Most Enterprises Shouldn't Train Their Own Models
The durable enterprise advantage is the learning loop that turns real failures into tested improvements.
4 min read
Hamza Mostafa
I build things on the internet and write about systems, creativity, and craft.
About
I'm a builder and tinkerer. Previously: Agent team at OpenAI. On leave from CS at the University of Waterloo, KP fellow.
Right now I'm exploring multi-agent systems, long-horizon agents, and reinforcement learning.
If you want to chat, say hi on X.
Writing
All postsShort notes on building, taste, and the kinds of problems that keep me interested.
SalesBench: The Long-Horizon Agent-to-Agent Eval
A long-horizon RL environment where a small model learns to manage an insurance sales pipeline against an LLM buyer, scored by revenue closed instead of by an LLM judge. The trained model vastly outperforms the untrained base, and the gap widens as the eval gets harder.
12 min read
The Agent Research Loop
What Karpathy's autoresearch really means, where agent systems are headed, and an open-source harness that ran 550 experiments over a weekend.
8 min read
I Run a Personal AI Agent 24/7 on a Mac Mini. Here's How It Actually Works.
A Mac Mini, some markdown files, and seven communication channels. Inside the setup that gives me a 24/7 AI assistant that monitors my email, iMessage, WhatsApp, and Twitter - and actually does useful things.
12 min read
I Let AI Agents Train Their Own Models. Here's What Actually Happened.
Two frontier agents, a pile of bugs, and a reality check on the future of autonomous AI research.
7 min read
Contact
The fastest way to reach me is via a DM on X or a note on LinkedIn.