The Real AI Skill Is Imagination

Most agencies are treating AI adoption as a learning curve problem, which tools, which prompts, which workflows, which model is newest. That stuff matters. But I think it misses the deeper shift happening right now.

The real bottleneck is not technical. It is imaginative.

The Constraint That Shaped How You Think

For most of your career, software was a fixed thing. You bought what existed. You learned the tools that were available, adapted your process around their limitations, and worked within what vendors decided to build for the average customer.

That constraint was so consistent and so long-running that it gradually shaped how you think. When you have a process problem, you probably start by asking “what tool does this?” Not “what would the perfect system for this actually look like?” You learned to think inside the box because the box was real.

That is the habit AI is now making obsolete.

Tools like Claude, Codex, and similar AI-assisted development environments are changing the math on what it costs to build a custom system. We wrote about this directly when we covered low-code AI-powered software: most agencies wait for SaaS vendors to solve their workflow problems, but you can now build exactly what your team needs, faster and cheaper than any generic product will ever deliver. And you can build it for yourself. A brief intake form that actually reflects how your team scopes work. A QA checklist that matches your actual delivery process. A margin calculator that thinks the way you do about utilization and overhead. A lead tracking system or a content calendar built exactly around your own marketing process, which assumes, of course, that you are actually running a marketing process and not just thinking about it every few months and then getting busy with client work again. No judgment. But worth naming.

That is not a technical accomplishment. The technical part is genuinely getting easier. The harder part is knowing what you want built.

What Operational Imagination Actually Means

I want to be careful here because “imagination” can sound soft, like a creativity pep talk. It is not that. Operational imagination is a specific and practical skill: the ability to look at your own agency’s actual work and ask, with some honesty, what would make this better if we could build exactly the right thing?

That requires you to know your operations well enough to see the friction. Where does a project slow down? Where does scope drift in the same way every time? Where are your people making the same judgment call over and over, because there is no system to catch it? Where does client communication get messy because there is no shared structure?

Those are the gaps where a custom tool pays off. Not because the tool is sophisticated, but because it is yours. It fits the actual work rather than a vendor’s approximation of it.

The agencies I see struggling with this are not struggling because AI is hard. They are struggling because they have never had to think this way. They have always been tool-takers, not tool-builders. And the mental shift required is not really about technology at all. It is about being a more deliberate observer of your own operations.

Prompting Is the Easy Part

There is a lot of coaching available right now on how to prompt better, which is fine and useful. But workflow thinking is the deeper skill. A prompt can produce a draft. A well-designed system can take a client brief, extract strategic goals, generate angles, produce draft assets, and create a review checklist, all without you intervening at every step.

The difference between those two outcomes is not the quality of your prompts. It is whether you sat down and thought through what the process should actually do. That thinking is yours. AI will not do it for you.

This is related to something we keep running into on the creative side too. AI defaults to statistically average output dressed up as intentional. The agencies getting real traction are the ones leading with judgment first, not prompts first. Same principle applies to systems and workflows. If you hand AI a vague problem, you will get a vague solution. The quality of what you build depends on how clearly you can articulate what you actually need.

And that is an owner-level responsibility. Not something you delegate to a junior team member experimenting with whatever tool is trending this month.

Where the Real Work Is Right Now

Here is a practical reframe. Instead of spending your next few hours reading about the newest model release or watching a demo of a feature you might use someday, try this: write down five things about your agency’s operations that have always been slightly broken or inefficient. Not big structural problems. Small, recurring friction. The thing that happens every project kickoff. The thing that makes scoping painful. The thing that causes the same conversation with a client every time you invoice.

Now ask which of those five things could be addressed by a lightweight tool or system if you could build exactly the right one. No vendor constraints, no pricing tiers, no integrations you have to wait for.

That exercise will tell you more about where AI can actually help your business than any amount of tool comparison will. As we noted in the piece on software for the market of one, AI code assistants let agency teams build lightweight, single-purpose tools for their own workflows without waiting for a SaaS vendor to put it on a roadmap. Brief generators. Intake forms. QA checklists. The thing you have always done on a sticky note but should probably be a real system.

The tools to build them exist right now. The question is whether you can imagine what they should do.

This Is the Owner’s Job, Not the Intern’s

AI adoption tends to get delegated downward in agencies. Give it to the team, let them experiment, see what sticks. That is not the worst approach for surface-level productivity gains, and those gains are real. But the bigger opportunity lives at the operations level, and operations is an owner conversation.

What should your agency become better at over the next two years? What constraints are currently limiting how you deliver, scale, or price your work? Where does capacity disappear in ways that are hard to explain or justify? Those are the questions that operational imagination is supposed to answer.

AI can help you build the system once you know what it should do. But the business judgment underneath it, what to prioritize, what to fix first, what your agency should be known for being exceptionally good at, that does not come from a model. It comes from you thinking clearly about the business you are actually running.

The technical part of AI is getting easier every month. The imaginative part does not get easier on its own. That is the skill worth developing.

Pick one broken thing. Build something small that fixes it. See how that changes the way you think about the next one.