Source: raw/How_to_Go_from_AI_Power_User_to_AI_Leader.md — Creator: Nufar Gaspar (presenter; captions render the name “Newfar”/“Nofar”), hosted by Nathaniel Whittemore (NLW) on The AI Daily Brief · URL: https://www.youtube.com/watch?v=wdXF-5nn-T4 · Platform: YouTube (yt-podcast; fetched 2026-10-21). The episode ends with a plug for the hosts’ paid AI-champions bootcamp, so the framework is also a course outline.
NLW opens by calling a well-structured internal champions program “the single highest ROI transformation activity” he has seen in enterprises. Gaspar then gives a practical playbook in two halves: what the company has to set up (role definition, nomination, training, and the support that keeps a network alive after launch), and what a champion should do with each use case, from deciding whether to build at all to writing the spec. It is opinion from someone who runs these programs, not research, but most of it can be turned straight into a checklist.
Key Takeaways
- A champion is “peer plus one”. “They sit within the team, they do the team’s work, but they’re a step ahead in AI.”
- Pick a role on a spectrum, and write it down. Ambassadors are the best users; they drive adoption, showcase use cases, train peers and voice what the team needs. Ambassador-builders build for themselves and their team within a defined scope. Internal FDEs (forward-deployed engineers) are “a professional builder who sits inside a business team and builds what the team needs”, from idea to production. Goals and expectations should be “agreed with everybody else who is involved like IT security”.
- Nominate on evidence. Good signals: they already use AI without being told, “their peers already come to them”, and their manager gives them time. Mistakes: senior self-nominators (they belong on an “AI committee” that makes strategic decisions), the busiest people, nomination “as a reward”, and new hires (“never nominate the person who just started working yesterday”).
- One test for training. “Every champion takes one real use case from their team and shows how they were able to build or promote it by the end of whatever training that you do. If they can do that, they’re ready.”
- Programs fail in “the continuation”. After launch, “nobody took ownership on making sure that the champions are being used effectively and learning continuously.”
- Two sponsors. A community sponsor from HR or the people team keeps the human side alive; a professional sponsor close to the chief AI officer sets priorities and clears approvals, access and tools. Her best example: “a dynamic duo of an L&D expert and an AI expert.”
- Brief the champions’ managers. Great champions stall when their managers “don’t have realistic expectations or don’t know how to use them”.
- Default to the simplest solution. A shared chat setup (prompt plus files) → a skill → an automation → a full agent, moving right only when there is a reason to.
The company side
- Define the role and the program’s purpose. One company decided it mainly wanted champions as “a good funnel that brings the relevant use cases from the ground to the central team”; others later changed their minds and wanted champions to build.
- Nominate using the signals and avoiding the mistakes above.
- Train, internally or as a concentrated bootcamp, ending with the one-real-use-case test.
- Run the day after. Name the two sponsors, brief managers, and make it easy for people to leave and to join.
- Revive a stalled network with the same checklist: name two sponsors, “check that every champion still have the one or two days or sufficient capacity to do the work”, recommit with their managers, and let people who can’t do the work step out.
The champion side: idea to production
- Identify the use case conservatively. “Assume that the investment will be higher than you anticipated and … that the value will not necessarily be as high in day one.”
- Decide deployment early: where it lives, who approves it, “and who supports it in day two, not just day one where everybody’s excited.”
- Evaluate before release. A champion must be able to sign off on whether it is “good enough to put in the hands of users”.
- Should we build at all? Check how often it comes up, what it is worth and how costly a mistake is; then choose between reusing something, buying, building, or adding it to the backlog. “There is such a huge bias towards building.”
- Same use case, three roles. For a team that spends hours each week assembling a customer-renewal summary, the ambassador describes it and hands a clean request to a central team or vendor, then makes sure the team adopts the result; the ambassador-builder builds a version for their own team that another team may productize; the internal FDE takes it all the way.
Spec checklist before building:
- Who it is for: “designing for everybody is designing for nobody.”
- What it needs to know (often a separate knowledge-curation project).
- How people will use it.
- “What it must never ever do”, i.e. concrete guardrails; starting here “creates much more important clarity”.
- Who checks the output.
- How you will know it works: write test cases early, which also “helps kill it before it’s built”.
Guardrails for non-technical builders. NLW raises that agentic coding tools are “powerful and tenacious enough that they tend to escape containment pretty quickly”. Gaspar’s answer is not to shut non-professional builders down but to “do the enablement and the infrastructure”.
Try It
- Write a one-page charter naming the role (ambassador, ambassador-builder or internal FDE), the time allowance, and what IT security has agreed to.
- Run the nomination list through the four mistakes (senior self-nominators, the busiest, rewards, new hires) before announcing it.
- Make the graduation test one real use case per champion, shown at the end of training.
- Put the six-question spec checklist into your intake form for any internal AI build request.
- Diary a quarterly health check: two named sponsors, real capacity per champion, managers briefed.
- Agency angle: the same structure fits a client rollout, with one champion per client team acting as the “funnel” for use cases.
Open Questions
- No outcome data is given for any program; “highest ROI” is NLW’s judgment from his own client work.
- The episode doubles as promotion for the hosts’ paid bootcamp.
Related
- Loops for Knowledge Workers (Nufar Gaspar) — Gaspar’s earlier webinar on goal cards and verification
- AI Rollout Resistance — Three Principles — the job-security side of adoption
- WalkMe State of Digital Adoption 2026 — adoption data behind the champion case
- AI Transformation Case Studies — company rollouts in practice
- Microsoft Autopilot — Five Habits — Nate B Jones on the 5% who get value, and how they share fixes
- Lenny’s Summit 2026 — Atlassian’s AI builder weeks and fluency index