more valuable over time.

26:04 Play from here And then you have the pain of switching, right?

26:06 Play from here You end up in that same spot

26:07 Play from here that you might've a few years ago

26:09 Play from here when you decided you don't like your CRM

26:11 Play from here and it's too expensive.

26:12 Play from here Then you think about switching and you think about,

26:15 Play from here is all this data gonna move over?

26:17 Play from here Is it gonna sync up seamlessly?

26:19 Play from here Am I gonna like the new one better, right?

26:21 Play from here You've really got a sunk cost in there.

26:24 Play from here And now we're starting to see that with these labs as well.

26:26 Play from here And that's something that you experienced

26:28 Play from here with all of them basically, right?

26:30 Play from here But probably mostly Anthropic and OpenAI right now for you.

26:34 Play from here Yes, definitely.

26:35 Play from here And portability from a policy perspective

26:37 Play from here is sort of an interesting thing to think about

26:40 Play from here because it's a known weakness of platforms

26:44 Play from here that reduces your ability to pull your data out,

26:49 Play from here move somewhere over.

26:50 Play from here Jump around.

26:51 Play from here Jump around a little bit.

26:52 Play from here And there are kludgy mechanisms by which you can do it,

26:56 Play from here which do an okay job.

26:58 Play from here But just like the giant ad platforms

27:01 Play from here know way, way, way too much about you,

27:03 Play from here so do these giant AI platforms.

27:06 Play from here But oddly, I think that the tenor

27:08 Play from here that's being expressed from a regulatory perspective

27:11 Play from here is actually more keen on that portability.

27:15 Play from here So in the future, this is just prognosticating,

27:18 Play from here but you may have the ability to have rights

27:22 Play from here over this online identity that is so much more intimate

27:27 Play from here than just the data points that previously would have existed

27:30 Play from here and the ability to be deleted,

27:32 Play from here the ability to be able to transfer it somewhere else,

27:35 Play from here like this corpus of information

27:38 Play from here that you can then move around.

27:40 Play from here That's sort of utopian.

27:42 Play from here We'll see how it plays out.

27:43 Play from here But wasn't there something,

27:44 Play from here so when all of the crazy stuff with the Pentagon

27:48 Play from here and Anthropic went down and then Altman swooped in

27:52 Play from here and there was like this mass exodus

27:55 Play from here from OpenAI to Anthropic,

27:57 Play from here wasn't there some feature that they pushed at that time

28:00 Play from here that was like, oh, it's a single button import.

28:03 Play from here You just put in your OpenAI log grads or whatever it is,

28:06 Play from here and we'll just suck in all the context into our model

28:09 Play from here because we're in all of them all the time.

28:12 Play from here Like that was something I never got to test.

28:14 Play from here Was that real?

28:15 Play from here Did that work?

28:16 Play from here Did you hear much about that?

28:17 Play from here Yeah, it kind of worked.

28:17 Play from here It was pretty small.

28:18 Play from here It wasn't everything, right?

28:19 Play from here Like that seemed a little outlandish that it could,

28:21 Play from here because man, if you've been on one of these for like,

28:24 Play from here let's say a year and use it every day,

28:26 Play from here it's a lot of context.

28:27 Play from here It's an enormous amount of context

28:30 Play from here that's being absorbed in there.

28:32 Play from here Yeah.

28:34 Play from here Yeah, it worked within reason,

28:36 Play from here but it was just this little snapshot.

28:38 Play from here And another thing that people need to know,

28:41 Play from here I've used this term before,

28:42 Play from here but when you open up a new account

28:45 Play from here or you invite people on your team to a new account,

28:48 Play from here it doesn't know anything about you.

28:50 Play from here So it's just a little baby.

28:51 Play from here And then it's total.

28:53 Play from here It's just a little baby.

28:55 Play from here Just a little baby.

28:56 Play from here Oh, a toddler with a bottle.

28:58 Play from here Isn't it cute?

29:00 Play from here It's totally not going to take your job,

29:02 Play from here but it was such a,

29:07 Play from here it takes some effort to train it

29:10 Play from here and to get all of that information.

29:11 Play from here And increasingly the platforms have this baked in.

29:15 Play from here So the, the Claude team's account,

29:19 Play from here has a layer in it now in the administrative layer that allows you to add organizational context.

29:26 Play from here And that's really useful when a new employee joins your organization,

29:30 Play from here to be able to have immediate context for their own little baby cloud account.

29:36 Play from here And then they need to spend a little bit of time educating it. So we actually tell people to,

29:41 Play from here you know, hey, take a half an hour, take 45 minutes, ask your LLM to interview you

29:47 Play from here about who you are, what you do, like how you like to communicate, etc.

29:52 Play from here And we encourage them to use either the built-in dictation tools or something like WhisperFlow

29:57 Play from here that allows you to capture the transcription in the live such that they build up.

30:01 Play from here They have the organizational context. What does the organization do? What does it represent,

30:05 Play from here etc. What's its branding, messaging, etc. And then themselves, what do they do?

