#Is AI Coming for Software Jobs? Why Tech Firms Are Panicking — and What It Means for You

On February 26, 2026, Jack Dorsey cut Block's workforce from over 10,000 to just under 6,000 — a 40% reduction — while the company was profitable and growing. He tied it directly to AI efficiency gains from Block's internal agent platform, Goose, and predicted most companies would do the same within a year. Two weeks later, that prediction is landing harder than anyone expected. Microsoft's Satya Nadella says AI now writes 20–30% of the company's code. Google's Sundar Pichai says the same. The U.S. recorded 108,435 announced layoffs in January 2026 alone — up 118% from a year prior and the highest January total since 2009. The "are AI tools really replacing jobs" debate that has been running since 2023 now has a clear, documented, public case study with a name on it. This post explains what actually happened at Block, what the three biggest tech companies are doing differently, why some experts think Dorsey is right and others think he's wrong, and what it means if you work in software.

#The Moment the Debate Stopped Being Theoretical

For the past three years, the tech industry has been arguing about whether AI would actually replace jobs or just help people do their existing jobs faster. The argument had two camps, and both had good points.

Camp one said: AI tools are powerful but shallow. They speed up certain tasks but they can't replace the judgment, architecture decisions, product instincts, and organizational knowledge that experienced engineers carry. The productivity gains are real, but they're gains for humans who use the tools — not replacements for humans.

Camp two said: The tools are compounding. Every six months they get meaningfully better. The tasks AI can automate today are a narrow slice of what it will automate in eighteen months. If you're building headcount assumptions around today's capabilities, you're already behind.

On February 26, 2026, Jack Dorsey picked a side — loudly, publicly, and with 4,000 people's jobs as the evidence.

Block, the parent company of Square, Cash App, and Afterpay, cut its workforce by more than 40% in a single announcement. The company had just reported Q4 gross profit of $2.87 billion, up 26% year-over-year. Cash App gross profit was up 33%. The business was not struggling. Dorsey was explicit: this was not distress. This was a proactive bet that AI tools had crossed a threshold that made a team of 6,000 as productive as a team of 10,000.

"I think most companies are late," Dorsey wrote in his shareholder letter. "Within the next year, I believe the majority of companies will reach the same conclusion and make similar structural changes."

That sentence has been sitting in a lot of C-suites for the past two weeks.

#What Block Actually Did: Goose and the Numbers Behind the Bet

#The Internal Tool That Changed the Math

The specific catalyst Dorsey pointed to wasn't a general observation about AI progress. It was a December 2025 leap in model capabilities that he described on his analyst call as something where "the models just got an order of magnitude more capable and more intelligent."

That shift showed up directly in Block's own internal platform. Goose — an agentic coding harness Block built internally that sits on top of large language models to execute actions, draft emails, and automate workflows — has been in production use at Block for about 18 months and has since been open-sourced. Block's CFO Amrita Ahuja disclosed the clearest public data point in the AI productivity conversation yet: since September 2025, developer productivity at Block improved with a 40% increase per engineer in production code shipped.

That's not a benchmark. That's a real internal number from a real company that just bet its entire org chart on it being accurate.

Ahuja gave another detail that illustrates the scale of the shift. One risk underwriting model that previously took a full quarter to build was completed in a fraction of the time with Goose. A task that once occupied an entire team for three months became something a smaller group finished in weeks.

#The Efficiency Target That Set a New Benchmark

The ambition behind the cuts is visible in one number: Block is now targeting gross profit of over $2 million per employee. Before Covid, that figure sat around $500,000. By 2024 it had reached $750,000. By 2025 it hit $1 million. Dorsey is betting it can double again in a single year.

If Block achieves $12.2 billion in gross profit — their 2026 guidance — with a workforce of 6,000, that benchmark becomes industry-wide evidence that lean, AI-enabled teams can outperform larger, traditionally staffed ones. For the boards and leadership teams at every public software company watching Block's stock jump 24% on the announcement, that's a number that's very hard to unsee.

