The market cheers. Engineers leave. And the industry may be quietly eating itself.
The question had been circulating in developer conversations for weeks before ABC journalist Julian Fell was one journalist that put it into print: did AI really take Block's jobs? It was the right question to be asking publicly. But I think the more consequential one sits just beneath it. Not whether AI is taking jobs today, but what happens to the industry's capacity to invent tomorrow if it keeps behaving as though it has.
I have spent enough time as a developer and watching this sector evolve to know that the current moment feels different. Not necessarily in the ways that are making headlines, but in subtler ways that do not show up in share prices or headcount figures.
The financial restructuring dressed as transformation
Let's start with what the data actually shows, because it deserves more scrutiny than it has been getting.
When Block announced in late February 2026 that it was cutting 40% of its workforce, roughly 4,000 people, CEO Jack Dorsey framed it plainly as an AI story. Intelligence tools, he said, had changed what it means to build and run a company. Investors responded immediately. Block's stock, which had closed at $51.94 (NYSE) on the day of the announcement, surged to around $67 in after-hours trading, a jump of nearly 25%.
But watch what happened next. Within three weeks, the stock had settled back to around $59. The AI narrative moved markets for a day. Fundamentals did the rest.
This is not an isolated pattern. Australian productivity software company Atlassian cut 1,600 people, around 10% of its workforce, in March, also framing the decision around AI and competitive repositioning. Its NYSE-listed stock was at $79.43 on the day of Block's announcement. Today it sits at $58.96. WiseTech Global, the Australian logistics software firm, announced cuts of around 2,000 roles, nearly a third of its workforce, citing generative AI efficiencies. Its ASX-listed shares have fallen from $49.00 to $37.90 in the same period.
What these numbers show, taken together, is that the AI framing is providing a brief investor sugar hit in some cases but it is not, on its own, a durable market signal. And this matters because the framing has real consequences beyond the share price.
Block had grown its headcount from around 3,800 employees in 2019 to more than 10,000 before these cuts, a ballooning driven largely by pandemic-era demand. The February 2026 announcement was also, notably, the company's third significant round of layoffs in roughly two years, following cuts in January 2024 and March 2025. This is a business undertaking financial restructuring. Calling it an AI transformation is not necessarily dishonest, the tools are genuinely changing how engineering work gets done, but it is a framing that flatters the story.
The term "AI washing" was coined precisely for this dynamic: executives attributing to AI what is partly, or substantially, a correction of pandemic-era over-hiring and balance sheet repair.
The developer's sugar hit
But here is where I want to move away from the financial picture, because I think it is actually the less interesting story.
What is unfolding for developers right now is its own version of the sugar hit. AI delivers immediate, tangible gains. Tasks get done faster. Problems get resolved more quickly. A developer who once spent an afternoon digging through documentation can surface an answer in minutes. Open source communities are feeling this acutely. AI can now sweep through well-documented repositories and extract insights that once required genuine expertise and significant time investment. The productivity uplift, for experienced developers especially, is real. Everyone is getting that hit right now.
And it does feel like progress, because in many measurable respects, it is.
But there is a critical distinction being glossed over in much of the public discourse around AI and software development: AI is not creating. It is retrieving, recombining, and accelerating. That is genuinely impressive. It is also genuinely different from invention.
This distinction matters enormously for what comes next.
The friction we are optimising away
Software development, at its most productive, is not a clean or efficient process. It involves bottlenecks. It involves mistakes, dead ends, misguided architecture decisions, and the slow, sometimes painful process of figuring out why something does not work the way you expected it to. These are not inefficiencies to be eliminated. They are, in many cases, where new understanding comes from.
The developer who spends three days debugging an obscure edge case learns something that does not exist anywhere in a training dataset. The team that builds a framework that turns out to be the wrong approach, and has to reckon with why, develops intuitions about system design that cannot be retrieved from a model trained on previous work. The open source contributor who argues about architecture in a public thread for two weeks is doing something that looks like friction but is actually the process by which ideas get tested and refined.
If developers lean too heavily on AI tools, or more importantly, if economic pressure drives enough of them out of active software development altogether, we risk losing the very friction that drives genuine innovation. An industry that optimises this away may find it has also optimised away its capacity to invent.
This is not a theoretical concern. We are already seeing early signals in open source communities, where the nature of contribution is shifting. AI can answer questions that once required a community of humans to maintain shared knowledge. That sounds purely beneficial until you notice that maintaining shared knowledge is also how communities develop new knowledge.
The longer question: what does AI actually learn from?
There is a deeper structural question here that deserves serious attention, even if it remains somewhat speculative.
AI coding tools, the ones driving the productivity gains that Dorsey, WiseTech's leadership, and Atlassian's Mike Cannon-Brookes have all cited, are trained on human-written code. The quality and diversity of that training data is a function of the breadth and depth of human experimentation in software development. Programming languages, frameworks, and architectural patterns advance because developers experiment, iterate, publish, argue, refactor, and publish again.
If fewer developers are doing that work, either because AI has automated parts of the role, or because the economics of the sector have driven people into adjacent careers, or simply because the experience of early-career development becomes less viable as AI handles more routine tasks, then the pool of new ideas that the next generation of AI tools can learn from begins to thin.
The tools do not just stagnate in this scenario. They begin to recombine an increasingly dated set of ideas. The acceleration continues, but it is accelerating along a narrower and narrower corridor.
This is the part of the AI-and-jobs conversation that I think is genuinely underexplored. Most of the debate is about displacement: will AI take jobs? The more unsettling question is about feedback loops. Does an industry that displaces the humans who generate new ideas eventually undermine the AI systems that depend on those ideas?
What the honeymoon phase obscures
We are, right now, in something of an AI honeymoon phase. The tools are impressive. The productivity gains are real. The opportunities, for developers who can use these tools well and for companies that integrate them thoughtfully, are genuine.
But the honeymoon phase is precisely when the structural questions get papered over. Everyone is focused on what is being gained. Fewer people are asking what is being quietly lost.
What is being lost, I would argue, is not primarily jobs. The labour market question will resolve itself over time in ways that are genuinely hard to predict. What is being lost is something harder to measure: the density of human experimentation in software development, the breadth of the ecosystem in which new ideas emerge, and the apprenticeship pathways through which developers accumulate the kind of hard-won judgement that AI tools can accelerate but cannot replace.
Fell's reporting on Block asked whether AI really took those jobs. The honest answer is probably some of both: genuine AI-driven efficiency gains and financial restructuring dressed in Silicon Valley vocabulary. The market's short-lived enthusiasm for most of these announcements suggests investors are beginning to work out the difference.
But the question that will matter more, five or ten years from now, is whether an industry that is aggressively optimising for short-term efficiency is also quietly eroding the conditions that make long-term innovation possible.
The sugar hit is real. The question is what comes after it.