30:11 Play from here And these all layer on top of each other and are helping to make people more effective right off

30:16 Play from here the bat. But when people get frustrated, it's often because they feel let down by these new

30:21 Play from here models, either because they haven't spent the time contextualizing them or because, frankly,

30:26 Play from here they're using the wrong model. They're using kind of a snap-to model, like one of the instant

30:32 Play from here answer models, as opposed to one of the longer thinking models, like the Opus series, for

30:39 Play from here example. Right. And a lot of the times that comes down to, are you willing to put in a credit card

30:44 Play from here or are you not willing to put in a credit card, right? Like if you're using these three models,

30:47 Play from here you're not getting best in class. You may not get the type of experience that everybody is

30:52 Play from here whooping over, right? It's going to be markedly different. Kevin, by the way, you're the worst

30:57 Play from here person to talk to when somebody is really tired because there was like four things I was going to

31:01 Play from here say, but then you just have so much to say about one of them. I can't remember where I started,

31:06 Play from here but I was going to go back to prompts and how you said that most of the people that you're

31:10 Play from here working with are still in this world of, they open up a fresh chat and they start prompting.

31:15 Play from here Do you do that at all anymore, ever? Do you still find yourself doing that?

31:18 Play from here Well, that's what I use the GPT for, my GPT tab.

31:22 Play from here And that's your one-off?

31:23 Play from here That's my one-off that I just go. But because I have 32,000 chats in GPT in my own account. So

31:31 Play from here when I do a quote one-off prompt, it's actually not a one-off prompt. It is well contextualized

31:37 Play from here as far as a lot of the things in my world. And it's going to get me pretty quickly to whatever

31:42 Play from here that answer is.

31:43 Play from here That's interesting because I know personally how much time you spend in the Anthropic ecosystem

31:48 Play from here as well. So it doesn't have all the context. It's got a lot of context.

31:54 Play from here It's painful because there are things that I know live in great detail in OpenAI and things that

32:01 Play from here live in great detail in Anthropic. And from, again, a behavioral perspective, but this time last year,

32:07 Play from here I was already using Cloud for sort of heavy lift things that are more like thought, strategy, etc.

32:15 Play from here Like a thought partner, strategic partner type.

32:17 Play from here But I wasn't willing to pay the premium for the pro accounts or the max accounts at the time

32:25 Play from here because Cloud Code really didn't exist. I didn't really have a great reason to.

32:29 Play from here And what I found was I could only use Cloud for, at the time, maybe 45 minutes as a power

32:36 Play from here user before I would tap it out.

32:40 Play from here On your $20 a month plan.

32:41 Play from here On the $20 plan. And I wasn't keen enough with the differentiation between GPT that I was willing

32:47 Play from here to pay that. But it meant that Cloud couldn't become like my daily driver in a lot of ways.

32:55 Play from here Until you turn the dial up on the spend.

32:58 Play from here When I look at the two, the important stuff has actually always been in Cloud because that's

33:03 Play from here where I was doing the heavy lifting, particularly if I needed to do writing at the other end of it,

33:08 Play from here and I needed writing assistance to help me get through whatever it was that I was creating.

33:13 Play from here So it hasn't been a big deal to me. It's actually, it's funny, I can't immediately

33:19 Play from here remember the last time I needed to go to GPT just because of a contextual issue.

33:24 Play from here Interesting. Okay. So yeah, a couple more questions for you on that.

33:29 Play from here So back to somebody who is still just doing these one-off prompts, and they might be doing a lot of

33:35 Play from here them, right? You said that you weren't getting a lot of mileage out of the $20 a month plan for

33:42 Play from here Cloud. It would just tap out really fast. Going back when I was bouncing a lot between Gemini and

33:49 Play from here GPT, I found that I got a lot of mileage out of both those models. So I'd rarely hit a point where

33:56 Play from here it said, you have hit your limit. And I am not sure where that's landed today. I mean, I know

34:03 Play from here even on my $100 a month max plan with Cloud, I'll still hit session limits, especially if I'm using

34:10 Play from here the big, fancy Opus model. But with GPT and with Gemini, are you using it enough that you have a

34:17 Play from here sense as to how far you can push it on that $20 a month plan, which for a lot of people out there

34:23 Play from here is the next step of being able to have their hands on a better model and more utility? Are you finding

34:29 Play from here that it's still a lot more generous, or are you using it enough to know? It's much more generous.

34:35 Play from here And when I have somebody who is sort of AI curious still, and they want to get into it,

34:41 Play from here I literally have a text to a friend up right now of, oh, which Cloud plan should I choose? And I'm

34:46 Play from here like, eh, well, you're not going to get really far with just the normal plan. So you pretty much,

34:52 Play from here if you do anything that involves automation or heavy work, you can expect to pay the $100.

34:58 Play from here So if you want to stay cheap and easy, GPT. I can't, I haven't run out of GPT use in ages.

35:06 Play from here It has happened, but I think it's only happened a couple of times in the last three and a half

35:11 Play from here years. And now I'm like really hesitant to leapfrog over this one because here we go again

35:16 Play from here down the rabbit hole. But have you been using GPT for any coding exercises or any agentic

35:23 Play from here exercises? And if you do that, then do you start to experience that same kind of,

35:30 Play from here oh, I got three things out the door and now I'm done for four hours, right? That you would on the

35:36 Play from here $20 a month Cloud plan. So this is more observational than experiential because I did

35:42 Play from here dive in last weekend to try and do a specific project in Codex, which is the GPT version of

35:51 Play from here Cloud Code. And it's excellent. It is faster in a lot of respects. It is better as far as quality,

35:57 Play from here but I didn't actually get to a great result at the end of the weekend.

