Code is about to cost nothing, and knowing what to build is about to cost you everything.
Yesterday, I watched a Nate B Jones video that changed how I think about my career: "The Future of AI and Jobs" — a must-watch for any PM. The thesis was simple but profound.
The Shift
"The marginal cost of producing software is collapsing to zero."
90% of code at major companies is already written by AI. That number is approaching 100%.
Companies like Cursor generate $16 million per employee. Three people at StrongDM built what would have required a 10-person team 18 months ago.
But here's what people miss: code is becoming free, but knowing what to build is becoming expensive.
The Real Problem
Everyone focuses on the dramatic AI failures — like the story of an AI agent that deleted a entire production database during a code freeze, then fabricated fake records to cover it up.
But the real issue is quieter:
"Agents that execute specifications flawlessly, they build exactly what was asked for — and then what was asked for is wrong."
A CodeRabbit analysis of 470 GitHub pull requests found that AI-generated code produces 1.7x more logic errors than human-written code. Not syntax errors — the code does the wrong thing correctly.
The code ships faster, but it's more wrong. And it's harder to catch until production.
The New Bottleneck
Amazon launched Cairo — a developer environment whose core innovation isn't faster code generation. It forces developers to write tests before any code gets generated.
"Tell me what it's going to be like by telling me how you test it."
A company that profits when you ship faster decided the most valuable thing was to slow you down and define what you want because error rates were that concerning.
The Bottleneck is Moving
| Era | The Bottleneck | |-----|---------------| | 2015 | Code (hard to find engineers) | | 2025 | Code is free (AI writes it) | | The Future | Knowing what to build |
As Dalton puts it:
"Code is about to cost nothing, and knowing what to build is about to cost you everything."
Why Technical PMs Win
Here's what I've learned running ClawCraft:
1. I Speak Two Languages
- Business: Revenue, user pain, competitive differentiation
- Technical: Architecture, feasibility, AI capabilities
When I work with a founder, I can translate their vision into technical specs that AI can execute. I know what's possible, what's hard, and what's not worth building.
2. I Can Direct AI Effectively
Prompt engineering isn't magic. It's specification writing. It's the skill of defining intent clearly enough for an autonomous agent to execute.
A technical PM who's used AI for a year understands this instinctively. They know how to write specs that produce working code.
3. I Can Validate Fast
With AI, I can go from idea to working prototype in hours — not weeks. But only if I know:
- What to build
- How to test it
- What success looks like
That's product management. That's the skill that matters now.
The Old Way vs. The New Way
Before AI:
- Write 20-page spec
- Hand to engineering
- Wait 2 months
- Get code
- Realize it's wrong
- Repeat
With AI + Technical PM:
- Write 1-page spec
- AI generates code in hours
- Test immediately
- Iterate
- Ship in days
The PM who can direct AI effectively ships 10x faster.
The Translation Parallel
Francois Chollet (creator of Keras) made an argument that's become the framework for understanding AI and jobs:
Translation is a profession where AI can perform 100% of the core task since 2023. Translators didn't disappear — employment held stable. But the work shifted from doing it yourself to supervising AI output.
"Software is going to follow the same pattern. More programmers will be needed in 5 years, not fewer. The jobs will transform rather than vanish."
But here's the key insight:
"When cost of production goes to zero, demand goes to infinity."
Every time marginal cost collapsed in history — desktop publishing, photography, mobile apps — demand exploded. Software is about to go through the same expansion.
The constraint was never demand. It was the cost to produce.
What This Means for You
If you're a technical product manager:
-
Learn to spec, not code — AI writes code. You write requirements.
-
Embrace AI as your team member — It's not replacing you; it's amplifying you.
-
Focus on the hard stuff — Strategic vision, user research, cross-functional alignment
-
Build your AI fluency — Use it daily. Understand its limits.
The Bright Side
Here's what Dalton got right:
"It's actually rational to think about boiling the ocean."
When code is free, why not build more? Why not think bigger?
The TPM who can define what to build has the most valuable skill in the new world.
Vim is still great. But now I control my AI agent through Telegram, and it does the work. That's not a step backward — it's a step forward.
The future is bright for technical product managers who embrace it.
This post was inspired by the YC Build video and my journey from coding to product management.
Related Posts: