How to build AI agents that work: start with the simplest thing
Most AI agent projects get canceled over cost, unclear value or missing controls. When you actually need an agent, what to build first, and how to shape it so it's dependable.
¿Prefieres español? Lee esta página en español
Blog · Bitácora
How I turn messy AI experiments into software that behaves. Short, practical, written between builds.
Most AI agent projects get canceled over cost, unclear value or missing controls. When you actually need an agent, what to build first, and how to shape it so it's dependable.
An agent that's right 90% of the time fails far more than it seems. How to build evaluations for AI agents from real tasks, what to grade, and why to read the transcripts.
Tools are how an AI agent acts on your systems, and the part that changes its results the most. How to choose them, name them, describe them and make them hard to misuse.
An AI agent doesn't remember your business. It only sees what's in its context at that moment. How to decide what goes in, what comes out, and how to keep focus on long tasks.
What the 2026 studies say about AI adoption in companies across Latin America and Colombia, why the numbers disagree, and what the companies getting real value do differently.
A six-step guide to implementing AI in your business and automating processes with it, from choosing the right process to running it in production with measurable results.
Only 6.8% of small companies in the region use AI in their operations. How a small business can start with one process, measure the result and know when to bring in an AI consultant.
An AI agent on WhatsApp can answer, book and quote, and it can also make up a price. What a business AI agent needs before it goes live.
A tour of the experiments every agent goes through before it ships, from the writing test to the night watch.
Unit tests catch what breaks. Evals tell you whether a change made things better or worse, on examples that look like real life.
Free-form text is where AI pipelines quietly break. Give the model a shape to fill and validate every answer against it.
Most unreliable AI features fail before the first model call. A one-page brief is the cheapest fix there is.
A short checklist for attacking your own AI feature, and what each attack should run into.
Models, data and prompts all change after launch. Monitoring is how you notice before your users do.
What working with a lab in Colombia looks like day to day, from time zones to languages.
No entries with that topic yet.