Let me name an uncomfortable truth I see every week. I make keynote speeches on the “Future of Work and AI” to packed rooms and watch the same pattern. Leaders assume everyone is already fluent in AI.
It’s true that many people are dabbling, often using a chatbot like a smarter version of Google. Real integration of an AI copilot into daily work feels like a leap too far.
If that’s you, you aren’t behind. You’re still early in the curve, and you can move fast without breaking trust. Here is the 90-day plan I teach and have seen work across industries.
Week zero: Set the guardrails
- Pick one team, one workflow, one tool. Publish redlines for use: no personal or regulated data in prompts, every external message gets a human review, and every AI draft is labeled. You don’t need to invent policy from scratch. Use a neutral framework and adapt it. NIST’s AI Risk Management Framework and its Generative AI Profile offer clear controls for data handling, transparency, and human oversight.
- Agree on a simple scorecard you will review weekly for 90 days.
- Track cycle time, first draft quality, human edit rate and hours saved. Don’t forget to review defect escapes (AI copilot errors or mistakes that slip through before going live).
Take a quick trust pulse, as well. This is a lightweight, recurring survey or check-in that measures how much the team actually trusts the AI copilot’s output. Something as simple as: “On a scale of 1–5, how confident are you in the AI’s work this week?”
The word “pulse” signals it’s fast and frequent—not a deep annual survey but a regular gut check. It matters because if people don’t trust the copilot, they’ll either waste time over-editing perfectly good output or quietly stop using it altogether. Either way, you lose the efficiency gains.
Days 1 to 30: Prove value with lightweight jobs
Start where the impact is small and wins are clear—meeting notes and action items. Use AI to record and transcribe, then let the AI copilot draft notes and next steps. Humans confirm and publish.
There are a few ways to do this:
- Internal messages: Copilot drafts options, humans choose and edit
- Project updates: Feed the copilot issues and commitments, then generate a one-page update for stakeholders.
- FAQ drafts: Use past tickets or chats. Humans prune, while your legal department reviews anything external.
- Research briefs: Copilot outlines claims, sources, and what would falsify them. A human validates links.
- Summaries of long documents or threads: Ask for a one-paragraph summary and three open questions to guide a decision.
Rigorous studies show sizable productivity lifts on routine knowledge tasks, with the largest gains for novices. A National Bureau of Economic Research field study involving more than 5,000 support agents found a 14% to 15% increase in issues resolved per hour, particularly for less experienced workers.
In another controlled experiment published in arXiv, developers with a copilot finished a standard task 55.8% faster. Your mileage will vary, which is why you measure.
Trust habits for this phase:
- Label every AI draft.
- Keep a changelog of human edits.
- Publish a living list of approved and banned uses.
When the copilot makes an error, add it to an error library so that everyone can learn from the mistake more quickly.
Days 31 to 60: Graduate to structured outputs
Next, move to jobs where the AI copilot produces repeatable artifacts from known inputs
- Standard operating procedures: Feed recorded walkthroughs and tribal knowledge, and get a draft SOP. Two operators must run it end-to-end before it can be published.
- Job descriptions and interview guides: Calibrate for skills and outcomes, not buzzwords.
- User stories and test cases: Pull from specs and bug reports, generate edge cases and acceptance criteria. Engineering curates.
- Customer reply templates: Train on successful past replies. Humans run A/B tests for clarity and accuracy.
- Data hygiene and tagging: Let the copilot propose categories for feedback or tickets, then audit weekly samples.
As a culture multiplier, name your meeting types. This enhances AI summaries and action list accuracy while ensuring everyone understands the stakes, goals, and expected outcomes.
In our system, a “cockroach” is a 15-minute, single-topic huddle with optional attendance. “Tiger team” sessions are longer, mandatory, and produce specific owners and actions. “Tsunami” sessions address larger hypotheticals that stress-test the business.
Trust habits for this phase:
- Set explicit quality bars by artifact type.
- No customer template goes live without legal and support sign-off.
- No SOP ships until operators confirm it works as written.
Days 61 to 90: Scale what works, kill what doesn’t
Feed prior policies and regulator guidance, then have the AI copilot produce a redlined update with traceable references. Humans settle any conflicts and finalize the draft.
- Scale rules:
- Only scale workflows hitting quality targets three weeks in a row.
- Promote prompt libraries and examples to a shared space with named owners.
- Add a light-review rotation so no single gatekeeper becomes a bottleneck.
- Publicly thank people who flag errors, improve prompts, or catch risks early. Recognition helps reinforce behaviors that align speed and integrity.
What your scorecard should show by day 90
If all goes well, you should notice the following:
- Time saved on routine knowledge work.
- Faster handoffs and fewer status meetings.
- Lower edit rates on templated outputs.
- Clearer boundaries for when humans must decide.
- Higher perceived trust because rules and reviews are visible.
If results are flat, change the job mix.
Patterns across studies are consistent. Copilots help some roles and tasks far more than others, and the gains are largest where work is standardized and context is well captured. Treat the entire experience like an experiment, not a religion.
A short script you can borrow
- Share this with your team on day one:
- Copilots draft. Humans decide.
- No personal or regulated data in prompts.
- Every external message gets a human review.
- We measure speed, quality, and trust. Any workflow that fails one of the three stops.
Maybe you felt behind when you started reading this piece. But you now have a plan that respects trust and earns quick wins. It’s a good place to start.
Chris Dyer is a leadership and keynote speaker.