We explore how AI can help people develop software and investigate investment ideas. Meet our two projects, both in development.
CSL Laboratories brings research questions into working projects. Our focus is on how people direct AI agents, review their work, and use evidence to make decisions.
Our current projects are Recursion, an agent harness for developers, and AI Investment Cockpit, a workspace for investment research.
An agent harness for software development.
Recursion explores how developers can coordinate AI agents and review their work. The aim is to connect model selection, task context, and review in a workflow developers can follow.
Interface concept · Scripted example This example illustrates a proposed workflow. It is not a live agent run.
Recursion interface concept showing a scripted example of model selection, a draft file change, and a second-model review. The example illustrates a proposed workflow, not a live agent run.
User Open the project spec.
Assistant The sample spec describes a summary task, a review step, and one open question.
Summarize the spec, then ask a second model to review the draft.
Selecting the preferred model…
Provider A is unavailable in this example. Continuing with the local model.
# Project spec
- Task: summarize the project structure.
- Review: ask a second model to check the draft.
- Open question: what should happen if a model is unavailable?docs/SUMMARY.md+2−1 ## Summary- - TODO: summarize+ - The example covers a summary task and a second-model review.+ - Open question: how should the workflow handle an unavailable model?Reviewed by a second model
A workspace for investigating investment ideas.
The Cockpit is a project to bring research notes, supporting sources, and strategy experiments into one workspace. It starts with a simple question: what would help a researcher examine an idea and decide what to investigate next?
Interface concept · Fictional data This example shows a proposed research and paper-trading workflow. No live trade or actual investment performance is shown. It is not investment advice or a promise of future returns.
AI Investment Cockpit interface concept with fictional research notes, sample strategy checks, and a simulated paper-trade approval. No live trade or actual investment performance is shown. This is not investment advice or a promise of future returns.
In this fictional example, a filing excerpt describes lower input costs under a new supplier contract1. A call transcript says the contract starts next quarter2.
A supplier note raises a delivery concern3. The potential benefit depends on that concern being resolved.
Can the supplier meet the delivery schedule before the contract starts?
Next step in this example: Review the supplier update when it becomes available.Lower-cost scenario
An illustrative line made from fictional values. It does not represent actual investment results or predict future returns.
Illustrative paper-equity chart · Fictional data · Not investment performance
12:04:31 · Sample paper-trade approval recorded · SAMPLE-AWe care about how a result was reached and what still needs checking. That means making room for sources, assumptions, and review as we develop our tools.
Explore Recursion and AI Investment Cockpit on GitHub.
View the projects on GitHub ↗