TL;DR. I tested whether an independent AI critic improves an agent’s research more than an agent reviewing its own work. An agent wrote 9 short equity research answers on Deere (3 questions, each run 3 times). A separate judge agent flagged a material error in 3 of the 9 first drafts, all the same omission. By its assessment, self-review fixed 2 of the 3 and an independent critic fixed all 3. The judge also classified 13 of the critic’s 41 complaints as invalid, yet the writer accepted 12 of them. For a task this size, self-review looks like the more efficient choice, though I’d want to test more questions before deciding whether the critic’s extra fix justifies 22.5% more cost. ...
The Evening Sweep: One Coordinator for an Overnight Agent Fleet
This past summer, I set agents to work through my backlog each evening and ended up with pages of open PRs I didn’t know what to do with. The program assigned batches of issues by priority and selected agents based on complexity. But whenever an agent encountered ambiguity, it would leave a comment and move on. I tried updating the label and categorization schemes, but they were all quite brittle and the unresolved work kept accumulating. Nothing in this process owned the outcome, so I never felt in control. ...
I Deleted My Multi-Agent Research System
Originally posted on LinkedIn on September 1, 2026. Earlier this year I built a multi-agent system for investment research. Several agents, each with a role, coordinated to produce a view on a set of stocks. I removed it in July. The system was complex and expensive to run, and I could not test it properly. You can’t write an assertion against a non-deterministic graph, so I had no reliable way of knowing whether the extra complexity was doing anything useful. When I eventually measured it against a much simpler quantitative approach, the simpler approach came out ahead. ...
Keeping an Agent Fleet Consistent: Policy Routing with oiax
Originally posted on LinkedIn on August 4, 2026. In my last post I introduced Nous Ergon, an applied AI lab. Since then, while building inside my multi-platform agentic workflows, I have been running head-first into a notable pain point: how to keep a fleet of agents doing things the same way? Every agent does things slightly differently. Same task, different direction, different quality. As a solo operator supervising this fleet, I have a strong interest in making sure the agents are delivering a standardized, conforming set of deliverables — so I wrote my expectations down. That corpus is now 35 policy documents and well over 100,000 tokens. ...
Introducing Nous Ergon
Originally posted on LinkedIn on July 22, 2026. In February, Block generously freed up 100% of my time. Since then, I’ve spent almost every waking hour building Nous Ergon, an applied AI lab. I have three products in flight: Crucible — an experiment harness for testing agentic and financial strategies Metron — an agent-powered investment dashboard Vires — a strength-training app with an AI coach I’ve also built mnemon, an open-source memory server for LLMs, which came out of a problem I kept hitting along the way: every LLM platform now ships with its own memory, but each one is a silo. Mnemon gives every LLM I run, across any client, one shared long-term memory. ...
Multi-Agent Research: How 6 LLM Teams Analyze 900 Stocks
In Post 1, I introduced Nous Ergon — an autonomous trading system that splits intelligence across four layers: LLM agents for research judgment, ML for pattern recognition, deterministic rules for execution, and a backtester for system-wide learning. This post goes inside the Research module — the layer where LLMs are found hard at work. What a Weekend Run Looks Like Over the weekend, an AWS Lambda fires. It loads the S&P 500 and S&P 400 — roughly 900 mid-to-large-cap US stocks — along with recent price history, and then distributes them across six sector-specialized teams that run in parallel. Each team screens, analyzes, and debates their sector’s best opportunities. A CIO agent evaluates the top picks across all teams and decides which stocks enter or exit the portfolio. ...
Nous Ergon: Building an Autonomous Alpha Engine with AI
The Thesis Can AI generate sustained market alpha — not through a single model making predictions, but through a system of specialized components, each contributing what it does best? That’s the question behind Nous Ergon: Alpha Engine (νοῦς ἔργον — “intelligence at work”), a fully autonomous trading system I’ve been building that combines AI-driven research, quantitative prediction, and rule-based execution. Quantitative finance — using mathematical models and statistical analysis to make investment decisions — has traditionally been the domain of institutional hedge funds with massive engineering teams. Large language models and modern machine learning tooling are changing that equation. ...