Dear Dario, Sam, and Jensen: We Are Not Your Productivity Statistic.

Every quarter, a billionaire tells us AI will replace us. Then tells us not to worry. Then tells us to “upskill.” We are tired. And we have questions.

Let me tell you what it feels like to be a software engineer in May 2026.

You wake up. You open LinkedIn. Another think piece about how AI writes 90% of code now. You get to work. You spend four hours reviewing AI-generated pull requests for subtle logic errors the model confidently introduced. You attend a meeting where someone says the word “leverage” fourteen times. You go home. You read that a CEO just said engineers are “10x more productive” because of AI. You feel neither 10x more productive nor 10x more valued. You feel exhausted and vaguely replaceable.

You scroll. Another post. “The skills you need for the AI era.” Another framework. Another roadmap. Another person who has never had to worry about their rent telling you to pivot to “systems thinking.”

You close the app.

You stare at the ceiling.

This is the feeling nobody at Davos is describing.

Let Me Frame What Is Actually Happening

Dario Amodei gave an interview recently. He said Claude writes 90% of the code at Anthropic. He said engineers are now 10x more productive. He said the bottleneck has shifted from writing code to “high-level architectural design.”

He said this from a position of having built the tool. From a reported net worth that insulates him from the labor market he is describing. From a stage, at a conference, to an audience of investors and other CEOs.

He was not speaking to you.

He was not speaking to the junior developer in Bangalore whose entry-level contract rate has fallen 40% since 2023 because clients now say “we’ll just use AI for that.”

He was not speaking to the mid-level engineer in Chicago who has been passed over for a promotion three years running because the company is “doing more with less.”

He was not speaking to the bootcamp graduate who spent $15,000 and eighteen months learning to code and graduated into a market where that specific skill is now the most automated thing in the economy.

He was speaking to other people who run companies.

And the story he told them is very convenient for people who run companies.

The Rhetoric. Let Me Name It Precisely.

What Amodei and others are doing has a structure. It is not conspiracy. It is not malice. It is something more mundane and more pernicious.

It is the reframing of a cost decision as an opportunity.

Step 1: The productivity claim.

“Engineers are 10x more productive now.”

This sounds like good news. In the framing of labour economics, productivity is value. More productive workers deserve more compensation.

But watch what happens in practice.

If an engineer is 10x more productive, a company has two choices. Pay that engineer 10x more for the same headcount. Or reduce headcount by 10x and maintain output.

Which one do you think venture-backed companies are choosing?

The productivity gain flows to shareholders and to the AI company selling the tool. The “10x productivity” talking point is a justification for doing more with fewer people. Not a promise to compensate the people doing more.

Step 2: The “skills shift” story.

“You just need to upskill. Learn systems thinking. Learn to direct AI. The bottleneck has moved from coding to architecture.”

This is true in the way that saying “the bottleneck has moved from manual typesetting to editorial judgment” was true when desktop publishing replaced typesetters in the 1980s.

It was true. And it did not help the typesetters.

The skills that are now “premium” — systems thinking, architectural design, AI orchestration — these are skills that require years of experience to develop. You develop them by doing the lower-level work first. You write the syntax before you design the system. You debug the unit tests before you architect the service. You learn by doing the “grunt work.”

The grunt work is now automated.

Which means the ladder is gone.

Not for the people already at the top of it. They are fine. They can direct AI agents all day. Their accumulated context is valuable.

But for the person at the bottom rung? The rung has been removed. And the people at the top are telling them they should be able to jump directly to the third rung.

Step 3: The reassurance that does not reassure.

“We still need humans. We always will. AI is a tool.”

This is the part I want to sit with for a moment.

Because it is designed to sound comforting. And it is technically true. And it accomplishes nothing.

Yes, humans are still needed. To supervise AI. To validate AI outputs. To catch the confident hallucinations. To provide the “direction and intent.”

These are real jobs. They require real skill.

But they are not the same jobs. They do not have the same career progression. They do not offer the same entry points. They do not build the same skills over time in the same way.

When someone says “we still need humans, just in a different role,” they are telling you the old role is gone while pretending this is neutral information.

It is not neutral.

The Paradox That Nobody Will Name Out Loud

Here is the thing that bothers me most.

Dario Amodei runs a company whose stated mission is the responsible development of AI for the benefit of humanity.

He also just told the world that his AI already writes 90% of the code at his own company, and that this represents the future of software engineering work.

Those two things are in tension.

If AI is writing 90% of the code, and that is the direction of travel, and the “new skills” required are possessed primarily by people who already have seniority, context, and economic stability — then the technology that is supposed to benefit humanity is, in the near term, primarily benefiting:

People who own AI companies. Companies that use AI to reduce labor costs. Senior engineers who can direct AI workflows.

And it is hurting:

Junior engineers who cannot get their first job. Mid-career workers whose skill sets are being commoditised. Contractors and freelancers in markets where AI undercuts their rates. Students who are being told to learn skills that were valued yesterday.

This is not a surprising outcome. Technology concentration of benefits at the top is one of the most well-documented patterns in economic history.

But the people describing this concentration are the people at the top. And they are describing it as if it is inevitable, natural, and ultimately fine.

It is not inevitable. It is a choice about how to deploy the technology.

It is not natural. It is a consequence of specific funding structures, IP regimes, and corporate incentives.

And whether it is fine depends entirely on who you are.

What I Want to Say to Every Engineer Reading This

You are not broken.

You are not slow to adapt. You are not failing to “embrace AI.” You are not insufficiently visionary.

