Overview
The parts of an agent product that seem like an afternoon's work, and what they turn into once real users arrive.
Most of what an agent product needs sounds simple. Save the conversation. Ask before sending an email. Keep going if the server restarts.
Each one is simple in a demo. Then a user opens a second tab, a server restarts in the middle of an answer, a teammate approves from Slack while someone else approves in the browser, and the simple version starts doing things nobody intended. Often quietly: nothing errors, you just pay twice, or show the wrong answer.
These pages walk through those parts one at a time: what the simple version misses, and what Latch does instead. Each ends with the few lines it takes.
Conversation historySaving messages is easy. Two tabs, retries and forged messages are not.ApprovalsAn Approve button, clicked tomorrow, twice, from Slack.Running on many serversA crash mid-answer, five servers, and exactly one of them finishing it.Who sees whatCustomers, users, teams, and the runs nobody is logged in for.Usage limitsCaps that hold when five requests arrive in the same moment.Files and imagesWhere files live, how they reach the model, and what never leaves your servers.StreamingThen the user closes the tab, reloads, or sends another message.SchedulesA run with nobody there, on every server at once.Connecting accountsSigning in mid-conversation, staying signed in, tokens the model never sees.SubagentsA second conversation that can pause, crash and cost money.Chat appsPlatforms that resend, bots that answer themselves, buttons anyone can press.Agent versionsSomeone will edit an agent, break it, and ask what ran last Tuesday.