Takeaways

If you sit on the executive committee

• Put a light skeleton of governance in place before the first project: what is off limits, what needs care, what is safe to start with, and where your data goes.
• A GenAI project commits your people and your accountability, not just a budget. Approve it only if the business is ready to own it, and only if it is meant to go live.
• Choose a first project where a wrong output costs little and is caught by a person, and that leaves something useful even if it stops.
• Fund it in stages. Each stage must return something useful and end with a decision: move on, adjust, or stop.
• The acceptable level of error, and what the system does when it doesn't know, are your decisions. Don't let them be made by default.
• Ask what lies under every demo, and put the cost of doing nothing on the table. Knowledge that leaves with retiring experts doesn't come back.

If you are the CTO or CIO

• Your role is not to decide whether the answers are good enough. It is to make sure someone with the authority does, and that the decision is recorded.
• Insist on a business product owner, named content owners and protected expert time before the first cycle. Without them, the project isn't ready.
• Make the test set and the Definition of Done the contract between your teams and the business.
• CTO: build for production from the Walk stage. A pilot built on throwaway foundations turns success into a second project.
• CIO: prepare the hosting, contract and sovereignty decisions rather than accepting the defaults, and treat the knowledge base, access rights and test set as assets that will outlive the first use case.
• Watch your teams. Engineers who no longer define success still need to know why their work matters.