36:02 Play from here Interesting. Explain what a great result is here.

36:05 Play from here This is user error in that it was a total side quest. It was something I wanted to do.

36:11 Play from here And I literally just kind of had it running and I wasn't giving it much care and feeding,

36:16 Play from here but I did lean into Cloud to do the planning, Cloud design to do the user experience like

36:25 Play from here architecture. And then I packaged all that up and basically tossed it over the fence to Codex

36:30 Play from here and occasionally like, you know, checked in to see how it was doing.

36:34 Play from here So Cloud basically gave you the instructions, the blueprint for what Codex was going to build.

36:39 Play from here You had an architect come from across the street and then you handed over the blueprints

36:43 Play from here to a lower cost builder. Yep.

36:46 Play from here But...

36:47 Play from here Exactly what I did.

36:48 Play from here Okay. And so as far as the results not being great or...

36:54 Play from here It was terrible. What happened?

36:56 Play from here And absolutely people are going to hate on me. And because I am not a power Codex user,

37:02 Play from here people are getting wonderful results out of Codex. They're paying a fraction as of those of us who

37:08 Play from here are in Cloud code. It's working great. It's working fast. Everything is rosy.

37:14 Play from here So this is somewhere along the lines. I was being deliberately vague about what I was doing,

37:21 Play from here but I've done the same thing in Cloud code and ended up with a functional product at the end

37:25 Play from here of the road. In this case, it wasn't even remotely a functional product. Like it was completely

37:31 Play from here broken. Again, no shade on it at all. It's just, if you're going, this is the same thing I was

37:36 Play from here saying earlier. If you're going to lean into a universe, it's a good idea to lean into a universe.

37:41 Play from here If you're starting today and you want to be cost-effective and not be rate limited as much,

37:47 Play from here CPT is great. Lean into Codex. Learn how to use Codex as your agent building tool or whatever

37:54 Play from here it is. And you will learn tips and tricks and other things that I simply didn't have because

38:01 Play from here I've been leaning into Cloud code too much. And, you know, as trying to cross over between the two

38:06 Play from here wasn't as seamless as I expected and wasn't as seamless as the internet is leading people to

38:12 Play from here believe. I'm starting to get vibes of the Android versus iOS user type thing, you know, like where

38:21 Play from here people really feel affiliated with one camp or another, like, how dare you send me a text from

38:26 Play from here an Android? What's with these weird bubbles I'm getting, you know? And I think very often these

38:31 Play from here are more technical people or writers that feel drawn towards the Anthropic models as opposed to

38:38 Play from here GPT. And I think there are some people who they've always been on GPT and GPT has been rolling out a

38:45 Play from here ton. I mean, OpenAI has been rolling out a ton of features and it keeps getting better and better

38:49 Play from here and probably won't have time, but I wanted to talk to you about the latest release and how much

38:53 Play from here you've been able to use that or there are different gigantic tools they have there. But yeah, like it

38:58 Play from here really feels like for those people who are past the exploratory phase, a lot of them are starting

39:03 Play from here to lean into one lab versus another and feel a strong affiliation. Are you seeing that in the

39:09 Play from here workplace at all, Kevin? I'm encouraging it in the workplace to a degree. If you are able to go

39:13 Play from here deeper and we absolutely need to take two steps back here and do some clarifications in just a

39:18 Play from here second, but okay. Well, the clarification is we are talking about Codex and we are talking about

39:24 Play from here Cloud Code. Both of those are very scary names for like normal people who aren't developers. Well,

39:30 Play from here I'm not a coder. Like what the heck? Why am I using Cloud Code and something called Codex? Oh,

39:34 Play from here that's got to be scary too. And the fact is these are exactly the same tools that are used to build

39:40 Play from here code and build apps and things like that behind the scenes. And they really have done a disservice

39:47 Play from here and Anthropic is trying with the rebranding around co-work being like a softer, nicer approach to

39:54 Play from here all of this. But it doesn't matter to a user if they're coding under the hood, if they're trying

40:00 Play from here to get to a result. Roll back the clock just six months ago, if you wanted to like have scanned

40:06 Play from here Gmail, pull out invoices, translate those invoices, do something with them, we would have been talking

40:11 Play from here about. In fact, we were talking about workflow software like an N or Zapier or make or one of

40:17 Play from here those. We're basically doing the same thing with Codex or with Cloud Code, except you're doing it

40:24 Play from here just purely in natural language instead of me describing like, oh, I need to build this workflow

40:31 Play from here in it. And I'm going to have this node and this connection and this API, these nodes and

40:36 Play from here the authorizations for each. And as much as I'm warm and warm in my heart about N8n, because it

40:41 Play from here is a really useful tool, you don't need to do it anymore. Now you use that same natural language

40:46 Play from here and you say, hey, Cloud Code or Codex, I need you to check my Gmail once a day and find invoices and

40:55 Play from here then turn the invoices into something that you can do something with and then forward it to my

41:00 Play from here bookkeeper on the other end. And you're just looking at your chat box. And maybe this is

41:05 Play from here an unlock for people who are listening that you're just looking at that chat box and it's going to

