Mercado Libre Is Betting 23,000 Engineers on 90% Autonomous Coding by Q3
At Code w/ Claude 2026, Anthropic put a number on the next phase: Mercado Libre is targeting 90% autonomous coding across 23,000 engineers by Q3. The new Routines feature is the primitive that makes it sane.
NeuroX AI · May 20, 2026
At Code w/ Claude 2026 last week, Anthropic put a number on where this is heading: Mercado Libre is targeting 90% autonomous coding across its 23,000 engineers by Q3 2026. Not "AI-assisted." Autonomous.
That target wouldn't be sane without the other thing Anthropic shipped on the same stage — Routines. Higher-order prompts that run on cron schedules, GitHub webhooks, or API endpoints. The pitch was literal: "wake up to merge-ready PRs." A queue of tasks the agent picked up overnight, drafted, ran the tests against, opened a PR, waited for review.
The supporting numbers are heavy. Anthropic's API volume is up 17x year-on-year. Claude Code's five-hour rate limits doubled for Pro, Max, and Enterprise. Managed Agents added multi-agent orchestration, Outcomes (success criteria the agent iterates against autonomously), and Dreaming (agents inspect prior sessions to build memory).
The shift this implies is one we've been pointing at for six months: the unit of work is no longer the prompt or the session. It's the routine. A pre-scoped task, a trigger, an eval, a PR target. The engineer designs the routine. The agent runs it on every webhook.
Most teams are not ready for that yet. No eval suite to gate the PR, no monitoring to catch a bad run, no staging path the agent can poke at safely. That gap is the thirty days.