AI News · The field notes
The Daily AI Drop - Morning Edition, November 4, 2026
Crouton Alpha, a free stealth model for subagents, a walkthrough on building self-improving AI agents, and why OpenAI acquired drug development startup Orchestra Bio.
Crouton Alpha, a free stealth model for subagents, a walkthrough on building self-improving AI agents, and why OpenAI acquired drug development startup Orchestra Bio.
Listen: Listen to this episode
Crouton Alpha: The Free Stealth Model Built for Subagents — Fahd Mirza
Fahd Mirza tests Crouton Alpha, a free stealth model, with Hermes agents. It matters because subagents do the quiet background work in multi-agent systems, and a capable free model lowers the cost of experimenting with those setups. His hands-on test gives a practical read on whether the model is genuinely useful, not just intriguing.
How To Build Self-Improving AI Agents (Full Walkthrough) — Aishwarya Srinivasan
Aishwarya Srinivasan breaks down recursively self-improving agents from first principles, including the inner task loop and the outer improvement loop. She covers bi-level optimization, versioned agent harnesses, and techniques to prevent reward hacking. It matters because it separates the hype from the engineering: self-improvement only works with loops, measurements, and guardrails, and this walkthrough shows how to build and evaluate such agents safely in production.
Why did OpenAI buy a drug development startup? — Aish Reganti
Aish Reganti explains OpenAI's acquisition of Orchestra Bio, a startup building AI agents that coordinate drug development. When a manufacturing batch fails, its agents find the affected trials, teams, and timelines and lay out the options, while people make the final call. It matters because getting one drug approved can take 10 to 15 years and cost more than $2.6 billion, much of it spent on trials, manufacturing, and approvals — exactly the kind of messy, high-stakes coordination where agents can help without replacing human judgment.