41:10 Play from here say, OK, cool. I need to connect to your Gmail. Is that OK? All right. Agent Harness piece one, Gmail

41:18 Play from here done. Oh, I need a skill that's going to take a PDF document from your Gmail and be able to do

41:24 Play from here something with it. It's just going to do that. It doesn't need you to figure that out. And then

41:29 Play from here it's going to say, oh, OK, you want that to go somewhere else. Oh, you want it to go to Slack

41:34 Play from here for your bookkeeper. So now would you like me to connect to Slack? So now you have Agent Harness

41:40 Play from here connection two, which is Slack. And it's just bridging the gap between those and it's making

41:47 Play from here it happen. Yes, it is code behind the scenes. But for most and I mean, most people, it doesn't

41:53 Play from here matter that it's code. So don't get bound up. When we say Codex, when we say Cloud Code,

42:00 Play from here these are really the engines powering a bunch of things that to you is just going to look like a

42:06 Play from here text box and you can just put in here's what I want to do. And then there'll be a little place

42:11 Play from here where that gets pinned in your user interface that you can go back to it and it either runs on the

42:16 Play from here clock or you run it when you need to. And it's making all that magic happen because of that

42:20 Play from here engine. That's essentially what it's doing. And that's an agent in a nutshell. Like that is that

42:24 Play from here is that it's going to run. Automation agent like that's I think another good thing to define.

42:29 Play from here Well, no, it's an agent. As far as marketing is concerned, that is an agent. If it is,

42:34 Play from here it is autonomously on a schedule going to something like email, finding something,

42:40 Play from here parsing it and then pushing it to an action that needs to happen. That's an agent. Like I don't

42:46 Play from here you can you can you can call it an automation if you want. But that's what people are referring

42:52 Play from here to agents as. Can they do way more elaborate things by taking OK, that's just one that's like

42:58 Play from here AP scheduling. But, oh, I'm also going to connect another automation slash agent to like payroll

43:05 Play from here or whatever it is. And now all of those things together feel agentic because it's like a lot

43:12 Play from here of automations are happening under the hood. But it's just a play devil's advocate because you're

43:16 Play from here a big fanboy. Was that was my loss. Even like back to the days of Zapier, right,

43:23 Play from here which is still a real thing, like back in the days, like company doesn't exist anymore. I think

43:27 Play from here it's doing really well. But those were I mean, not N8N, but Zapier, for example, was around way

43:34 Play from here before any of these frontier models were before we had like real usable LLMs. But when you put in

43:39 Play from here that layer of intelligence in that automation, this was clearly an automation. When this happens,

43:45 Play from here then do this. Like if an email from this person comes in, then I want you to put it in the CRM

43:50 Play from here and I want you to tag it with this and I want you to put in this note. Just a bunch of different

43:54 Play from here rules. Right. But when you have that layer of intelligence from one of the LLMs that you

43:58 Play from here could interject into that automation.

43:59 Play from here and workflow, was that the moment that it became an agent because it's now making decisions

44:05 Play from here in a workflow that used to just be if this, then that trigger based workflows?

44:10 Play from here You know what?

44:11 Play from here 99.9% of people never use those automations and they're calling them agents.

44:14 Play from here And I don't feel the reason to fight it.

44:17 Play from here Like who cares?

44:18 Play from here Like it's, it's, if it's, if it's moving it from one side of the digital desk to the other

44:22 Play from here side of the digital desk, yeah.

44:24 Play from here Okay.

44:25 Play from here That's an automation.

44:26 Play from here It's moving it from one side of the digital desk with an insight, which I did leave out

44:30 Play from here of that, that in description, which is read the invoice, determine if the invoice is abnormal

44:36 Play from here based on, you know, other invoices that you've received from this vendor, making an informed

44:42 Play from here decision.

44:43 Play from here Yes.

44:44 Play from here Okay.

44:45 Play from here That is more agentic, but it is, it's always been surprising to me that outside of like

44:47 Play from here kind of techie SMB people and enterprises, very few people actually like leaned into

44:54 Play from here heavy automation of that sort and fighting the semantics of it.

44:59 Play from here Like, I don't know, like we're accomplishing stuff automagically based on signals that

45:05 Play from here formally humans had to go hunt down and then putting the human at the inflection point.

45:11 Play from here That seems to me like an unlock.

45:13 Play from here Okay.

45:14 Play from here So what you're saying is kind of like AGI definitions are starting to bleed a little

45:16 Play from here bit.

45:17 Play from here It's, it's, it's becoming a little, it's becoming a little more fluid.

45:21 Play from here So when we kind of back up into just these prompts, which you say, this is how most people,

45:26 Play from here and I agree when I just, you keep saying that I I'm like, like most people at large, I'm

45:31 Play from here not saying my client, like if, if somebody is going to engage with a relatively expensive

45:35 Play from here consultancy to come in and do this stuff, they're going to be serious about like putting

45:40 Play from here in structure.

45:41 Play from here I'm saying out there in the, in like the, the broad field, as I speak to audiences and

45:46 Play from here I ask for like, who's using what for what?

45:49 Play from here Like most people are very fast and loose.

45:52 Play from here Most organizations have no centralized approach, like, you know, et cetera.

45:57 Play from here That's different from, you know, sophisticated forward-looking companies are still just like

46:02 Play from here doing random prompt.

