#agentesia

La Inteligencia Artificial conversacional ha dejado de ser un lujo exclusivo de las grandes...

Prowl como Termómetro de Utilidad Prowl asigna scores de relevancia a APIs según su uso y...

The most common mistake when implementing MCP tools is coupling them to the provider's SDK. The spec exists to prevent exactly that. A practical architecture guide: the input/output contract that makes a tool work across Claude, GPT, and local models without rewriting anything.

Two weeks using Cline as an autonomous coding agent on a TypeScript project. What tasks I delegated, where it screwed up, how it compares to Claude Code, and which workflows I'd never hand over to it. Architecture-grade analysis, not hype.

La semana pasada me llegó un post viral de LinkedIn con este titular: «🚨 Chinese AI agent created...

I ran git blame on a project where I used Claude Code heavily. 61% of the lines aren't mine. That's not a legal problem yet — it's an accountability problem for when something blows up in production and nobody knows whose signature is on it.

I ran GPT-5.5 against my actual production prompts and compared it to GPT-4o on latency, cost, and output quality. The marketing leap doesn't match the leap in my metrics. Here are the numbers.

The HN post has 127 points and nobody's talking about the real problem: when an async agent fails, you don't get a stack trace. You get silence. And silence in production is the worst bug that exists.

I've been running MCP in production for weeks and the problem I found isn't in any Dev.to post. MCP assumes context doesn't change between calls. My agents live in contexts that mutate. Three bugs that passed every test and failed in prod in ways that took me days to understand.

Anthropic just said Claude CLI usage in the OpenClaw style is allowed. Last week I was doing it with a knot in my stomach. Today I'm doing the exact same thing. What changed was the paperwork. And that tells me something pretty uncomfortable about what it means to build on platforms that rewrite the

Anthropic acaba de decir que el uso de Claude vía CLI al estilo OpenClaw está permitido. La semana pasada lo hacía con cierta incomodidad. Hoy lo hago igual. Lo que cambió fue el papel. Y eso me dice algo bastante incómodo sobre qué significa construir sobre plataformas que reescriben sus reglas sin avisarte.

El CAPTCHA nació para demostrar que sos humano. Ahora mis agentes necesitan demostrar que son bots para funcionar. Medí los retries, el overhead, y la fricción real. Los números son feos.