Ensayos y notas de campo sobre IA, ingeniería de software, diseño y el oficio de construir equipos de producto que entregan. Escrito por los ingenieros que hacen el trabajo. Publicaciones en inglés.

At VeeamON 2026 on May 12 Veeam launched the DataAI Command Platform — six capabilities (Command Graph, Security, Governance, Compliance, Privacy, Precision Resilience) built on the Securiti AI acquisition, with 300+ connectors and a stated design principle that control belongs at the data source, not at the agent. Three weeks after Microsoft Agent 365 made governance-of-agents a category, Veeam shipped governance-of-the-data-the-agents-touch as a category. They're not in competition — they're the two halves of an enterprise agent posture every regulated buyer is suddenly putting together this quarter.

On May 4 Anthropic announced an enterprise AI services firm backed by Blackstone, Hellman & Friedman, and Goldman Sachs — with Apollo, General Atlantic, Leonard Green, GIC, and Sequoia rounding out the cap table at ~$1.5B total. The thesis isn't Fortune 500 — it's the mid-market: community banks, regional manufacturers, regional health systems. One week before, OpenAI launched DeployCo at $10B for the upper-end of the same buyer conversation. The deployment layer just split into two tiers, both capital-rich, both routed through PE-portfolio distribution channels, and the boutique that's been quietly selling "embed senior engineers into mid-sized companies" now has Anthropic as a same-shape competitor.

At MongoDB.local London on May 7, MongoDB shipped a coordinated push to make Atlas the production data layer for AI agents: Voyage AI Automated Embeddings (generated automatically as data is written), GA LangGraph.js Long-Term Memory Store (persistent cross-conversation agent memory backed by Atlas), MongoDB 8.3 (45% more reads, 35% more writes), and cross-region private connectivity. The headline is "production-ready enterprise AI." The substance: the five data surfaces every production agent needs — operational state, vector search, memory, embeddings pipeline, reranker — just got a single-vendor answer, and teams running three or four stores have a real consolidation conversation to schedule.

On May 12 The Information reported Anthropic in advanced talks to acquire Stainless — the AI-powered SDK compiler that generates the official client libraries for OpenAI, Google, Cloudflare, and Meta — for at least $300M, partly in Anthropic equity. The headline framing is "developer-tools acquisition." The substance is one tier deeper: the SDK is the thinnest layer the developer sees of an API, it's where multi-language ergonomics get built or broken, and whoever owns the generator owns a quietly leveraged control point across every vendor that uses them. Every multi-model routing team should be paying attention this quarter.

On May 11 OpenAI launched the OpenAI Deployment Company — a standalone $10B Delaware LLC backed by $4B from TPG, Advent, Bain Capital, Brookfield, plus McKinsey, Bain & Company, and Capgemini as distribution partners, with ~150 forward-deployed engineers acquired through Tomoro. The pitch is engineers embedded inside enterprises to redesign workflows around frontier AI. That's the same buyer conversation every senior AI consultancy is having today — just with OpenAI's name on the contract and three of the world's largest SI/PE channels carrying the bag. The deployment layer is now its own market.

On May 14 OpenAI shipped Codex inside the ChatGPT mobile app for iOS and Android, in preview, available across every plan including free. From the phone, a developer can switch between active Codex tasks, approve commands, redirect a running agent, switch models, or kick off new ones — all connected to a desktop Codex instance back at the workstation. The same release graduated Hooks to GA, rolled out programmatic access tokens for CI use, and added HIPAA-compliant Codex for ChatGPT Enterprise. The mobile surface is the headline. The CI plumbing is the production change.