Our request for startups fits in one sentence: computer-use agents that actually finish the job. Agents that companies can trust to do real work at scale, and that work in the software companies already use.
Each part of that sentence is doing work. Here is what we mean by it.
“Actually finish the job”
Completion is the whole point. An agent that gets most of the way through a task and then hands a half-finished mess back to a person hasn’t saved much time — and may have created new work, because someone now has to figure out what was done and what wasn’t.
Finishing means:
- Completing the task end to end, across every system it touches.
- Checking the outcome, not just the action — reading back the record that was updated, confirming the email was sent, reconciling the totals.
- Recovering from the unexpected: a timeout, a pop-up, a changed layout, a missing field.
- Knowing when to stop and ask, and asking a precise question with the full context attached.
“Trust to do real work at scale”
Real work means production systems, real money, real customers. Scale means thousands of runs a day, not one supervised session.
At that volume, trust can’t rest on someone watching. It has to be built into the product: least-privilege access, a complete record of every action, approvals at the moments that matter, predictable behavior from run to run, and a clear way to see — and undo — what happened.
“In the software companies already use”
Companies are not going to replace their ERP to make an agent’s life easier. The winning products will meet work where it already lives: browser-based SaaS, supplier and partner portals, legacy desktop apps, virtual desktops, spreadsheets, and email.
The best teams will use APIs where they exist and the interface where they don’t. Customers care that the job gets done, not how.
Measure success in jobs finished and verified — not demos delivered.
What we’d love to see
We don’t have a fixed list, and the best ideas will likely come from founders who have lived the problem. But here are the kinds of workflows that fit:
- Finance operations: invoice capture, three-way matching, and reconciliation across bank portals, ERPs, and spreadsheets.
- Insurance and claims: intake, document collection, and status-chasing across carrier and broker portals.
- Healthcare administration: eligibility checks, prior authorizations, and claim follow-ups across payer portals and practice systems.
- Logistics: booking, tracking, and exception handling across carrier and customs portals.
- People operations: onboarding and offboarding across HR, payroll, identity, and IT systems.
- Compliance: gathering evidence and entering the same information across many different portals.
The common thread: high volume, many systems, no clean integration, and a clear definition of done.
What we look for in a team
- You’ve lived the workflow. You know where it breaks because you, or people close to you, have been the human API.
- You’re obsessed with reliability. You think about the ten-thousandth run, not just the first demo.
- You have an evaluation culture. You measure completion on real tasks, and you know your failure modes by name.
- You have a wedge. One painful workflow you can own end to end before expanding.
- You’re honest about what doesn’t work yet. Clarity about the gaps is more convincing than a perfect demo.
How to reach us
Send a few lines about what you’re building, the workflow you’re starting with, who it’s for, and how you measure a job as done. Early is fine. Get in touch here.