Apuntes de ingeniería desde el terreno.
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.
Notion's Developer Platform Just Made the Workspace the Agent Control Plane
On May 13 Notion shipped a Developer Platform built on three pieces — Workers (a hosted runtime for custom code), Database sync (live mirror of Salesforce, Zendesk, Postgres into Notion databases), and an External Agents API that brings Claude Code, Cursor, Codex, Decagon, and custom-built agents into the workspace as typed, first-class participants. Microsoft Agent 365 made governance a category. Notion just made the workspace itself the place agents live. They're not in competition; they're in adjacency — and the buying conversation just got more complicated.
Claude Platform on AWS Hits GA: Anthropic-Operated, IAM-Authed, Inside Your AWS Account
On May 12 Anthropic moved Claude Platform on AWS to general availability — the full direct-API feature surface (Managed Agents, Skills, MCP connectors, batches, caching, citations), but authenticated through AWS IAM, audited through CloudTrail, and billed through the AWS invoice. Anthropic still operates the service and processes the data, which is what makes this different from Bedrock. Procurement friction inside AWS-shop enterprises just collapsed; the workload-allocation conversation is the one that didn't exist yesterday.
Microsoft Agent 365 Went GA on May 1 — Agent Governance Is Now a $15/User/Month Line Item
Microsoft Agent 365 became generally available on May 1, bundled into the new E7 Frontier Suite at $99/user/month or standalone at $15/user/month. It discovers, observes, governs, and secures agents across Microsoft, AWS Bedrock, and Google Cloud — registry sync, lifecycle controls, runtime policy, the whole control-plane category in one SKU. Three months ago the question was "do we need an agent governance plane." Today the question is "do we let Microsoft own it for us, or do we build the cross-vendor version ourselves."
Coder Agents Hits Beta: Self-Hosted, Model-Agnostic Coding Agents and the Regulated-Industry Door Just Opened
Coder went to beta on May 6 with Coder Agents — a native agent architecture that runs entirely inside the customer's network perimeter (cloud VPC, on-prem, or fully air-gapped), with control plane, orchestration, and execution all on customer infrastructure, and bring-your-own model support including self-hosted weights. Source code, prompts, and model traffic never leave the network. For every regulated enterprise that's been telling its engineers "no Cursor on production code," the answer changed this week.
Snyk Embedded Claude and Shipped Evo: AI-Generated Code Just Got Its Own Security Stack
On May 7 Snyk announced it had integrated Anthropic's Claude across its AI Security Platform and brought Evo — the first agentic security orchestration system with red-team agents that continuously probe running AI applications for prompt injection and data exfiltration — to general availability. The framing in the press release is "AI-native AppSec." The framing that matters for buyers: 65–70% of production code is now AI-generated, nearly half of it contains vulnerabilities, and the agents shipping that code operate almost entirely outside traditional AppSec tooling.
GPT-5.5 Instant Replaced the Default on May 5 — Every Team Without an Eval Suite Just Got Upgraded by Surprise
On May 5 OpenAI flipped GPT-5.5 Instant to the new default for ChatGPT and to chat-latest in the API, replacing GPT-5.3 Instant. The model claims 52.5% fewer hallucinated claims on high-stakes prompts, 30% shorter responses, fewer emojis, and stronger personalization. Every team consuming chat-latest just absorbed a behavior change without explicitly opting in. That's what default-model swaps do — and it's why 2026 is the year an eval suite stopped being optional.