My Agent Stack For Automating My Personal Life
My agent manages my emails, SMS, WhatsApp, Telegram and pretty much everything to automate my personal life.
Nicolas Bustamante
Nicolas Bustamante's personal website.
Latest posts
Long Running Agent Engineering
Why the hard part of long running agents is not making the model think longer. It is building the harness, state, judges, and recovery loops around it.
Model-Harness-Fit
Why mixing a frontier model with a foreign harness quietly tanks performance, and what the open source code tells us about why.
Agent Memory Engineering
How do agents actually remember you? And why is moving from one agent's memory to another's so much harder than just copying files?
Hard Lessons Building Agents Since GPT-3.5
Three years shipping AI agents at Fintool. The hardest lesson isn't architecture. It's people. The second hardest: everything you build has a life expectancy of a few months.
Microsoft Has Acquired Fintool
Fintool, the AI-powered equity research platform I co-founded, has been acquired by Microsoft. Here's the story behind it.
10 Years Building Vertical Software: My Perspective on the Selloff
A decade building vertical SaaS, now building the AI that disrupts it. Which moats LLMs actually dissolve, and why the software selloff is only half right.
The LLM Context Tax: Best Tips for Tax Avoidance
The Context Tax is what you pay when you fill your context window with useless tokens.
The Crumbling Workflow Moat: Aggregation Theory's Final Chapter
Software's real moat was never data. It was the interface. LLMs absorb that layer, and Aggregation Theory finally reaches its endgame: API versus API.
ChatGPT Killed the Web: For the Better?
After 30 years of clicks and blue links, websites are becoming obsolete. LLM agents now read and act for us, ending search engines as we know them.
LLMs Eat Scaffolding for Breakfast
We just deleted thousands of lines of code. Again.
Lessons from Building AI Agents for Financial Services
Two years of building AI agents for financial services, distilled into battle scars and lessons on sandboxes, context, skills, S3, Temporal, and evals.
Model-Market Fit
Product-market fit has a prerequisite most AI founders ignore: the model itself must be capable of doing the job. That's Model-Market Fit.
Are LLMs Plateauing? No. You Are.
When you test frontier models on yesterday's benchmarks, you'll miss today's breakthroughs. You're weighing an elephant with a bathroom scale.
Domain names I own
The full list of domains I own.