What Happens When a Public Company Goes All In on AI

What It Covers

Block (parent of Square, Cash App, Afterpay) executed a 40%-plus workforce reduction in early 2026, restructuring around AI agents and squads of one to six people. Owen Jennings details how late-2025 model improvements broke the decades-old headcount-equals-output equation and what operating inside that transformation looks like.

Key Questions Answered

• •Workforce restructuring trigger: A binary capability shift in late November–December 2025, when models like Opus 4 and Codex 5.3 became proficient with existing complex codebases—not just greenfield projects—made one or two AI-enabled engineers 10x to 100x more productive, directly causing Block's 40%-plus reduction in force concentrated on the development side.
• •Squad model replacement: Block replaced classic feature teams of roughly 14 engineers with squads of one to six people working alongside AI agents. Designers and PMs now ship pull requests directly. Internal tool BuilderBot autonomously merges PRs and completes features to 85–100%, with humans handling only the final 10–15% requiring deep contextual judgment.
• •Parallel agent workflow: Individual contributors now run 8–14 simultaneous agent instances rather than working linearly through a single pull request. The workflow shifts from sequential task completion to context-switching across multiple agents building in parallel, then reviewing, nudging, and committing outputs—a fundamental change applicable to engineers, PMs, and growth marketers alike.
• •Compliance and regulatory carve-outs: Block deliberately left its compliance team and compliance technology team nearly untouched during the restructuring. When operating in complex regulatory environments, companies should treat compliance headcount as a non-negotiable floor regardless of AI capability, maintaining human oversight to avoid regulatory and reputational risk during the transition period.
• •Defensibility through proprietary signal: Long-term competitive moats will belong to companies that deeply understand something structurally hard for others to replicate—proprietary data and domain insight—then run a continuous feedback loop using that signal plus agentic build tools like BuilderBot. Companies unable to articulate their unique signal risk being displaced entirely by AI-enabled competitors.