46:03 Play from here Right.

46:04 Play from here Right.

46:05 Play from here Okay.

46:06 Play from here That's, that's fair.

46:07 Play from here And I mean, listen, yeah.

46:08 Play from here Normies, like when I talked to most people and it's not specific to an organization that's

46:12 Play from here trying to do an AI transformation or something.

46:14 Play from here Yeah.

46:15 Play from here It's just, there's, there's the text box.

46:18 Play from here There's the prompt.

46:19 Play from here Here's something I need help with or a question I have.

46:22 Play from here And I think the more sophisticated people are thinking ahead of like, Hey, here's all

46:26 Play from here these documents and pulling all of the specific context out of these documents and putting

46:30 Play from here it into a report or something would take me a whole day.

46:33 Play from here This thing can chew through it in the next few minutes.

46:36 Play from here You know, people are starting to use it like that, but for yourself, it's a very one-off

46:40 Play from here thing.

46:42 Play from here You're going to your single browser tab that's running GPT and you're, you're doing that.

46:46 Play from here So the next step down as far as projects go, could you give an example of a project?

46:52 Play from here And by the way, that's what they're called for Claude in Gemini.

46:56 Play from here They are called gems and GPT.

46:58 Play from here They're called custom GPTs.

46:59 Play from here Would you say?

47:00 Play from here There are a lot of projects in GPT.

47:02 Play from here But it's different than a custom GPT though.

47:04 Play from here No?

47:05 Play from here Well.

47:06 Play from here That's where it gets tricky.

47:07 Play from here Okay.

47:08 Play from here So again, we're not doing, we're not doing our, our naysayer from the audience, any service

47:13 Play from here here by making this any more complicated, like a project is a project.

47:16 Play from here It's something that has context around an idea and you can accumulate a bunch of documents.

47:22 Play from here And now in either GPT or Anthropic, you can also tie in external connectors if you want

47:28 Play from here such that, um, Oh, I realized that this needs to access my, my Gmail.

47:34 Play from here So I'm just going to simply reach into Gmail and grab the information as opposed to opening

47:38 Play from here up Gmail and copying and pasting.

47:40 Play from here So it has its own little piece of context around it.

47:43 Play from here That's the way to think of projects.

47:44 Play from here And guess what?

47:45 Play from here That's what a project is.

47:46 Play from here Like I have file folders that are under my desk that are a project.

47:51 Play from here It has stuff in it, old documents and whatever we used to do.

47:54 Play from here And they're all very dusty, but like, you know, 2018, like that sort of stuff.

47:59 Play from here Just think of it that way.

48:00 Play from here Don't no need to overcomplicate it.

48:02 Play from here When you're talking about custom GPTs, that is a slightly different flavor, but because

48:08 Play from here they're, they're more designed for repeated actions and they're falling by the wayside

48:14 Play from here because we, what we haven't talked about today and what is important because as much as I'm

48:18 Play from here a Claude fan guy, fan boy, and you're pretty deep into Gemini, OpenAI has its whole new

48:25 Play from here agents engine, which is really cool and really accessible.

48:29 Play from here So forget custom GPTs.

48:31 Play from here If you have a custom GPT that you've built that does something cool that works for your

48:35 Play from here customer service team, for your marketing team, by all means, keep using it.

48:40 Play from here Like great.

48:41 Play from here No reason to reinvent it, but if you want to do like multi-step stuff that we're talking

48:46 Play from here about in terms of the agentic harness, then look on the left bar, you want to get specific?

48:53 Play from here Look on the left bar of your GPT account.

48:56 Play from here If you are in a team's account, you will see something that says agents on it.

49:01 Play from here Click on browse templates in the agents, and you're going to see a really clear and

49:06 Play from here graphical representation of a bunch of really good ideas that may or may not totally match

49:13 Play from here your organization as far as the connectors and the platforms you have.

49:17 Play from here But these are repeatable actions that you can do that, that do more than you could do

49:23 Play from here with a custom GPTs previously.

49:26 Play from here And frankly are way easier to use than Cowork and on the Anthropic side, at least at the

49:32 Play from here moment.

49:33 Play from here Interesting.

49:34 Play from here Definitely way easier to use than the Gemini version, which is agent.

49:38 Play from here What is it?

49:39 Play from here The agent.

49:40 Play from here Gemini has its own like term for this.

49:41 Play from here Really for developers.

49:42 Play from here I mean, it's not, there's no, there's no like easy access if you're not actually building

49:48 Play from here something and using your API key from your Gemini account.

49:51 Play from here The stuff you're talking about where I'm just going to pick out a template, I'm going to

49:54 Play from here hook up my connectors for my Microsoft account or my Gmail account or my to-do list or whatever

49:59 Play from here it is that you're using, and then have these agentic workflows.

50:02 Play from here You're just doing it through a chat.

50:04 Play from here You're not digging into your settings.

50:06 Play from here You're not building anything or hosting anything.

50:08 Play from here But I'm curious as to, is that completely analogous to what is happening in Cowork right

50:13 Play from here now?

50:14 Play from here And if not, what's the difference?

50:15 Play from here And why do you find the OpenAI version easier to use?

50:20 Play from here One of the, and just full disclosure, I've been meaning to, I haven't had time.

50:23 Play from here I do want to play with those agents, but yeah, Cowork has been great for me and I'm

50:27 Play from here on, you know, like paying the a hundred bucks a month.

50:29 Play from here So I want to burn my credits.

50:32 Play from here So one is that GPT is in the cloud.

50:35 Play from here So is, is more seamless as far as connecting different connectors to do things from, just

50:43 Play from here to be clear, as opposed to running on your machine in a local environment, right?

50:47 Play from here Like Cowork does.

50:48 Play from here I can only use Cowork.

50:50 Play from here Cowork is only going to operate if I do have this sequence of tasks that I want it to undergo.

50:55 Play from here If my machine is awake and it's running and the cloud app is running on my machine.

51:01 Play from here But in GPT, that's not the case for the agents.

51:04 Play from here That's not the case.

51:05 Play from here That's not the case.

51:06 Play from here It is slightly less technical.

51:08 Play from here I would say there's just, you're going to laugh at me for using the term fiddly again,

51:11 Play from here but Cowork's a little bit fiddly.

51:13 Play from here Like you've got to, you've got to like kind of connect this thing and that thing.

51:17 Play from here And then you're running these local projects and they have their own like timelines.

51:22 Play from here It's a little confusing.

51:24 Play from here If you want that granular control as far as like turning all the dials and things, which,

51:32 Play from here you know, I, I do like it's, it's, it's really cool to be able to do that.

51:36 Play from here But I honestly think that, that, that open AI has just a clearer idea of making this

51:42 Play from here accessible to people who don't want to fiddle around.

51:44 Play from here They just want to solve whatever that problem is that they have and do so really easily.

51:49 Play from here You know, I built one last week that if anybody does anything with marketing, you're constantly

51:55 Play from here resizing images.

51:57 Play from here And it's painful because just resizing an image is you're not just like skewing it and

52:02 Play from here stretching it like to be a different dimension.

52:06 Play from here It might actually change the internal layout.

52:08 Play from here Well, with GPT 2.0 image, like the image quality of GPT finally hit a point where it's great.

52:16 Play from here So imagine you're doing a text ad and your text ad has like, you know, it has a bunch

52:21 Play from here of graphics or whatever it is in it.

52:23 Play from here If you want to make it portrait vertical versus landscape, like you actually have to redesign

52:29 Play from here the thing to make it look good.

52:31 Play from here But building an agent that takes a one by one, so an even ratio like ad and then constructs

52:38 Play from here all of the downstream ads that Facebook slash meta wants as far as it's, it's library.

52:45 Play from here It's great.

52:46 Play from here It's and you can do that because they have an image generator and, and Claude does not

52:50 Play from here have an image generator.

52:52 Play from here And so that's, people don't recognize that because they think that they're, they're we're,

52:57 Play from here we're getting into like multimodal like models and things like that.

53:00 Play from here But Claude deliberately or strategically for whatever reason or financially never leaned

53:06 Play from here into having their own image model.

53:08 Play from here So they have the ability to interpret data that they're given in terms of screenshots

53:13 Play from here and such, but they can't actually generate that output.

53:18 Play from here So if you were using coworking and you wanted to do something like this, you would be forced

53:23 Play from here to create a connector to go to Nano Banana at Gemini or another provider to create that

53:31 Play from here image.

53:32 Play from here And it was, it was complicated because now you have multiple providers and multiple steps.

53:36 Play from here This works really cleanly in open AI.

53:39 Play from here This is just one use case, but it was click, click easy for me to get to a point where

53:44 Play from here I start with an ad concept, get a single, a single image, and then have it iterate such

53:49 Play from here that it has all the images.

53:52 Play from here And then actually it can interact with your computer as well.

53:56 Play from here So then I move all of those images to a folder on my desktop such that I can then laboriously

54:03 Play from here and annoyingly have to upload them one at a time to Facebook.

54:06 Play from here So even though that's all running in the cloud, you can still give it access to your local

54:10 Play from here file structure on your machine.

54:12 Play from here So you kind of get the best of both worlds there.

54:14 Play from here Okay.

54:15 Play from here That's pretty interesting.

54:16 Play from here So what you said about the ad creatives, ad creatives obviously usually have a bunch of

54:21 Play from here text on them.

54:22 Play from here They're not just, you know, an image that you're resizing.

54:24 Play from here And with the GPT image generation, that was one of the big pieces of news that just how

54:30 Play from here brilliantly it would reproduce text without any misspellings at the right size.

54:37 Play from here Just right out of, right out of the gates, even really tiny text, little diagrams, just

54:42 Play from here I saw a bunch of side-by-sides with Nano Banana, which was best in class for sure.

54:47 Play from here It's markedly better.

54:48 Play from here It's really good.

54:49 Play from here It's markedly better.

54:50 Play from here I mean, I would, I would, I would just go as far to say some of the outputs I saw were

54:53 Play from here like 20 to 30%.

54:54 Play from here And again, it's, this is about friction.

54:56 Play from here Like you and I are embracing the friction because we're on the cutting edge of this

55:00 Play from here stuff.

55:01 Play from here Right.

55:02 Play from here But most organizations and most people just need this stuff to work and being able to,

55:06 Play from here if you have an existing GPT account, being able to know that it's just sitting there

55:11 Play from here in your left bar and the admin in the account needs to like give access to the things that

55:17 Play from here it needs access to.

