Agents and automations, together
Outcomes
Two kinds of help
Workflows carry the routine; a worker stands at the judgment moments. Composing the two deliberately starts with telling them apart: the automation is a workflow you design, and the AI Employee is a worker you brief. Each deserves exactly one sentence:
- An automation is deterministic: the same trigger runs the same steps and produces the same result, every time.
- An AI Employee is agentic: it reads each situation and exercises judgment, so no two interactions are handled identically.
Determinism is a feature, not a limitation. A receipt should not be creative. A record update should not exercise judgment. When the logic is fixed, an automation executes it perfectly, instantly, and without asking for anyone's attention. And judgment is a feature that would be wasted on fixed logic: an AI Employee earns its keep where every instance is a little different.
The sorting test
Rule-shaped or judgment-shaped: that old sorting tool is the whole decision.
- Rule-shaped work goes to an automation. If you can write the steps down and they never branch on judgment, you have written the automation already.
- Judgment-shaped work goes to an AI Employee. If every instance needs interpretation, brief a worker instead of drawing a flowchart.
Partners in the field put a sharp edge on the first half: if your automation requires manual input, it is not an automation. A workflow that pauses for a human to fill something in is a task list wearing a costume. Design automations to run start to finish on their own, and give the judgment calls to a worker built for them.
Better together
The two systems are at their best composed, because each hands the other exactly what it lacks. Here is the composition partners deploy most, end to end:
Asking for the review is rule-shaped: every completed job gets the same request, reliably, with nobody remembering to send it. Responding to the review is judgment-shaped: a glowing five-star note and a lukewarm three-star one deserve different words, and the AI Reputation Specialist writes each response to fit. One outcome, a review engine that runs itself, built from two systems doing what each does best.
The handoff runs in both directions. Automations can put AI Employees to work, and what an AI Employee does can start an automation: a lead captured in conversation can kick off a follow-up sequence the moment the chat ends. Build an automation with an employee at the end of it and you are composing, not just configuring.
Take one client process, start to finish, and draw a line through it: which parts are rule-shaped, and which need judgment? That line is an implementation plan. The rule side becomes an automation; the judgment side becomes an AI Employee brief.
The complete picture
The full vocabulary fits in four lines:
- Knowledge is what an AI Employee knows: the business facts behind every answer.
- Capabilities are what it does: the instructions that shape its behavior, present in every conversation.
- Tools are how it acts: scoped doorways into the systems where work happens.
- Automations are how work finds it: deterministic workflows that trigger, and are triggered by, your workforce.
Every configuration screen you meet from here maps onto one of those four, and any requirement a business brings you belongs to one of them.
Knowledge Check
Three quick questions on sorting work between automations and AI Employees, and on the handoff between them.