Last Thursday, Jack Dorsey announced Block is reducing its organisation by nearly half — from over 10,000 people to just under 6,000. The stock surged 24%. His words on X were blunt: "Intelligence tools have changed what it means to build and run a company. I don't think we're early to this realization. I think most companies are late."
Here is the connection most people are missing. Block is the company that built Goose — the open-source AI coding agent that started as an internal experiment by one of their engineers, Brad Axen. They open-sourced it in January 2025 under Apache 2.0. Block's engineers were already reporting 8–10 hours saved per week. Then Stripe forked Goose, built their own version called Minions, and started shipping over 1,000 merged pull requests every week — with zero human-written code.
Dorsey did not just announce layoffs. He announced what happens when a company actually follows through on AI-driven engineering at scale.
The AI Is Not the Interesting Part
This is the bit that should make every engineer and IT leader pay attention. Stripe's agent is a fork of an open-source tool. The AI model is essentially a commodity. What Stripe actually built was the infrastructure around it.
There is an excellent breakdown of Stripe's architecture that makes a distinction worth highlighting: this is not vibe coding — it is agentic engineering. The difference matters. Vibe coding is prompting an AI and hoping for the best. Agentic engineering is building a deterministic system with an AI engine inside it, where you know what will happen because you have designed the guardrails.
Stripe's system has several layers that make it work:
A blueprint engine that interleaves AI creativity with deterministic code. Not everything needs an LLM — git operations, linting, test execution are all handled by regular code. The agent handles the creative problem-solving.
Isolated dev boxes — full cloud instances per agent run, pre-warmed in seconds with the entire codebase and services. Engineers spin up multiple agents in parallel, each in its own sandbox.
A tool shed — a meta-tool that dynamically selects from 400+ internal tools, preventing the token explosion you would get from giving an agent access to everything at once.
Conditional rules files that scope context to specific subdirectories. Stripe's codebase is hundreds of millions of lines — no agent can read all of it. The rules ensure agents only get relevant context.
A validation layer with over 3 million tests, selectively run on each push, giving agents automated feedback. If the agent cannot fix a CI failure in two rounds, it stops and flags a human.
The result? An engineer writes a Slack message, walks away, and comes back to a production-ready PR that has passed CI and is ready for review. No interaction in between.
Why This Should Be a Wake-Up Call
Some analysts have called Block's layoffs "AI-washing" — cover for pandemic-era overhiring. And there is probably truth in that. Dorsey admitted he overhired during COVID and built redundant structures. But whether it is 60% AI and 40% restructuring or the other way around does not change the signal.
A CEO of a major public company explicitly said AI made this possible, predicted most companies will do the same within a year, and the market rewarded him with a 24% stock price jump.
Meanwhile, Stripe is demonstrating what the engineering side of this actually looks like in practice — not theory, not a demo, but 1,000+ PRs a week in a codebase that processes over a trillion dollars annually.
The Real Lesson
The companies winning here are not the ones with the best AI models. They are the ones building the best infrastructure around commodity AI — the orchestration, the guardrails, the feedback loops, the developer tooling.
And that infrastructure layer is where the real decisions live: how you isolate agent environments, how you manage context at scale, how you integrate with existing CI/CD, and critically, how you keep the costs sustainable as you scale. When you are spinning up dedicated VMs for every agent run, the cost management conversation becomes essential, not optional.
We are entering a phase where the competitive advantage shifts from who can write the most code to who can build the best system for agents to write code on their behalf. Stripe just published the reference architecture. Block just showed us the workforce implications.
The question for the rest of us is whether we are paying attention.
Further reading:
John Holmes — Azure Solutions Architect, Caleta.io