Testing Jev on Financial Documents

Jev identifies earnings changes more accurately than the tested classifiers. Trained models do better on filing categories and further SEC comments.

Oct 13 • Stefan Jansen

7

1

What Leading Hedge Funds Disclose About AI For Investment Research

What funds delegate, where investment expertise enters, and what their reported results measure.

Oct 6 • Stefan Jansen

7

1

1

September 2026

How AI Labs Build Training Environments for Agents

Three research systems construct tasks, tools and evaluators together. Their experiments show why solving a problem and executing it through tools need…

Sep 30 • Stefan Jansen

1

Measuring Self-Improvement in AI Agents

From better outputs to systems that improve their own improvement process.

Sep 24 • Stefan Jansen

1

Recursive Self-Improvement: A Reading List

A practical path through feedback, memory, code evolution, automated research, and the limits of current evidence.

Sep 22 • Stefan Jansen

3

1

4

How to Design and Evaluate Skills for Research Agents - and the ML4T Skills Release

We are releasing 61 open-source ML4T skills for quantitative research. Two new benchmarks show where written procedures help and why strong verification…

Sep 1 • Stefan Jansen

7

1

2

August 2026

How to Evaluate a Financial Research Agent as a System

From task interpretation and evidence retrieval to calculation, judgment, synthesis, and repeated-run reliability.

Aug 27 • Stefan Jansen

3

1

Six Libraries for Quantitative Research in the Age of Agents

Data, features, models, diagnostics, backtesting, and live trading are now stable on PyPI. Why we built them, and why coding agents make a tested tool…

Aug 13 • Stefan Jansen

16

5

July 2026

How Bridgewater Engineers a Research Agent

Five design decisions from PAT that do not need a 50-year archive, and the point-in-time test the talk leaves open.

Jul 29 • Stefan Jansen

7

1

The Third Edition Publishes Tomorrow

What is new since 2020: the research process front and center, nine cross-asset case studies, six open libraries, 459 notebooks, 112 primers, and 61…

Jul 23 • Stefan Jansen

8

2

2

The Coding-Agent Toolkit: A Workflow Loop You Can Install

Seven host-neutral steps to improve agent productivity, now public, that run the same on Claude Code and OpenAI Codex.

Jul 7 • Stefan Jansen

7

2

A Coding-Agent Reading List: Behind the Loops

Loop engineering is only the surface. A reading path through the older control problems underneath — and the line between what a coding agent may change…

Jul 2 • Stefan Jansen

9

4