#Three Companies, Three Strategies: Block, Microsoft, and Meta

Block is the bluntest case. But it's not the only one. The three largest software companies in the world have all acknowledged that AI is writing code at scale — and each one is responding differently.

#Block: Slash and Rebuild

Dorsey's strategy is the most aggressive and the most honest. Cut now, deeply, from a position of strength rather than be forced into it reactively later. Retain the people who can work effectively alongside AI tools. Target $2M gross profit per head. The market rewarded the move immediately. The employees who lost their jobs are a different story.

The argument against Dorsey is that Block had pandemic-era over-hiring baked into its headcount already. Block employed 3,835 people at the end of 2019 and nearly tripled its workforce to over 10,000 by 2023. Critics on X pointed out, reasonably, that unwinding a bloated workforce isn't the same as proving AI caused it. Dorsey disputed this framing, and the Goose productivity data gives him some ground to stand on.

#Microsoft: Quiet Cuts, Loud Claims

Microsoft's Satya Nadella disclosed last year that AI now writes 20–30% of the code inside the company's repositories. Google's Sundar Pichai has reported similar figures. Microsoft CTO Kevin Scott has predicted 95% of all code will be AI-generated by 2030.

In May 2025, Microsoft laid off approximately 6,000 employees. Among the 2,000 cuts in its home state of Washington, state filings showed that over 40% of those affected were in software engineering roles — by far the largest single category. Microsoft declined to comment when asked whether AI-assisted coding contributed to those engineering-specific cuts. The company has not made the same explicit connection Dorsey made. But the pattern is the same.

#Meta: Spend Big, Concentrate Talent

Mark Zuckerberg is taking the opposite approach on headcount — for now. Meta guided for $115 to $135 billion in 2026 capital expenditures and is hiring elite AI researchers aggressively. Zuckerberg said that "projects that used to require big teams can now be accomplished by a single very talented person" while simultaneously planning to grow sales headcount and acquire AI talent at any cost. He noted during LlamaCon that he didn't even know what percentage of Meta's code was AI-generated.

The Meta strategy is to spend its way into the AI era rather than contract. Whether that means it avoids AI-driven headcount reductions or is simply delaying them is something the next two years will answer.

#The Skeptics Are Not Wrong Either

It's worth being honest about what we don't know.

The Citrini Research "Global Intelligence Crisis" post — a viral thought experiment imagining 2028 with unemployment topping 10% and the S&P 500 in freefall due to AI-driven layoff cascades — got enormous attention the same week as Block's announcement. Deutsche Bank analyst Jim Reid called its vibes-to-substance ratio "undeniably high." Citadel Securities pushed back on it directly.

There is a legitimate counterargument. An Oxford Economics report found that many layoffs CEOs described as AI-related were actually the result of pandemic over-hiring corrections. Block's own history — tripling headcount between 2019 and 2023 and then unwinding it — gives that argument some force.

There is also the historical pattern. ATMs did not eliminate bank tellers — they allowed banks to open more branches and employment in banking grew. The internet reduced the number of employees needed to generate a million dollars in revenue, but it also created entire industries that didn't exist before. The current AI panic has echoes of every previous automation panic, and those panics were, broadly, not borne out as catastrophically as feared.

And the productivity data is messier than headlines suggest. One enterprise study found a 26% increase in completed tasks with AI coding tools. Another found no change in cycle time and a 41% increase in bugs. The signal is real. The magnitude is contested.

#What This Means If You Work in Software

The honest answer is that this depends enormously on what kind of software work you do and at what level.

The roles most immediately at risk are the ones AI coding tools already do well: writing boilerplate, generating tests, building standard CRUD applications, producing documentation, completing well-specified tasks in well-trodden languages. Entry-level positions that primarily involve implementing clear specifications are under real pressure. The evidence from Block's Goose data, Microsoft's headcount patterns, and Q1 2026 earnings calls is consistent enough to take seriously.

