You probably already have one. You just do not have the language for it yet.
This person on your team who quietly figured out how to actually get work out of agents. More than simply prompt engineering, the Agent Operator is what we call the discipline they have already started building.
They design, run, evaluate, and improve workflows that use AI agents.
They sit between human intent and machine execution: mapping work, shaping context, connecting tools and data, supervising agents, verifying outputs, writing the practice down, and keeping workflows current as models change.
They came from engineering, product, design, ops, finance, GTM, support, data, or IT. The background gave them domain judgment. What we provide is the shared vocabulary, the method, and a community of people figuring out the same things in parallel.
The skills they are already building.
Workflow literacy
reading how a piece of work actually moves
Context engineering
briefing the agent so it can do the work well
Tool orchestration
connecting the agent to data, APIs, and other agents
Delegation
choosing what to give the agent, what to keep
Evaluation
judging whether outputs are good, with evidence
Verification
building loops that catch the wrong answers
Documentation
writing the practice down so it transfers
Change management
keeping the workflow current as models change
Continuous learning
reading the field, swapping notes, staying sharp
Six rungs. Most teams are at L1.
A few are stuck at L2 calling it L4. The diagnostic places you on the ladder honestly. Office hours keep you climbing.
you cannot skip rungs · the work shapes the discipline
- L0
Tool dabbling
A few people use chat. Nothing is shared.
- L1
Personal AI workflows
Individuals have private rituals that work for them.
- L2
Repeatable individual practice
Some workflows are written down somewhere by someone.
- L3
Shared team playbooks
Two or more humans can run the same workflow and get the same shape of result.
- L4
Instrumented agentic workflows
Workflows have evals, owners, and a measurement plan.
- L5
Agent Operator function / team
Named role. Operating cadence. The practice is part of how the org works.
Adjacent roles people confuse with this one.
What the role looks like, hour by hour.
Morning
Reads the overnight session traces from agentic workflows that ran without supervision. Flags one for review.
Mid-morning
Sits with a PM whose research-synthesis flow has been drifting. Walks through the prompt, the eval, the failure cases. Updates the playbook.
Lunch
Reads two field notes from the team Slack. Adds a comment to one. Does not pretend to read the third.
Afternoon
Pairs with an engineer to install verification on a coding-agent loop that has been merging too fast.
Late day
Writes 200 words of pattern notes from the morning session. Pushes to the team notebook.
Want to install the role on your team?
The 90-day program is how we install it: workflows, playbooks, evaluation loops, and trained internal champions, left behind in your team's voice.