When I ask audiences what they’re using AI for, the honest answer, the one that cuts through the excitement and the strategy decks and the pilot program announcements, is that most people are still opening a chatbot, typing a question, and closing the tab.

That’s not a failure. It’s a starting point. But it’s worth being clear about where most organizations actually are versus where the technology actually is, because the gap between those two things is where most AI strategies are going wrong.

Listen to the full episode: https://youtu.be/3JhQEUYqCCo?si=bioMYsVOFEBpuOCe

The Four Levels, A Map That Actually Helps

In a recent conversation with Eli, we mapped out the four ways AI actually gets used at work. Not in theory. In practice, right now, across the organizations we’re working with and inside our own workflows.

Level one is prompts. You open a chatbot, Claude, GPT, Gemini, whatever you have access to you ask a question or describe a problem, you get a response, and you close the tab. This is where the majority of people and the majority of organizations are operating today. It’s useful. It’s genuinely valuable. And it’s the floor, not the ceiling.

Level two is projects. The distinction here is reusability. Instead of re-explaining your context, your preferences, your documents, and your specific situation every time you open a new chat, you save that context in a project. Claude calls them projects. GPT calls them projects too, though custom GPTs covered similar ground before that. The idea is simple: it’s a folder with memory. You build it once, you update it as things change, and every conversation inside it starts from a richer, more contextual place.

Level three is automations. Fixed workflows running on a trigger or a clock. When this email comes in, do this. Every morning at 8am, run this sequence. Tools like Zapier, Make, and N8N have been the home of level three for years connecting apps, moving data, executing rules. This is where the more technically sophisticated teams have been living. It’s powerful. And until recently, it was close to the ceiling for most non-developer teams.

Level four is agents. And this is where the map changes.

What Level Four Actually Looks Like From the User Side

Six months ago, if you wanted to automate a specific workflow let’s say: check Gmail daily, find invoices, parse the relevant information, and forward it to your bookkeeper via Slack you were looking at a meaningful technical project. Building nodes in N8N. Configuring API connections. Specifying the logic for each step. Understanding enough about how the pieces fit together to make the whole thing work.

Today, you describe it in plain language.

“Check my Gmail once a day. Find invoices. Parse them. Forward to my bookkeeper in Slack.“

And the agent handles what comes next. It doesn’t need you to figure out the connections. It figures them out and asks permission at each step.

“I need to connect to your Gmail. Is that okay?” You say yes. Done.

“You want that to go to Slack. Should I connect to Slack?” You say yes. Done.

“You want it to be able to read PDF attachments. I’m going to add a skill for that.” It adds the skill. Done.

Code is happening behind the scenes. For most people and I mean most, not just the non-technical ones it doesn’t matter that it’s code. You’re watching a text box build the thing you described, asking you to confirm each connection, and handling the technical layer without asking you to understand it.

That’s an agent. That’s what level four looks like from where the user is sitting.

Why This Changes the Conversation for Most Organizations

The practical implication of the technical barrier collapsing at level four is that the question most organizations are asking “when will we be ready to do agentic AI?” is the wrong question. The better question is: what are the specific workflows in our organization that currently require a human to move information from one system to another, and which of those could be described in a sentence?

Because if you can describe it in a sentence, it’s probably buildable right now. Not by a developer. By whoever currently does the work.

The four levels give you a map for a conversation most AI strategies aren’t having. Not “are we using AI” most organizations are, at level one but “which level are we actually operating at, and what does it look like to move one level up?”

Level one to level two is about building reusable context. Take the prompts your team runs most often and turn them into projects with saved instructions and connected documents. That’s a half-day of setup that pays back every time someone opens the project.

Level two to level three is about removing manual repetition. Identify the workflows where a human is executing the same sequence of steps on a schedule or in response to a predictable trigger. Automate those sequences.

Level three to level four is the one most organizations are closest to and furthest from simultaneously. The tools are available. The natural language interface means the technical barrier is lower than it’s ever been. The gap isn’t capability. It’s awareness most teams don’t know that the thing they’re manually doing in N8N today can now be described in plain English to an agent that builds it for them.

That’s the gap worth closing first. Not by replacing everything at once. By identifying one workflow, describing it out loud, and watching what happens when the agent starts asking permission.

If you want a framework for mapping your organization’s current level and identifying what the next step looks like: launchpad.ascendlabs.ai

If you want to talk through where your team is on this map and what moving up a level actually requires: tidycal.com/kevinwilliams