The roles with more durability are the ones that require judgment that can't be specified in a prompt: architecture decisions, understanding a system's undocumented history, translating a vague business problem into something buildable, managing the human dynamics of a cross-functional team, spotting what's missing from a specification rather than just executing it. These skills compound over a career in ways that are harder to automate.

The roles growing fastest are in AI infrastructure, evaluation, and oversight: ML engineers, AI safety researchers, prompt engineers, AI product managers, and the emerging field of AI auditing. Demand for these roles is growing at rates between 35% and 140% depending on the specialty.

The practical frame: The question is not "will AI take my job" as if it's a binary event. It's "what percentage of my current tasks are the kind that AI tools are getting better at, and what am I doing about the rest?" That's a question worth answering now, not in eighteen months when the restructuring announcements have already gone out.

#What Should You Do Right Now?

If you're an engineer: Don't avoid the tools. Use them aggressively. The engineers who are being retained are the ones who can leverage AI to produce output previously requiring teams — not the ones who are demonstrably replaceable by it. Knowing how Goose, Claude Code, or Cursor works inside your stack is no longer optional.

If you're in tech leadership: The Block announcement has changed what boards are asking. Even if you aren't planning cuts, having a clear answer to "what is your AI productivity strategy and what is your gross profit per head" is now part of the job. Ignoring the question is not a neutral act.

If you're thinking about entering software development: The market for entry-level software engineers is the most directly affected segment right now. The path forward involves either specializing in areas AI is worst at (novel systems, complex architecture, human-facing product decisions) or specializing in AI itself. A general-purpose junior developer role at a large company is a harder sell in 2026 than it was in 2022.

#FAQ

Did AI really cause Block's layoffs, or was it just pandemic over-hiring?
Both things are true simultaneously, and Dorsey acknowledged it. Block over-hired during Covid — workforce tripled from 2019 to 2023. The correction was coming regardless. But the Goose productivity data — 40% more production code per engineer since September — is a real internal figure tied to a real tool. Dorsey's claim is that AI efficiency made it possible to cut more deeply and more confidently than a plain over-hiring correction would have justified. Whether you believe the AI story or the bloat story depends partly on how you weight those two factors.

What is Block's Goose platform?
Goose is an agentic AI platform Block built internally, sitting on top of large language models to execute tasks, draft communications, and automate workflows. It has been in production at Block for approximately 18 months and has since been open-sourced. Block's CFO credited it with a 40% increase in production code shipped per engineer since September 2025.

Is Microsoft's AI code figure (20–30%) accurate?
Nadella described it as "maybe 20–30%" with a lot of hedging language. How Microsoft measures AI-generated versus human-written code is not disclosed, and some engineers have pointed out there's no reliable way to track it at scale. The figure is best treated as a ballpark directional signal, not a precise measurement.

Which software jobs are safest from AI displacement?
Roles requiring complex judgment, system architecture, novel problem-solving, and human relationship management have more durability. Roles primarily involving well-specified implementation tasks in standard languages are under more pressure. The fastest-growing roles right now are in AI infrastructure, evaluation, and oversight.

Will the job market recover as AI creates new roles?
Historically, major technological waves have destroyed categories of jobs and created new ones. AI engineering, auditing, and oversight roles are growing rapidly. The honest answer is that the transition period — which is now — is turbulent, and the new jobs being created require different skills from the ones being automated. Retraining is a real requirement, not a platitude.

How does Dorsey's prediction hold up so far?
Two weeks in, the structural pressure he described is visible across the industry. Whether the majority of companies make similar cuts within a year — as he predicted — will depend on how much AI model capability continues to compound and how much confidence leadership teams develop in productivity gains translating to safe headcount reduction. The Block data is the first public benchmark. More will follow.