Give ten people the same job and ask each to build an agent that does it. You’ll get ten agents and ten different results.
Matt made that point on this week’s episode, and it’s stuck with me since. He used Alex Lieberman as the example. Alex is very good at storytelling and content. If Alex and a junior employee both automated the same content job, the prompts, guardrails, and thinking Alex would program in would look nothing like what the junior person would produce. Matt put it plainly: he’d buy Alex Lieberman’s setup over Joe Schmo’s.
That changes how leaders should think about which roles to protect.
Agents are good at clicking. They copy whatever you tell them good looks like.
Matt and I both run what he calls long-tail days. We touch a lot of different systems for a lot of different reasons. Most of those tasks don’t take much thinking. Make a virtual card in Ramp. Categorize some bookkeeping. Agents are getting very capable at that kind of work, even if authentication and 2FA still trip them up constantly.
The narrower, judgment-heavy jobs are different. Video editing. LinkedIn posts. SEO content. Here the agent’s output depends on whether someone told it what makes a clip good or bad, which tone to use, how to open and how to close. Matt’s line was that the best outcome comes from the person who understands what makes a great end product and can articulate it to people and to their agent.
That’s taste and discernment. Those are very human things at the moment.
There’s another route. You can build something that scans 10,000 Facebook ads, finds the patterns that perform, and replicates them through the hard law of statistics without caring why they work. I’ve seen that approach. The trouble is that an LLM is always backwards looking. It’s trained on what was cool six or twelve months ago. Keep feeding the grinder and you end up with mediocrity. You need fresh takes, and those come from people.
I compared it to the olfactory experts whose noses get insured for huge sums. People with that kind of taste are going to be employable forever.
Finding your taste people is the hard part
I pushed back on Matt a little here, because wishing for an Alex Lieberman in every department doesn’t make one appear. Most organizations don’t teach the skill of distilling a process into something an agent can run. They don’t hire for it. Just because someone does a job well doesn’t mean they can explain how. Companies go hire bright young people who have vibe coded something, and those people often have no experience turning work into a repeatable process.
At Ascend Labs we hire process engineers and people with Six Sigma backgrounds, and even they have to be curious about technology to make this work.
So the practical questions for a leader are these. Who in your organization has the taste? Who has the technical skills? You might get lucky and find both in one person. You probably won’t. Then the job is pairing them so they can cook up something that fits your business.
Don’t let everyone build their own
This is where I’ll be a buzzkill about the dream that tools like Lovable are selling. Letting every person in your company build their own AI solution sounds empowering on a slide. In a small organization, fine. In a large one, they’ll all do it differently, and you’ll get performance gaps and cost gaps. Align on how your organization builds before you scale it.
There’s a related problem I see with clients. Two departments build similar AI capabilities without knowing about each other, because nobody keeps a central record of AI projects. The go-to-market team builds something that distills communications into messaging, and the same plumbing would work elsewhere, but the left hand doesn’t know what the right hand is doing.
Where the taste conversation fits for most companies
I was in Charleston last week with 17 CEOs at a Vistage meeting. It was the first time every person in a room like that was leaning in. Owners of a big paving company and an HVAC company had spent years telling themselves people would always need paving and HVAC. They’ve stopped saying that.
If that’s where you are, start small. I give people two places to look. First, where are your most expensive human mistakes? Often that’s quoting. Second, where are your most skilled people spending time beneath their ability, like salespeople spending 30% of their time on CRMs instead of customers?
Once you start automating those, the taste question shows up fast. Someone has to tell the agent what a good quote or a good follow-up looks like. Figure out who that person is before you decide whose job the agent is taking.
