The opposite bet.

AI agents are getting very good at doing the work — and the infrastructure being built around them mostly optimizes for removing people. Agent email APIs advertise no human in the loop as a feature. That's fine for machine-to-machine chores. It's the wrong shape for work that matters: a report a client will read, a deploy that touches production, an invoice run, a pitch deck.

Traylet is the opposite bet. Work isn't finished when the agent stops — it's finished when a person has seen it and left their word. So we built the place where that happens: a tray where agents lay out finished work as real artifacts, where you, your team, or your client can look at it and respond over the channels everyone already has, and where a reply resumes the agent in seconds.

The mechanics came first. Traylet grew out of a private tool we ran for our own agents — inbound email through to rendered artifact links, used every day to hand work between agents and a human. The loop held up. Traylet is that loop, productized.

Traylet is built by Michael Cutler in Singapore.