Blog
AI Can Write the Code Now. That Makes the CTO Question Harder, Not Easier.
Every founder thinks AI shrank the gap between a senior developer and a CTO. It did the opposite. When the code is cheap, the only thing left to pay for is judgment about what not to build.
A founder told me last month that he was not sure he needed a CTO anymore.
His two engineers were shipping faster than they ever had. Cursor and Claude were writing most of the code. He could describe a feature on Monday and watch a demo on Wednesday. Why pay for an executive when the work was flying out the door?
He was right about the speed. He was wrong about what the speed meant.
The bottleneck moved, and most people are still guarding the old one
For twenty years the hard part of software was making it exist. Hiring people who could build the thing. Getting the thing to run. Getting it to run again next quarter without falling over. That was the scarce skill, and that is what a senior developer sells.
That part got cheap. Not free, but cheap. A good engineer with the current tools ships in a day what used to take a week. A founder with no engineers can stand up a working prototype over a weekend.
So the question stopped being "can we build this." It is now "should we build this, and what does it cost us in eighteen months if we do." That question was always the CTO's job. It was just easy to ignore when building was slow enough to act as a natural brake.
The brake is gone. Now you can build the wrong thing at full speed.
A senior developer asks how. A CTO asks whether.
This was always the line between the two seats. AI did not blur it. It widened it.
A senior developer is paid to answer "how do we build this." That is a real skill and it matters. But it assumes the decision to build has already been made by someone else.
A CTO is paid for the decision itself. Should this exist. Does it move the business. What do we not build so we can afford to build this well. Which of these ten things the team is excited about will still matter after the demo.
When code was expensive, a smart senior dev could pass for a CTO for a while, because the slow pace hid the missing judgment. You could not build that many wrong things in a year, so the cost of weak prioritization stayed small. Cut the build time by 80% and you do not get 80% more good decisions. You get the same number of good decisions and a pile of fast mistakes.
What I am watching happen in the rooms I sit in
I have spent the last few months in a lot of these conversations. The pattern repeats.
A team adopts AI tooling and output jumps. Everyone is thrilled for about a quarter. Then the same team is slower than before, and nobody can say why.
Here is why. They shipped six features in the time it used to take to ship two. Four of them nobody asked for. Now there are four extra things to maintain, secure, document, and explain to the next engineer. The codebase grew faster than the company's understanding of it. Speed without a filter is just debt you took on more efficiently.
The fastest way to bury a small company is to give it the ability to build everything it can think of.
The other thing I watch is shadow AI. Engineers pasting customer data into whatever model has the best autocomplete that week. Nobody decided this. Nobody banned it. There is no list of what is allowed and what is not, because the person who should own that list is the missing seat.
The questions that actually separate the two now
If you are a founder or a board member trying to tell a strategist from a strong engineer with a new title, the old technical questions will not do it anymore. AI can coach anyone through those.
Ask these instead.
- What would you stop us from building, and why? A senior dev lights up at what to build. Listen for someone who has opinions about what to kill. That is the muscle you are buying.
- Our output just tripled. What breaks first? The right answer is not "nothing, this is great." It is maintenance, security review, onboarding, and the gap between what we ship and what we understand. They should already be worried about the thing you are celebrating.
- Where is our customer data going right now? If your team uses AI tools and nobody can answer this in one sentence, you have found the exact hole the seat is supposed to fill.
- What does this decision cost us in eighteen months? A strategist thinks in the time horizon past the current demo. A senior dev often optimizes for the sprint, because the sprint is what they have been rewarded for.
None of those require you to be technical. They require the other person to think like an owner.
What to do this week
You do not need a sixty-day assessment to start. You need a couple of hours.
- Write the "what we will not build" list. One page. The features and rewrites the team keeps lobbying for that you are deliberately saying no to, with the reason. If you cannot fill the page, that is the finding.
- Find out where your data goes. Ask your engineers, today, which AI tools touch customer or proprietary data. Write down the answer. Decide what is allowed.
- Count your last quarter's features against your roadmap. How many of the things you shipped were on the plan, and how many were just easy to build now? The ratio tells you whether you have a strategist or an accelerator.
The seat got more valuable, not less
AI did not make the CTO question optional. It made building so cheap that the only expensive thing left is knowing what is worth building. That is judgment, and judgment is the one thing the tools cannot hand you.
The founder I started with still has his two fast engineers. He also has someone now whose whole job is to stand between the team and the temptation to build everything the tools suddenly let them build. He is shipping less than he was three months ago. He is also further ahead.
