---
title: "Labs: what we are building next | Agrim Intelligence"
description: "What Agrim is building next, in the open: Sigma Loop, experiments in progress, and notes on what does not work yet."
source: https://agrim.ai/labs
---

Labs

# What we are building next, in the open.

Labs is where our work sits before it is a product: an open-source tool, experiments, and notes on what did not work. Every experiment is dated and says plainly what stage it is at.

[Open Sigma Loop](https://agrim.ai/sigma-loop) [View on GitHub](https://github.com/Agrim-Intelligence/sigma)

1

## Open source

Sigma Loop is the tool we use to build our own products.

Sigma Loop is an open-source tool that makes AI coding agents plan, get reviewed and leave a record.

### Sigma Loop

In use · Open source · updated November 2026

What it is A fixed process for AI coding agents: the plan is written and challenged, the work is reviewed, and a record is kept.

Stage In use. We build our own products with it, and anyone can read the code.

Not yet Sigma Loop has run end to end only on Claude Code, by its maintainers. Codex is partly validated, Cursor is experimental, and some of its rules are instructions that no code enforces.

[About Sigma Loop](https://agrim.ai/sigma-loop) · [View on GitHub](https://github.com/Agrim-Intelligence/sigma)

2

## Experiments

Work in progress, labeled by how far along it is.

In design. An idea on paper. Nothing runs yet.

Prototype. It runs, on our own work only.

In use. We rely on it. It may still change.

### Genovis

Prototype · updated October 2025

What it is A tool that helps a company's own content appear in the answers AI assistants give, now that buyers ask an assistant where they used to search.

Not yet It runs on Agrim's own content only, and we have not measured whether it changes what assistants answer.

### Litmus

In design · updated August 2026

What it is A tool that opens your application as different kinds of user and records, with the time, whether each rule holds: for example, whether one user can reach another user's records.

Not yet Nothing runs yet. It will report each check as passed or failed, and leave the judgment to the auditor.

Careful is faster.

We write down what failed and why, so that the next step is a smaller one.

3

## Notes

What did not work, and why.

### Prompting AI one piece at a time

Note · from building our own products

What happened We built software by prompting AI one piece at a time. Each piece was written without knowledge of the rest. It solved the same problem a different way each time, and the product broke.

What we changed We gave the AI a shared memory of the whole project and a fixed process to follow. That fixed most of it.

### AI writes faster than anyone can read

Note · from building our own products

What happened By the time you notice a wrong turn, a great deal has been built on it. Left alone, the work becomes a black box.

What we changed The plan is written and challenged before the work starts, and a person approves each step that matters.

[Longer write-ups are in Insights](https://agrim.ai/insights)

## See how Sigma Loop runs one real piece of work. The plan, what review found, and the finished change.

[Open Sigma Loop](https://agrim.ai/sigma-loop) [View on GitHub](https://github.com/Agrim-Intelligence/sigma)
