Judgment on a trigger
Custom agents
A custom agent is an assistant that runs without you in the chat. Something wakes it (a time of day, new data, a message from another system) and it handles the job, then goes back to sleep. It does not replay a script: it looks at what is actually there and responds to it, so Tuesday's run can come out different from Monday's, because Tuesday is different.
Agent or automation
The same event, two different tools.
A new lead arrives. An automation saves it to your table: the same exact step, every lead, every time. A custom agent reads what the lead actually asked for and writes a welcome that fits them: different for every lead.
If the steps are identical every time, use an automation. It is faster and cheaper. If each run needs a judgment call, that is an agent. Reach for the automation first, and use an agent where the judgment is genuinely worth it.
Triggers
Three ways an agent wakes up.
On a schedule.“Every weekday at 8 AM”, “every Monday morning.”
When data changes. Whenever specific information is added, updated or removed, like a new lead landing.
When a webhook fires. When an outside system sends it a message, such as a notification from your payment processor. The incoming details are handed to the agent so it can act on them.
Any agent can also be run on demand: ask in the chat and it wakes immediately, whatever its normal trigger, and you can hand it information to work with. That is the easiest way to test one before you rely on its schedule.
What it may do
Every agent gets the smallest set of abilities its job needs.
An agent can do most of what Inscendo IQ does in the chat: look things up, search the web when you allow it, read and work with your data, send through connected services. But you decide which of those each agent gets, per agent, not as one blanket setting shared by all of them.
Two limits always hold. An agent can read your widgets and automations to understand them, but it can never create or change them. And it never acts outside the abilities you granted. If it needs something it was not given, it stops and leaves a note saying exactly what was missing rather than working around it.
What it may change
No chain reaction you did not sign off on.
If an agent will write or change data, you approve up front which kinds of changes are in scope, including which other automations those changes are allowed to set off.
Changes inside that scope happen quietly: that is the agent doing its job. Anything outside it is held for your review instead of being performed, and the agent carries on with the rest of its work rather than stalling. So it can neither surprise you nor get stuck waiting.
Cost and oversight
You set how much thinking each run gets.
Every agent has its own capability level, chosen when you create it and changeable later. It applies on every run, which matters more here than in chat because an agent may wake many times a day. Simple logging jobs run at the cheapest level; reserve the most capable one for runs that genuinely need judgment. Inscendo IQ walks you through that trade-off before it creates anything.
Each wake is recorded as its own run, and you can open any run to read what it looked at, the steps it took and what it concluded. Or just ask in the chat: “is my morning briefing agent working?” and it will go through the history and explain.
Describe the job and when it should happen.
Custom agents are created the same way as everything else here: say what you want to happen and when, review what it proposes to do and what it may change, and confirm.