I wrote a chapter on imposter syndrome in my book, Think Like a CTO, a couple of years ago. I thought I had said everything worth saying on the subject for the technologist. I had read the research. I had talked to dozens of CTOs who felt the same way. I had built the framework for recognizing it, naming it, and moving past it.

I believed I had it handled. Then AI happened.

The chapter I wrote was about the pre-AI version of imposter syndrome. The version where you look around a room of your peers and wonder if you belong there. The version where a new technology emerges and you feel a little behind until you remember you have twenty years of context that the new thing cannot replace. The version where your accumulated knowledge is still your armor.

That version feels quaint now.

I say this not as someone who observed the shift from the outside. I say this as someone who feels it. I suffer from imposter syndrome. I have for most of my career. I wrote the chapter to help others, but I also wrote it to help myself. And I have to be honest. The last two years have made the coping mechanisms I wrote about feel insufficient.

I was in a room with many CTOs last month and in quiet hushes they noted the same sentiment: “I feel like I should be doing more“. Not about their companies. About themselves. About keeping up.

This is a room full of people who have run engineering organisations for a decade or more. People who survived the dot-com bust, the 2008 crash, the crypto winter. People who have built teams that shipped products used by millions. And they are sitting there, genuinely worried that they are falling behind.

Not because they lack skills. Because the ground is shifting under them in a way that feels personal.

I feel it too.

The history of optional

Every major technology wave of the last thirty years had an opt-out option. You could skip it and your career was fine.

Web in the 90s? You could stay in enterprise, mainframes, embedded systems. Your knowledge was still worth something. You were the person who understood the database when everyone else was busy learning CGI scripts. Twenty years of COBOL experience got you paid more, not less.

Cloud? You could stay on-prem. You could specialise in security, compliance, performance. You could own the infrastructure layer that cloud had not replaced yet. There was a lane, and it was well lit. People spent entire careers on Solaris admin and never felt obsolete.

Mobile? You could stay backend. Nobody was telling the API architect they needed to learn Swift. The app devs needed you more than ever. Your REST design skills translated directly. Nothing was lost.

Crypto and web3? You could ignore it completely. Most of us did. No career damage. In fact, ignoring it was probably the smartest move you could make.

Every wave had a “not my problem” escape hatch. It kept your accumulated knowledge intact. It meant your twenty years of experience still made you the most valuable person in the room for the work that mattered. The investment you made in learning payloads, transaction rollbacks, thread safety, distributed consensus that stuff never expired. It just accumulated.

AI does not have that escape hatch.

Why this one is different

There is a reason this wave hits differently. It is not about learning a new syntax or a new deployment target. It is not about swapping out one database for another. It is about the thing you built your entire professional identity on being rendered less valuable by a tool that costs $20 a month.

The Clearing’s 2025 AI Fatigue Survey of over 2,400 engineers found something that should stop you short. Engineers with 6 to 10 years of experience showed the highest fatigue rate at 72%. Not juniors. Not people fresh out of bootcamps. The mid-career cohort who spent a decade building their craft. The study called it a psychologically distinct burden. These engineers spent years building the ability to ship complex systems from memory. AI shortcuts that process feel like evidence that those years of investment are becoming obsolete.

Entry-level engineers showed 31% fatigue. They have not built the muscle yet, so they have less to lose.

Staff and principal engineers showed 44%. They have enough organizational power and context to feel insulated.

The people in the middle, the ones who built their identity on being the person who can execute without support, are taking the hardest hit. And this is the cohort most organisations rely on for delivery. Your senior engineers, your team leads, your architects. The people who actually ship your products.

Of over 8,000 professionals, found that 64% of tech professionals have experienced imposter syndrome at some point in their careers. Technology beats accountancy and finance. Beats engineering. Beats construction. The more senior you get, the more experience you accumulate, the more you feel like a fraud.

A peer-reviewed study presented at ICSE 2024 put a finer point on it. Out of 624 software engineers across 26 countries, 52.7% experienced frequent to intense levels of imposter syndrome. Women suffered at a significantly higher rate at 60.6% compared to men at 48.8%. Asian engineers at 67.9% and Black engineers at 65.1% reported higher rates than white engineers at 50%. The prevalence had a statistically significant negative effect on perceived productivity across every dimension measured.

The baseline was already high. AI poured fuel on the fire.

The mechanics

Here is how the loop works.

1. Step One
   You use AI to generate code. It works. You feel productive. But something is off. The code works but you did not write it. The sense of authorship fades.
2. Step Two
   You notice you are reaching for the tool more often. Things you used to write from muscle memory now require prompting first. You feel a twinge of anxiety. You generate faster than you can verify. You start accepting output you would have rejected from a junior developer.
3. Step Three
   You stop trusting your own ability to solve problems without the tool. When workers had high confidence in AI doing the task, their perceived enaction of critical thinking dropped significantly. The more you trust the tool, the less you engage your own brain.
4. Step Four
   You try to compensate by learning more. Every new framework, every new tool, every new release adds to the pile. The senior you get, the worse it gets.
5. Step Five
   The learning compounds faster than the application. You are collecting knowledge you never ship. The gap between what you know and what you actually do widens. This feeds the feeling that you are falling behind. You try harder. You burn out.

One reason the loop is so hard to break is that the market is designed to keep you in it. And the data on tool launch velocity is frankly absurd.