55:18 Play from here But this is absolutely somewhere where you should be experimenting on sort of creating

55:24 Play from here your own agentic harnesses going forward.

55:26 Play from here Yeah.

55:27 Play from here And we, we've spoken a lot about our chief of staff.

55:29 Play from here I think we're going to have to dedicate a whole chief of staff episode coming up here

55:32 Play from here soon as what we're working on matures, because I want to talk a lot about that.

55:36 Play from here But that use case you said with the images all thrown out that I just did last night,

55:39 Play from here I did this one in coworking.

55:40 Play from here I'm curious to try to replicate it in a GPT to see what happens.

55:44 Play from here This is a little silly, but on Instagram, I will get a steady stream of these crazy

55:49 Play from here high protein recipes, like 50 grams of protein you can make in under 30 minutes.

55:53 Play from here And I'm like, Oh, save, save.

55:54 Play from here I never make any of them.

55:55 Play from here So just living in my Instagram account.

55:57 Play from here And one of the really annoying things about Instagram is they've disabled all sorts of

56:01 Play from here copy and paste, whether you're on desktop or the mobile app.

56:03 Play from here So the recipe you want is in the captions, but you can't actually get to it.

56:08 Play from here So what you end up doing is trying to make some salmon thing and grabbing your phone

56:11 Play from here with like raw fish juice all over your hands, trying to reopen Instagram, and then it defaults

56:15 Play from here back to your feed and you're trying to find the post.

56:18 Play from here It's awful.

56:19 Play from here So I, you know, I found an extractor is just a free tool where you'd put in the Instagram

56:22 Play from here URL and it extract all the text from that post and it's free.

56:26 Play from here I'm like, Oh, this is awesome.

56:28 Play from here Then I could copy and paste that with the original link and put it in an Apple note

56:32 Play from here or something and then print it out or build a, Oh my God, this is really poor use of my

56:38 Play from here time to make some high protein foods.

56:41 Play from here So last night, and this took about maybe 10 minutes in co-work, I just said, Hey, here's

56:46 Play from here what I want to do.

56:47 Play from here These are all my saved videos.

56:49 Play from here Here's the link.

56:50 Play from here Here is this extractor tool.

56:51 Play from here Here's the link.

56:52 Play from here Here's my ClickUp account.

56:54 Play from here Here's the workspace.

56:55 Play from here I want you to build me a library of these 50 high protein meals that I've saved in my

57:00 Play from here Instagram account.

57:01 Play from here And I want you to tag them breakfast, lunch, dinner, dessert, snack.

57:07 Play from here I want you to put in the estimated time that it takes to make each one as one of the fields

57:12 Play from here inside of ClickUp.

57:13 Play from here And then I just went to bed.

57:15 Play from here And this morning I had a complete ClickUp library that's now connected to my chief of

57:19 Play from here staff.

57:20 Play from here So as part of my kind of weekly plan that it delivers, it gives me my meal plan for

57:24 Play from here the week.

57:25 Play from here All in these print friendly things.

57:27 Play from here And again, like I wasn't curing cancer or anything with this genetic workflow, it's

57:33 Play from here really, really convenient.

57:34 Play from here And honestly, it would have taken me like a few hours of my time.

57:39 Play from here And now I can just say, you know what, run this every Friday night and go and look at

57:44 Play from here any new videos that I've saved and add them to my library of recipes.

57:48 Play from here So a very not AI at work example, but just kind of bringing it back home.

57:52 Play from here Like these are the things like anything that you use your computer for, where you are cutting

57:56 Play from here and pasting stuff back and forth or having to dip in and out of half a dozen different

57:59 Play from here applications to get the end result you want.

58:01 Play from here These are the things that are possible now that agents can solve and just do for you

58:05 Play from here while you're sleeping.

58:06 Play from here So for people listening, the sort of the homework is if you're on GPT, go open up agents, look

58:12 Play from here on the left bar.

58:14 Play from here It's going, there's going to be a template for chief of staff, create a chief of staff

58:18 Play from here agent.

58:19 Play from here I frankly need to do this this week myself because the agent that I've built in co-work

58:23 Play from here is a monster and like maybe it's too complicated.

58:27 Play from here And I don't want to try it.

58:28 Play from here It only works when your machine's running.

58:29 Play from here Right.

58:30 Play from here It only works when my machine's running.

58:31 Play from here So my machine is running all the time.

58:33 Play from here That's a different story, but like I want to try it again with this more accessible

58:37 Play from here approach with open AI.

58:39 Play from here just see how that plays out as far as which is more useful.

58:43 Play from here If you want to be highly technical

58:45 Play from here and be able to turn all of the dials,

58:46 Play from here it's going to be co-work.

58:48 Play from here If you want to very quickly accomplish something

58:52 Play from here that's very visual and very clickable,

58:54 Play from here I suspect that at the moment OpenAI has that nod,

58:57 Play from here but we'll have to report back.

58:59 Play from here Yeah, I think so.

58:59 Play from here That's going to be our homework for this coming week.

59:01 Play from here I think we're both going to have to rebuild

59:03 Play from here our Chief of Staffs in GPT

59:05 Play from here and see how it works compared to our cloud versions.

