Tag
#ai-agents
The Hidden Cost of “I Can Clone This in Five Minutes”
Claude Code makes the first version easy. Building my own personal agent, Maria, reminded me what the hype hides: testing, learning, improving and maintaining, forever.
What I Learned Building Carla and Andrea, My Two AI Agents
Two production agents, one on the phone and one on WhatsApp and email, rebuilt almost every day: evals and judges, shorter prompts, tools that know when not to fire, costs and model prices that keep moving, a North Star, and the humans in the loop, with the real numbers.
Let Your Agents Drive the Browser (With You Watching)
The internet was not built for agents, but agents now use it more than anyone. From Iceland bookings to Google Cloud, I log in and let them drive, at the speed I choose.
What If a Judge Fixed Your Agents?
An LLM judge checks two production agents every hour, tests each prompt fix on a copy first, then ships it. Three greens and the defect is solved.
I Used AI to Plan Iceland With Five Friends. The Hard Part Was the Humans.
Five friends, one Iceland road trip, a voting app, three PDFs and AI bookings. AI was the easy part: the real limit was context switching and humans in the loop.
The S.C.A.L.E. Framework I Use to Build AI Agents That Do Real Work
Five letters for thinking about AI agents — Scale, Consistent, Adapt, Learn, Escalate — told through Carla and Andrea, the voice and text agents I built for an Italian auctions consultancy.
Forget the Ladder: Learning Top-Down in the AI Era
Bottom-up learning is a ladder; with LLMs you can start from the top and go down the layers when you need them. How I learned ElevenLabs agents in months, and why my kids are learning Arduino the same way.
I Tested Claude Code’s Ultracode. It Was Amazing, Until It Burned My Whole Week in Half a Day.
I tested Claude Code's Ultracode orchestrator with a long epic and it ate my whole weekly limit in less than a day. Three lessons, and why I'm building a lighter version inside k0.