Hand off the work. Keep the decisions.
We built Tandem to queue coding agents on the right machine, ask the decisions that need a person, and hand back work ready to review.
At a glance
01 · Problem
AI agents are fast, but someone still has to babysit them.
Work we meant to hand off still tied us to a desk.
Why it hurts
Agents stop to ask a question with nobody there, run only on the machine in front of us, and forget how we like things done. Finished work gets scattered across terminals.
02 · User
- Answer agents from the phone
- One list of what's running
- Never miss a hard-to-undo call
03 · Flows
- QueueTask, agent, machine
- Pick upThe right runner starts
- WorkClaude or Codex
- DecideA person, or Autopilot
- ReviewCheck the returned work
- ShipPR or files + report
04 · Wireframes
3 screensFlow sketches reconstructed for this case study.
Why the decision stands out
The flow gives decisions their own step. Routine choices can settle through Autopilot or a timed proposal; anything hard to undo always waits for a person.
05 · Design system
Palette studyIllustrative colours; native type follows each platform.
Mac + iPhone: SF Pro, SF MonoWindows: Segoe UI Variable, Cascadia Mono06 · Final


07 · Results
- Agents can work across
- Mac + Windows
- Queue, decide and review from
- iPhone
- Hard-to-undo actions
- Wait for us
How we measured
As of 15 November 2026, native iPhone, Mac and Windows apps share one backend. The first commit was 26 October. Agents were observed running on Mac and Windows at the same time in one snapshot. We have not measured human time saved.
08 · GitHub
Source code is private. We can discuss the approach on a call.
Private repository