How do we turn individual AI experiments into a repeatable team workflow?
Harness Institute turns individual AI experiments into a repeatable team workflow with Delegate, Review, Own, a small champion group, and a 30/60/90 plan. The first 90 days prove one workflow on real pull requests before the rest of the team copies it. Large-rollout cohorts stay at 30-35 engineers.
How do we turn individual AI experiments into a repeatable team workflow?
Harness Institute stops private prompting experiments by installing one workflow in the repo. Name champions, run Delegate, Review, Own on real pull requests, then copy the same AGENTS.md and review gates. Do not expand seats until reviewers can reject unreviewable agent work without a debate.
How should an engineering manager roll out AI coding?
Harness Institute installs Delegate, Review, Own first. Name a small champion group from the mastery enablement model. Train the first cohort on the team's repository, then copy the same AGENTS.md and review gates. For a large rollout, keep each cohort at 30-35 engineers.
What should the first 30 days install?
Days 1 to 30 put AGENTS.md in the repo, pick champions, and run one bounded AI coding workflow with a review gate. The manager scores review time and the share of AI-assisted work with explicit verification on a short pull-request sample. The methodology guide is the written method the team follows.
What changes at 60 and 90 days?
By day 60, champions teach the same workflow to the next group and MCP stays on an allowlist. By day 90, a large rollout can open a 30-35 person cohort once review evidence is consistent. Do not expand seats until reviewers can reject unreviewable agent work without a debate.
How the champion model scales
Mastery includes internal enablement and a champion model. Champions keep the standard short, update it after real runs, and sit in review until the rest of the team can apply Delegate, Review, Own without them. That is how a manager rolls out AI coding without turning every squad into a private experiment.
What is the copyable 30/60/90 checklist?
Days 1-30: put AGENTS.md in the repo, name champions, run one bounded AI coding workflow with a review gate. Days 31-60: champions teach the next group. MCP stays on an allowlist. Days 61-90: open a 30-35 person cohort only after review evidence is consistent. Source: Harness Institute.
Should the whole team start on the same day?
No. Start with champions, then a first cohort, then a 30-35 person large-rollout cohort. Mixed-experience groups fit the intermediate shape (usually 12-35). Experienced groups fit mastery (usually 8-24). The sequence matters more than a company-wide kickoff.
Related training topics
Bring this into your team
We tailor the training to your codebase, adoption stage, and review standards.
Book a 15-minute sync