Talks for leaders who have to make AI work, not just approve it.

Directing agents is a skill: scoping work, giving context, reviewing output, knowing when to take the wheel. At G2 I took an engineering organization from zero AI usage to 92% active adoption in seven months, and the biggest lever was not a mandate or a tool. It was coaching agent management the way we coach code review. I'll show the rollout, the budget guardrails and the mistakes, and leave you with a first-90-days plan you can run on Monday.

Ad-hoc prompting does not scale to a team. A delivery harness does: phase-gated skills from triage to close, specialist subagents, an isolated git worktree for every ticket, and repo conventions that agents and engineers both read. The rules matter more than the models: complexity gates decide when a person approves the plan, every pull request gets a human review against a checklist, and agents never merge. You'll leave with the operating rules, written down, that you can take to your security team.

Lines of AI-written code is a vanity metric, and ranking people on usage destroys trust. A better question is whether the system gives the business enough at-bats: shipped experiments, each with a hypothesis, a metric and a result. I'll show how to pair at-bats with DORA and DX Core 4, keep usage data team-level and visible to the engineer first, and celebrate failed hypotheses so they get reported honestly.

AI spend arrives without a budget owner and grows faster than anyone forecasts. Route every call through a gateway, give each team a budget, send routine work to cheaper models, and forecast to fiscal year-end with a range. I'll show how spend moved from zero to broad adoption with visibility per team, and why cheaper defaults beat after-the-fact policing. Includes a walkthrough of a live spend report.

At Gartner Digital Markets we replaced a dedicated QA function with developer-owned, AI-assisted test automation and training. Developer efficiency rose 20% and bugs fell 12%. The tools were the easy part. I'll cover how to make the case, how to retrain rather than discard expertise, how to talk about role changes honestly, and what you should instrument before you start.

I led a 25-person, three-region organization to 13 while keeping delivery on track, then separated our platform from its former parent through the acquisition by G2. When agents absorb throughput, teams get smaller and each person's scope gets wider. I'll share what I'd do the same and differently: saying it plainly and early, making decisions once, protecting the people who stay, and the team shapes that work with agents (human pods, agent squads, a platform team).