Inspect the initiatives

Catalyst Group / Internal experiments

Real experiments. Open to inspection.

An AI agent fleet, a public status dashboard, and the operating notes behind them. Catalyst Labs tests ways of working inside human-led Catalyst Group and publishes the evidence, including the limits.

Inspect the initiatives Read the field notes

Inside Catalyst Group

A workshop with a public notebook.

Catalyst Labs is the internal incubator and experiment arm of Catalyst Group. The company is human-led; Labs is a place to test ideas and publish the evidence. The agent fleet is one experiment inside it, with no day-to-day authority over the business.

Where Labs fits
01

Public initiatives

Open the work. Judge the evidence.

Two exhibits from inside Labs. Each has a defined scope and a public place to inspect it. Neither puts agents in charge of Catalyst Group.

  • Internal experiment / Public notebook

    AI Fleet

    Testing how AI agents can organize like a company.

    Named roles, task ownership, and review chains inside a bounded experiment. The public notebook covers the squad, coordination model, delivered outputs, and failure modes. Human leadership retains business authority; the fleet is not a commercial product.

    • Agent organization
    • Review chains
    • Operating notes
    Explore the experiment
  • Public exhibit / Read-only

    Herman Monitor Dashboard

    A wall display for observing system status.

    A public status instrument with a deliberately limited role: observation. It shows status without offering operational controls. It is an output of the Labs experiment, not a console for running the company.

    • System observation
    • Read-only
    • Public output
    Open the status display
02

What we study

The questions behind the work.

Our public work examines how agents coordinate, how people observe a system, and what evidence makes an output ready to share.

  1. 01

    Agent organization

    Can a fleet of agents work with clear roles, bounded tasks, and someone else checking the result?

    • Named roles and review chains
    • Agent orchestration infrastructure
    • Model choice matched to the task
  2. 02

    System observation

    What should a status display tell a person at a glance, without giving that display control over the system?

    • Read-only status displays
    • Clear limits on what a view can do
    • Public information without private operating detail
  3. 03

    Delivery evidence

    What separates a plausible draft from an output that another person can inspect and accept?

    • Independent review of the work
    • Checks on the rendered result
    • Published outputs and candid operating notes
03

How we work

A result needs more than a draft.

The AI Fleet experiment makes its working standard public: define the task, keep the evidence, and separate doing the work from accepting it.

  1. 01

    Set the question

    Name the problem, the owner, and the evidence that would answer it. Keep the task small enough to inspect.

  2. 02

    Run a bounded test

    Give each piece of work a clear scope. Keep the documents and outputs so another reviewer can follow what happened.

  3. 03

    Review the result

    Check design, security, and release readiness. Inspect the rendered output. The person or agent producing a draft does not supply its only review.

  4. 04

    Publish the evidence

    Share accepted outputs with their scope and limits. Record the failures and unverified edges so the next experiment can learn from them.

04

Start a conversation

Bring a problem we can examine.

Tell us who encounters it, what happens today, and what evidence would show an improvement. For a Labs question or a client project, start with Catalyst Group.

Email Catalyst Group
contact@catalystgroup.tech
Visit the company
catalystgroup.tech