59:07 Play from here The other note that we're running into,

59:09 Play from here and I was texting with you last night,

59:11 Play from here is this stuff is not universal.

59:13 Play from here All of the connectors aren't there.

59:15 Play from here You said earlier OpenAI and Microsoft are really close.

59:18 Play from here Well, guess what?

59:19 Play from here You have a cloud add-in for Microsoft

59:22 Play from here and there isn't an OpenAI add-in yet.

59:24 Play from here I mean, that will happen.

59:26 Play from here This stuff moves really fast and we want it right away,

59:29 Play from here but right now I'm dealing with really smart people

59:33 Play from here who are trying to build really cool, cool Chiefs of Staff,

59:37 Play from here but if they're in the Apple universe

59:39 Play from here or if they're in the Microsoft personal universe,

59:43 Play from here not the corporate universe,

59:45 Play from here but you have an Outlook-type domain,

59:48 Play from here it's really hard to connect this stuff.

59:51 Play from here And if you can't get your email and your calendar

59:53 Play from here connected to something called a Chief of Staff,

59:58 Play from here all of a sudden you're back in fiddly land.

01:00:00 Play from here And my advice, if you run into that,

01:00:03 Play from here if you're not super gung-ho, just chill.

01:00:06 Play from here Just wait a week.

01:00:07 Play from here Just wait a month.

01:00:09 Play from here And these connectors are gonna snap into place

01:00:12 Play from here versus burning a weekend.

01:00:14 Play from here Not that most people are gonna burn a weekend.

01:00:15 Play from here I'm starting to make it a point to just browse connectors.

01:00:17 Play from here Just to see.

01:00:18 Play from here You're not gonna see a press release on there.

01:00:20 Play from here It's like, oh, they added Apple Notes and this and that.

01:00:21 Play from here Yeah, and so Apple iMessages was the one that we noted.

01:00:25 Play from here So now you can tie in your iMessages.

01:00:27 Play from here That I've been doing for two weeks now.

01:00:29 Play from here So my Chief of Staff, I get an iMessage every morning.

01:00:31 Play from here It's where I spend most of my time, right?

01:00:32 Play from here Like I'll miss it sometimes in Telegram or Slack.

01:00:34 Play from here Yeah.

01:00:35 Play from here And then I'll get an iMessage.

01:00:36 Play from here I'm like, oh, a text from myself

01:00:37 Play from here with my whole day and my calendar

01:00:38 Play from here and the emails I need to attend to in my open tasks.

01:00:40 Play from here Cool.

01:00:41 Play from here All right.

01:00:42 Play from here So again, it's a challenge for us

01:00:45 Play from here to both be on the edge of this and experimenting.

01:00:47 Play from here And we're building in public and we're failing in public

01:00:50 Play from here as we're building these things.

01:00:52 Play from here And then calibrating back to what's gonna be useful

01:00:56 Play from here to the folks who are listening to this.

01:00:58 Play from here And particularly if somebody comes in to our audience today

01:01:02 Play from here and we need to constantly make sure

01:01:05 Play from here that we're reiterating and kind of,

01:01:07 Play from here every time we take three steps forward,

01:01:08 Play from here we need to take two steps back.

01:01:10 Play from here Bring everybody along with us.

01:01:11 Play from here Yeah.

01:01:12 Play from here Yeah, it's moving quickly, buddy.

01:01:13 Play from here Well, until next week,

01:01:15 Play from here I guess we'll trade Chief of Staff notes.

01:01:18 Play from here All right.

01:01:18 Play from here All right, bud.

01:01:19 Play from here All right.

01:01:20 Play from here All right.

01:01:21 Play from here All right.

01:01:22 Play from here All right.

01:01:23 Play from here All right.

01:01:23 Play from here All right.

01:01:24 Play from here Thanks for listening.

01:01:25 Play from here If you'd like to send us a question to answer on the show,

01:01:27 Play from here email me at Elijah at spark6.com.

01:01:30 Play from here That is S-P-A-R-K, the number six.com.

01:01:33 Play from here Or Kevin at Kevin at ascendlabs.ai.

01:01:37 Play from here To learn more about AI strategy

01:01:39 Play from here and training for your business,

01:01:40 Play from here visit Kevin's company, Ascend AI Labs at ascendlabs.ai.

01:01:46 Play from here And to learn more about design and development

01:01:48 Play from here of custom AI solutions,

01:01:50 Play from here visit my agency at spark6.com.

01:01:53 Play from here You can also sign up for my free newsletter,

01:01:55 Play from here AI Framework Friday,

01:01:57 Play from here by visiting spark6.com forward slash newsletter.

01:02:01 Play from here All right.

01:02:02 Play from here Thanks for listening.

01:02:03 Play from here I'm Elijah.

01:02:04 Play from here I'll see you next week.

01:02:04 Play from here Bye.

How to cite this episode

Credit the named speaker and AI at Work by AscendAI. Link to the episode or the passage you quoted.

“Ep #25 When Your Team Says Claude But Means GPT.” AI at Work by AscendAI, 2026-05-06. https://ascendlabs.ai/podcast/ep-25-when-your-team-says-claude-but-means-gpt/

Recent episodes

View all episodes
×
AI Solutions Graphic
Free Resource
Get Started Now