mauriciosuarez.dev — WritingIdeas, guides, and lessons from building enterprise software and integrating AI in production. AI Engineering, Developer Experience, and System Architecture.https://mauriciosuarez.dev/enAI worms at home: slow does not mean harmlesshttps://mauriciosuarez.dev/en/writing/ai-worms-at-home/https://mauriciosuarez.dev/en/writing/ai-worms-at-home/15 TV boxes, a 26B model, and an uncomfortable question: how close are household AI worms? What Anthropic demonstrated, and what it projected.Wed, 14 Oct 2026 00:00:00 GMTHarness learning: is the model to blame?https://mauriciosuarez.dev/en/writing/harness-learning-mejorar-agentes-ia/https://mauriciosuarez.dev/en/writing/harness-learning-mejorar-agentes-ia/Harness learning aims to improve AI agents by revising their workflow. What the paper shows, and what to check before switching models.Mon, 12 Oct 2026 00:00:00 GMTThe bot isn't enough: why we built an MCP serverhttps://mauriciosuarez.dev/en/writing/mcp-operadores-admin/https://mauriciosuarez.dev/en/writing/mcp-operadores-admin/Why we built an MCP server for MaltaClean's operators and admins, and what permissions, OAuth and actions with real consequences taught us.Wed, 07 Oct 2026 00:00:00 GMTJev in Production: 703 Turns Later, Still a Complementhttps://mauriciosuarez.dev/en/writing/typesafe-jev-part-2/https://mauriciosuarez.dev/en/writing/typesafe-jev-part-2/Across 703 real WhatsApp turns, Jev responded in 128 of 131 main-classifier failures: new shadow-mode signals, not a proven replacement.Fri, 02 Oct 2026 00:00:00 GMTThe endpoint that took 29 seconds (and nobody complained)https://mauriciosuarez.dev/en/writing/endpoint-29-seconds/https://mauriciosuarez.dev/en/writing/endpoint-29-seconds/Autopsy of a ~29s production request: the AI call, the PDF render, and ~20 R2 downloads hidden inside an email — and the pattern that got it under 500 ms.Sat, 26 Sep 2026 00:00:00 GMTtry/catch won't help when the process is already deadhttps://mauriciosuarez.dev/en/writing/the-30-second-wall/https://mauriciosuarez.dev/en/writing/the-30-second-wall/A production FatalError at 30 s: the AI SDK was waiting for 60, PHP killed the worker first. The timeout hierarchy every backend needs.Sat, 26 Sep 2026 00:00:00 GMTWe tested cheaper models and decided not to migrate.https://mauriciosuarez.dev/en/writing/modelos-baratos-no-migrar/https://mauriciosuarez.dev/en/writing/modelos-baratos-no-migrar/A migration promised to simplify MaltaClean's concierge and lower the cost of each classification. Eight test cases showed why the savings weren't enough.Thu, 24 Sep 2026 00:00:00 GMTThe model interprets; code authorizes actions.https://mauriciosuarez.dev/en/writing/modelo-entiende-codigo-decide-permiso/https://mauriciosuarez.dev/en/writing/modelo-entiende-codigo-decide-permiso/A cleaning enquiry was mistaken for a job application. MaltaClean taught me why AI agents can interpret intent, but code must authorize actions.Wed, 23 Sep 2026 00:00:00 GMTThe AI Model That Can't Write (and Why I Want It in Production)https://mauriciosuarez.dev/en/writing/typesafe-jev/https://mauriciosuarez.dev/en/writing/typesafe-jev/I tested Jev, TypeSafe's first System One model: no text generation, just typed decisions. 292 cases, blind review, and 94.2% accuracy classifying a real cleaning company's WhatsApp.Sat, 19 Sep 2026 00:00:00 GMT