Introducing custom agents

Taito.ai ships with agents for the work every company does the same way: onboarding, pre-payroll, reporting, review cycles. You turn one on and it runs. The flows that aren’t the same everywhere still meant a checklist someone worked through by hand: the exception path unique to your team, the extra step your last audit added. Custom agents close that gap: describe the workflow once, and Taito.ai builds the agent that runs it.

Describe the outcome, not the steps

Building a custom agent starts with a plain description of what you want done, not a flowchart of how to do it. Tell the assistant the outcome: every new manager runs a 30-60-90 check-in with their report, or any contract that isn’t countersigned five days after the offer gets flagged. The assistant asks back only what it needs to run the thing: who approves the exception, which manager tier this applies to, what counts as done. It’s the way a competent hire asks before taking over a task.

Once it has enough to work with, the assistant hands back an agent scoped to that exact flow: the trigger that starts it, the steps it takes, and the point where it stops for you. You review it once, turn it on, and it runs from there. Nobody writes a workflow definition or touches a rules engine; the assistant did that translation in the conversation you were already having.

What an agent does that a checklist cannot

A checklist is a record of what should happen. It doesn’t send the welcome message, remind the manager their new hire starts Monday, or notice that a task is three days overdue. A person still has to read it and act. An agent runs the same flow and does the work: it drafts and sends the welcome messaging on the schedule you set, plans the first-week tasks and assigns them to the right person, chases the manager when their part is late, and books the recurring 1:1s that keep the process from quietly stopping after week one.

That’s the difference between a document and a worker. The checklist tells you what’s outstanding; the agent closes it out and only surfaces the parts that need a person:

• Sending the welcome message, the access request, and the first-week task list
• Chasing a manager task that’s gone quiet
• Booking and rebooking the recurring check-ins around calendars
• Flagging the one step it can’t complete on its own

Anything irreversible, such as a pay change, a termination, or a contract term, still stops and waits for you. The agent runs the process; it doesn’t make the calls only a person should make.

For the manual version of this same job, see how to automate employee onboarding. And if you’re weighing when a rule-based workflow is enough versus when it needs an agent, that distinction is covered in what is HR automation.

Where they run

Custom agents run wherever the ready-made ones do. Build and review one in the web app, where you see the whole flow before you turn it on. From there it runs quietly in the background and reports through Slack, where your team already gets a time-off approval or an onboarding reminder. Anyone connecting Taito.ai over the Model Context Protocol can also trigger or query a custom agent from Claude, ChatGPT, or whatever client they already have open, scoped to what that person can already see. The same agent runs across all three surfaces, with no separate setup for each.

Custom agents are generally available on every Taito.ai account today. Build one from the assistant in the app; the ready-made agents are still one toggle away.