Turn work into an agent that keeps doing it.
Describe what needs to happen in plain language. Mad Labs builds the agent and the workflows behind it, then puts them to work. The agent runs in the background, responding to schedules, triggers, and events.
You set the direction.
Agents carry the work forward.
You should not have to build and maintain another app to get the work done.
Mad Labs turns a description of that work into an agent that owns the job. Mad Tracker starts with one recurring process. We’re building toward agents that share context and coordinate work across sales, finance, and operations.
You set the goals, boundaries, and decisions that need your judgment.
Start with one recurring job.
Tell Mad Tracker the result you want. It creates the workflow, watches the sources you approve, and delivers the result in your Google Workspace. No code or workflow editor required.
“Keep a spreadsheet of the receipts that arrive in my inbox.”
-
You hand off the job
Review the plan and approve Gmail access.
-
A new receipt arrives
The agent extracts the supplier, date, and amount.
-
Your spreadsheet updates
Each new receipt adds to the record.
Business receipts
| Date | Supplier | Amount | Source |
|---|---|---|---|
| Sep 7 | Harbor Print | $36.00 | Receipt email |
| Sep 8 | Northline Supply | $52.50 | Receipt email |
| Sep 9 | Cedar Office New | $84.00 | Receipt email · 09:42 |
In your workspace, linked to the source. The agent works after you close the tab.
“Watch this project’s emails and files. Update the delivery brief and draft a client update if the launch is at risk.”
-
1 · Monitor changes
Supplier email + project plan“Proofs will arrive Thursday, not Tuesday.”
Print handoff: Wednesday. Launch: Friday.
-
2 · Update the brief
Google Doc · Delivery briefPrint handoff blocked.
Friday launch at risk. -
3 · Prepare a response
Gmail · Draft for review“Proofs are delayed. We can explore expedited printing or a revised launch date.”
You choose the next move and approve the message. The agent keeps watching.
Always on. Still in your hands.
Approve the plan and permissions before execution. Inspect the work, pause the agent, or revise the job. You decide what gets sent.
An agent should earn its keep.
An agent only makes sense if it costs less than the value of the work.
Open-weight models let us choose the right model for each job and run it locally or on hosted infrastructure.
Our goal: agents you can afford to leave on, with control over the models, hardware, and bill.
Built for the work after the first conversation.
Mad Labs is a small team of AI researchers and engineers. We have built large-scale distributed systems at Twitter, VMware, and Etsy; search and recommendation systems at Etsy and Fiverr; and startups in AI and data infrastructure.
That experience shapes the infrastructure behind Mad Tracker: agents that remember what happened, recover when work fails, and keep useful results current after the first conversation.
What would you hand off to an agent?
Tell us about one recurring job your team wants to delegate. We’ll talk through the workflow and how Mad Tracker could help. A short description is enough. Please leave out confidential data.