The cost of attention is invisible

Product Hunt data spanning 267,000 launches between January 2020 and January 2026 shows that AI-related launches went from 5% of all launches to nearly 40% in late 2025. In a single month, January 2026, 4,295 AI products launched. Total launch volume nearly doubled year over year. December 2025 saw 9,903 total launches compared to 4,444 in December 2024, a 123% increase.

The AI/TLDR index that tracks AI releases shows 60 new releases per week. Google ships every 2.8 days. Mistral every 6.1 days. Meta every 6.4 days.

You cannot evaluate this volume. You cannot even skim it. And here is the part nobody talks about. The cost of attention is not zero. Every hour you spend evaluating a new AI tool is an hour you did not spend shipping, thinking, or resting.

The difference is not the quality of the tools. It is the pattern of attention. The more hours you spend in the AI tool evaluation loop, the more exhausted you become, regardless of what you are actually learning.

The classic Stack Overflow 2025 survey confirms this indirectly. Only 33% of developers trust AI accuracy. Professional developers are the most skeptical. Despite this, developers are spending more time than ever on AI tooling. The disconnect between trust and attention is a signal. You are spending time on something you do not trust. That is not learning. That is anxiety.

The analysis found that 2,027 products have already pivoted to add AI positioning after originally launching without it. The median pivot time was 11.5 months. Half of these products took less than a year to decide their original angle was not working and rebrand as AI. The AI label is marketing as much as it is substance. Sixty-nine percent of AI-positioned launches use generic AI branding without mentioning specific models, techniques, or providers.

What you actually lose while chasing

The tools you are panicking about not knowing are, in many cases, built by companies that have no idea if they will exist in two years. The frameworks being hyped on Hacker News today are the abandoned GitHub repos of 2028.

55% of engineers fall into the significant or severe fatigue tier. Only 28% are in the mild tier. And recovery is not straightforward. Vacation barely helps. Only 29% found time off effective. The problem is not energy depletion. It is skill atrophy.

The academic research confirms the stakes, that the presence of imposter syndrome had a statistically significant negative effect on perceived productivity across every dimension measured. Satisfaction and well-being. Performance. Activity. Communication and collaboration. Efficiency and flow.

The study also found that imposter syndrome was less common among engineers who were married and had children. The hypothesis is unflattering but important. If your entire sense of worth comes from being the technical expert, erosion of that expertise feels existential. If you have other sources of meaning, it feels like a problem to solve.

AI has come along and put into question your very professional purpose.

1,100 developers found that AI already accounts for 42% of committed code, expected to reach 65% by 2027. But 96% of developers do not fully trust AI-generated code. Only 48% always verify it before committing.

Think about that. We are shipping nearly half our code through a tool we do not fully trust. And 38% of developers say reviewing AI-generated code takes more effort than reviewing human code (thanks to the sheer volume of it).

The productivity boost people promised is not a clean win. It is a trade. You get speed on generation but you pay for it on verification. The bottleneck did not disappear. It moved.

We are shipping nearly half our code through a tool we do not fully trust.

The person who can verify code, who knows when the AI is confidently wrong, who can look at a generated solution and say that works but it will break in production in six months; that person is still the most valuable person on the team.

The verification skill is harder to build than the generation skill. It requires everything you already have. Your experience. Your pattern recognition. Your scar tissue. The AI can generate a thousand solutions. It cannot tell you which one will survive a database migration, a team turnover, and three years of maintenance. That requires judgment. That requires having been wrong before.

This is the real cost. Not that you are falling behind. That you stop exercising the skills that made you valuable in the first place.

Breaking the loop

The Clearing survey identified the interventions that actually work:

• Retrieval practice at 84% effectiveness
• AI-free days at 79%
• Morning deep work blocks at 76%
• Explicit skill tracking at 71%
• Tool reduction protocols at 68%
• Peer programming with humans at 62%
• Vacation at 29%.

The pattern is clear active cognitive engagement without AI beats passive rest. The problem is not that you are tired. It is that you have stopped using parts of your brain that made you effective. Recovery means using them again.

The engineers who break the loop do not try to learn more. They learn less, more deliberately. They carve out time to work without assistance. They treat AI as an intern they supervise, not a brain they rent.

The most effective recovery intervention was retrieval practice – working without AI. Deliberately doing the hard thing without the crutch. Not to prove you can. To keep the skill alive.

The same survey found that light or sporadic AI users showed only 18% fatigue compared to 79% for continuous users. The difference is not the tool. It is the relationship with the tool.

What to do

First of all take a huge deep breath. You are not alone.

Stop competing with the churn. You cannot out-learn the market. Nobody can. The people who seem like they are keeping up are either lying or burning out. There is a lot of buzzwords flying around making people sound infinitely more intelligent than they are.

Start competing with yourself.

• What do you know that the AI does not?
• What patterns have you seen that it has not?
• What bad ideas have you already shipped and learned from?

The thing you built your career on is not obsolete. It has moved from execution to verification. From writing to judging. From knowing the answer to knowing which answers are wrong.

Here is your instruction. Pick one thing this week that you would normally use AI for. A code review. A design decision. A debugging session. Do it without AI. Not to prove you can but to keep the muscle alive.

If one task feels too small, take a full day. The first hour will feel slow. You will reach for the tool. Do not. By hour three, you will remember what it felt like to be in flow state without assistance. By the end of the day, the loop will have loosened its grip.

The tools will still be there tomorrow. Your skills might not be if you stop using them.

And keep breathing.

AI Disclaimer: Gemini Nano Banana Pro was used to generate the photo – Good Will Hunting (1997)