You made the agent useful for you. The natural next ask is “can the team use this?” — and the working answer is a rollout, not a memo. Rollouts succeed on the same shape every time: small start, visible wins, written guardrails, then widen.
1. Pick a pilot that can’t hurt much
Choose work that is high-volume and low-blast-radius: drafting, summarizing, meeting notes, first-pass research, form-filling. Keep payables and anything regulatory out of pilot one — not forever, just not first. Set a pilot window (30 days is plenty) and a success measure that maps to saved hours or turnaround time. A pilot without a measure is a demo; with one, it’s a decision.
Note what is NOT a pilot: turning it on company-wide with an FAQ. That skips the step that produces your guardrails — real usage.
2. Write the one-pager — before you need it
One page, plain language, three sections:
WHAT THE AGENT MAY DO / touch — tools & data it may reach WHAT IT MAY NEVER DO — money, people-data, deletion, sending as you WHAT TO DO WHEN UNSURE — stop, ask a named person (you), then act
The third section is the one people actually need; the first two are easy to write. The same page doubles as your pilot success criteria and your training handout — one artifact, three uses.
If this page isn’t written when your first incident hits, the incident writes it for you — badly. Write it while it’s cheap.
3. Run the pilot like a project
Real usage is the curriculum. Log the wins as they land: “drafted 40 emails,” “summarized 6 contracts,” “built the report skeleton.” Share them in the team’s existing channel — not a special AI channel. Keep a running list; it funds the go/no-go decision and becomes your “what good looks like” training material.
Two cheap accelerators: a weekly 30-minute clinic where people bring stuck prompts, and a shared prompt-snippet file so one person’s solved problem becomes everyone’s default. Competence spreads through examples, not memos.
4. Decide, then widen with structure
At the window’s end, review against the measure. If it holds up: widen access, formalize the one-pager into policy, and appoint the go-to person — formally, with a name. If it doesn’t: narrow the scope and run the focused pilot that failure just designed for you. Either outcome is progress; only skipping the review is a mistake.
When you widen, sequence the risk: new groups start on the pilot’s proven use cases, not on whatever they imagine. New capabilities are earned per use case, the same way connections are earned per tool in the Security module.
5. What actually sinks team rollouts
In the field, three failure modes repeat: the silent pilot (wins invisible, momentum never forms), the tool without a judge (nobody is empowered to say “stop, that’s out of scope”), and policy as PDF (a document nobody can find at the moment of decision). All three are fixed by the same move: a named owner, one visible habit, one page. This module is that page.
Knowledge Check
Three quick questions to test yourself. Nothing is submitted; this is just for you.
1. The first team pilot should be…
2. The one-pager section people actually need at decision time is…
3. The pilot window ends. You…