I've been coding for 25 years. AI made me faster.
Not because it does the work. Because I can evaluate the work.
When I prompt a coding agent, I see the architecture in my head before the AI writes a single line. The data model. How the pieces connect. How they'll break under load. When the output looks right but isn't, I catch it.
That's taste. It comes only from doing the work yourself, badly, for years.
You've shipped code that seemed fine, then watched it collapse in production. You've structured databases that worked at 100 users and exploded at 10,000. You've chosen frameworks that looked elegant in tutorials and became maintenance nightmares six months later.
These failures build intuition you can't get from documentation. AI can only leverage what you already have.
What Almost-Good Looks Like
AI output is convincing. The code compiles. The variable names are sensible. The structure follows patterns from any tutorial. A non-practitioner would ship it.
Then it breaks in production.
"Correct" and "good" aren't the same. Code can be syntactically correct, logically functional, and still wrong in ways that only surface under pressure. Wrong architecture for the scale you're heading toward. Wrong abstraction that makes future changes painful. Wrong assumptions about how users interact with the system.
Last fall I asked an AI to set up authentication for a SaaS payment system. What came back looked professional. Clean code. Best practices.
But I'd built authentication systems dozens of times. I'd seen the edge cases that don't show up in tutorials.
The AI's code didn't handle token refresh when users had multiple tabs open. It would have worked in testing. In production, sessions would expire randomly. Users would get logged out mid-task. Support tickets would pile up. Debugging authentication issues with angry customers locked out of their accounts is a bad place to learn these lessons.
I caught it because I'd shipped that exact bug before. 2019, different project, three days tracking down random logouts. The AI gave me a starting point. My taste turned it into something I'd ship.
The Media Buying Labyrinth
The same pattern shows up in paid acquisition.
If you've never spent considerable time inside Meta Ads Manager and Google Ads, you don't understand what you're looking at. These platforms are labyrinths. Thousands of options at campaign, ad set, and ad level. Always changing. Features appear and disappear. Best practices from six months ago are now anti-patterns. The algorithm shifts in ways the platforms never announce.
If you haven't built campaigns by hand thousands of times, you're at a contextual disadvantage no amount of AI can overcome.
When you set up a Meta campaign, you make dozens of decisions that interact in non-obvious ways. Attribution windows. Optimization goals. Bidding strategies. Audience targeting. Each choice affects the others. The right answer depends on factors invisible in the interface: account history, pixel data quality, creative velocity, current state of the delivery algorithm.
A non-practitioner prompts an AI with "set up my Meta campaign for lead generation." The AI produces something that looks like a campaign. All required fields filled in. Structure matching any course.
But the audience targeting might hit the wrong people. Or the right people too frequently. The optimization goal might not match the business goal. The bidding strategy might be wrong for the account's current state. The attribution window might hide the true signal.
The dashboard shows green. The metrics look reasonable. The non-practitioner can't see the leak because they don't know what healthy looks like. They've never felt the difference between a campaign that's working and one that's just spending.
Why Now
AI output crossed a threshold. It's now convincing enough to fool non-practitioners. The code compiles. The campaigns have the right fields. The strategies sound smart. Five years ago, AI output was obviously rough. You knew you were working with a draft. Now it looks finished. The gap between "looks right" and "is right" is invisible to anyone without taste.
AI also commoditized the strategist's toolkit. The frameworks worth $500/hour are now free. AI has read every business book, studied every case study, internalized every framework McKinsey ever developed.
Bad news for people who built careers on frameworks and delegation. Who operated at the "strategic" level and never got their hands dirty. Who thought execution was someone else's job.
Their edge was never the knowledge. It was the scarcity of people who could synthesize and present that knowledge.
That scarcity is gone.
The Practitioner's Advantage
If you're a doer, someone who developed hard skills through years of practice, this is your moment.
You now have an army of intelligent agents ready to execute. You know how to delegate because you've done the work yourself.
The bottleneck for practitioners was always time. You had the skills, the taste, the judgment. But only so many hours in a day. You couldn't scale yourself. You either stayed small or hired people and spent time managing instead of doing.
AI breaks that constraint. Not by doing the work for you, but by handling the first 70% while you provide the last 30% that matters. The part that requires taste. The part that separates "works" from "works well."
If you have deep skill, AI multiplies it. If you have no skill, AI multiplies zero.
I build in a week what used to take a month. Not because the AI writes all the code. Because it handles the parts I already know how to do, freeing me for architecture decisions, edge cases, the places where experience matters.
When I prompt a coding agent, I'm not asking "how do I do this?" I'm saying "do this, here's the approach, here are the constraints, here's what I'm optimizing for." The context comes from decades of doing. The evaluation comes from taste. The AI is a lever. My skill is the fulcrum.
The Risk
The moment you fully delegate, you start losing the taste that made you valuable. Skills decay. Pattern recognition fades. You become the strategist who can't evaluate.
The best position is practitioner-plus-AI. Hands still dirty, but with leverage. Still doing, but doing more. Still building taste, but applying it at scale.
The practitioners I know are having the best years of their careers. Building faster, shipping more, taking on work they would have turned down before.
The strategists I know are nervous. The smart ones realize their value was never the frameworks. It was relationships, trust, the ability to navigate organizations. They're doubling down on what AI can't do.
The people in the middle, the ones who thought they could operate at an abstract level without building deep skill, are discovering that the abstraction layer is now occupied by something that works 24/7 and charges pennies.
The Doers' Decade
Leverage shifted. The doers have the edge.
What's left when you can't tell good from almost-good?
The employees and consultants who skipped the doing are about to find out.
Ready to Work with Practitioners Who Do?
We're operators, not strategists. Every growth marketer on our team has built and shipped—not just advised. That's how we evaluate AI output, catch the almost-good, and deliver results that actually work.
Apply to work with us and work with a team that knows the difference.
Robbie Jack•Jan 06, 2026•7 min read
