Fire All the Coders? What I Got Wrong After Building Five AI Apps

Over the past few weeks, I built and published five working apps. I do not have a coding background.

ChatGPT and Claude helped me think through concepts, write prompts, troubleshoot, and iterate. Replit generated and deployed the code. Some apps were simple. Others were not — one pulls live entertainment data from external APIs, one assesses economic factors influencing the price of gold, one ingests live RFP opportunities and helps decide whether to pursue them.

I would never have imagined I could do this. A non-coder can now get shockingly far.

That experience nearly led me to the wrong conclusion.

The Conclusion I Was Tempted to Draw

It was tempting to be dramatic – in fact I joked about it with friends - 'coding is easy now - I'm proof of that; therefore software developers are about to disappear'.

Fortunately, I didn’t go there. I did some research and quickly understood there's a lot more to software dev and deployment (including apps) than my little adventure.

A Learning Experience

At one point while I was coding the entertainment app, I gave Replit a poorly considered prompt. It went off and worked confidently for many minutes — burning through tokens, rewriting code, "fixing" things — based on instructions that were wrong to begin with.

Ultimately, I had to shut Replit down. I had to stop it coding. Rethink the instructions. Increase my monthly investment! It was necessary, painful learning. I had carelessly asked for the wrong thing, clearly and expensively.

That is a small, personal version of a problem Boards should care about a great deal.

The Real Board Issue: AI Produces Output Faster Than Organizations Can Safely Absorb It

In my case, the cost was tokens and a long evening. In an organization, the same dynamic plays out with developer hours, infrastructure, rework, security exposure, and decisions made on output nobody had time to properly review.

AI can help developers generate more code. But Boards should care about trusted software, not just more software. The risk is an illusion of speed — teams producing more output than the surrounding system can review, secure, test, integrate, and maintain.

So the question is not only "how many coders do we need?" It is also "can our product development system safely absorb AI-accelerated output?"

That is a serious, governance-driven, Board-level question. It’s a question that Boards have to ask.

AI Changes the Staffing Question. It Does Not Answer It.

As a director, I would be skeptical of a CEO who says either of these:

• "AI doesn't change anything."  That's too complacent and it’s not true.

• "We can cut the team because AI writes code now."  That's just dangerous.

Here’s a more credible answer: 'AI should increase output, reduce some future hiring needs, move people away from routine work, and reshape the team over time. The first serious move is capacity reallocation, then role redesign, then — once there's evidence — selective hiring restraint or reductions.'

How I'd Think About Team Impact: Four Buckets

1. Routine feature builders. Most exposed. Simple screens, standard forms, basic API calls, predictable fixes — AI compresses this work significantly.

2. Junior developers. Complicated. AI makes juniors more productive but can hide weak fundamentals. A company that stops developing juniors creates a senior-talent problem in five years.

3. Senior engineers and architects. More important, not less. They decide what should be built, how systems fit together, where risks live, and when AI output is unsafe.

4. QA, security, DevOps, data, compliance, product. These roles shift rather than shrink. Faster development raises the value of everything downstream of the code.

Questions a Director Should Be Asking

• Show me a piece of AI-generated code we caught in review. What was wrong with it?

• Are we measuring real cycle time, quality, and stability — or just perceived productivity?

• What controls exist before AI-assisted code reaches production?

• Are we using AI to reduce headcount, increase output, reduce backlog, improve quality, or speed innovation? (These are different strategies.)

• What is our redeployment plan for freed-up capacity?

Bottom Line

AI will reduce the value of some coding tasks and increase the value of technical judgment.

Cutting coders because AI generates code is a shallow conclusion. The better question is what the organization can now build faster, test sooner, improve more often, and govern more intelligently.

Trusted software still requires judgment, discipline, controls, and accountability. Ensuring that is — and always has been — the Board's role.

Jim Crocker is an AI governance consultant and board director. He writes about what boards and senior executives need to know about AI at jimcrockerai.com. Here is Jim's LinkedIn profile.