You are a skilled person watching the economic value of your skills be repriced in real time by forces you did not choose and cannot individually control. And you are being told, simultaneously, that this is good news, that you should be grateful for the productivity gains, and that if you are struggling it is because you have not yet found your new “AI-augmented” identity.

This is an enormous psychological burden to place on individuals.

And it is being placed there so that the structural question — who benefits from this technology, and at whose expense, and what do we owe people whose livelihoods are disrupted — can remain unasked.

I am asking it.

Not because I think AI should be stopped. It should not and cannot be.

Not because I think software engineers deserve special protection that other workers do not. They do not.

But because the framing of “upskill or become irrelevant” locates the problem in the individual rather than in the system. And that framing conveniently excuses the system from any responsibility to the people it displaces.

The Things the CEOs Are Not Saying

They are not saying: “We are reducing our engineering headcount and passing the savings to shareholders.”

They are saying: “We are becoming more efficient.”

These are the same statement. One is honest. One is palatable.

They are not saying: “The pipeline for developing senior engineers has been disrupted, and we have no plan for where the next generation of senior engineers will come from if the entry-level pipeline closes.”

They are saying: “The bottleneck has moved to higher-level skills.”

As if those higher-level skills grow on trees. As if they are not downstream of years of lower-level experience. As if the people who currently have them got there without doing the work that is now automated.

They are not saying: “We don’t know how to train the next generation of engineers without the learning environment that entry-level coding work provided.”

They are saying: “The nature of work is changing.”

It is. But “the nature of work is changing” is not a workforce development strategy. It is an observation that displaces all responsibility for the transition onto the people being transitioned.

They are not saying: “We extracted enormous value from the open-source community, from the Stack Overflow answers of millions of engineers, from the GitHub repositories of developers who never consented to having their code used as training data — and we owe something back to those people.”

They are saying: “The future is exciting.”

It may be. For some people more than others.

What You Can Actually Do. Not Platitudes.

I am not going to tell you to “learn prompt engineering.” I am not going to tell you to “build your personal brand.” I am not going to tell you that if you just embrace the right mindset everything will be fine.

I am going to tell you three things that I actually believe.

First: Your accumulated context is real and it is yours.

The specific knowledge you have about your industry, your users, your codebase, your organisation’s history of decisions and their consequences — this cannot be replaced by a model trained on generic internet data.

The more specialised your context, the more valuable you are. Not because you can write Python faster than Claude (you cannot). Because you know why the Python looks the way it does. You know what was tried before. You know what the actual failure modes are in production.

Depth of domain knowledge is not a consolation prize. It is the actual differentiator. Invest in it relentlessly.

Second: The evaluation skill is now a core competency, not a nice-to-have.

Amodei is right that the bottleneck has moved. It has moved to verification. To catching the confident wrong answer. To knowing enough about the domain to audit the 90% that the model produces.

This is a skill. It is a learnable skill. It is a skill that requires deep domain knowledge to do well. A model cannot verify its own outputs reliably. You can.

“Knowing when the AI is wrong” is the most valuable skill in the 2026 labor market. Not because it is glamorous. Because it is rare. Most people who use AI tools cannot tell when the output is subtly broken. You can train yourself to be someone who can.

Third: Collective action matters more than individual pivoting.

Every “upskill” narrative individualises a structural problem. You cannot upskill your way out of a market where the economic rules have been rewritten.

What workers have always done when the economic rules change: organise. Build collective leverage. Make the cost of replacing you — as a group, not as an individual — higher than the cost of keeping you.

I know this sounds old-fashioned. I know tech culture has always been allergic to it. I also know that the last time a technology revolution repriced skilled labor this fast was the industrial revolution. And the people who eventually achieved decent working conditions from that transition did not achieve them by individually upskilling. They achieved them by acting collectively.

This is not a call for any specific political position. It is a historical observation about what works.

The Question I Want Dario to Answer

Not in a conference keynote. Not in a polished interview. In a direct conversation.

You say engineers are “10x more productive” because of Claude.

Are they being paid 10x more?

If not, where is the productivity gain going?

And if you know where it is going, and you know it is not going to the engineers, and you are still describing this as an unambiguously positive development for “humanity” — then I would like to understand your definition of humanity.

Because mine includes the people writing the code.

And theirs.

And their rent.

One More Thing

I want to say something that is genuinely true and that I mean.

The technology is extraordinary.

I use Claude. I have used it to build things I could not have built alone. I have been genuinely impressed and occasionally moved by what it produces.

The problem is not the technology.

The problem is the story being told about the technology. The story in which its benefits are framed as universal and its costs are framed as individual failures of adaptation.

The technology creates genuine abundance. It also creates genuine disruption. Both are true. And the distribution of that abundance versus that disruption is not determined by physics or by mathematics. It is determined by choices. About who gets to set the terms. About who captures the value. About what we owe each other during a transition of this scale.

The billionaires are making their choices.

We need to make ours.

Tell Me

If you are an engineer reading this in 2026:

What has actually changed in your day-to-day work in the last two years? Not what the think pieces say has changed. What you actually experience.

Because the people writing the think pieces are not in your chair.

And the people making the decisions are not reading your Slack.

The most important data about what this transition actually feels like from inside it is sitting in your head right now.

I want to know what it is.

I build AI systems for a living. I believe in the technology. I am also allowed to notice when the people who profit most from it are doing the majority of the talking about its human consequences. Both things are true.

AI #SoftwareEngineering #FutureOfWork #DarioAmodei #Anthropic #TechIndustry #Engineers #CareerAdvice #Inequality #Automation #MachineLearning #Claude #OpenAI #WorkplaceReality #Technology