Hadar BismutLately, the world's biggest AI companies have started building models for embedding AI inside organizations, in all kinds of formats: Forward Deployed Engineers, champions, and more.
Anthropic's offering for building a champions network in enterprise organizations made its way to me. As someone who builds champions communities herself, I wanted to see what they offer and how it compares with the programs I run. So here it is: a review of the offering, my takeaways, and more.
General overview
Anthropic breaks the process into 4 stages:
- Stage 1: Nominate and select
- Stage 2: Enablement, which splits in two. Train the trainer: how to teach, how to work with Claude, and so on. Champions: a workshop for identifying use cases.
- Stage 3: Launch
- Stage 4: Scale and sustain, adding more champions and supporting the champions who are rolling out use cases.
One important point: they call it a Network, and they avoid the word community. I'll come back to that.
What is a champion's role?
It's a big question, and the answer changes from organization to organization depending on its needs. The bottom line: champions are the people who learn the technology, figure out how to work with it, and take it back to their teams to implement real solutions.
Anthropic distinguishes between two roles:
- The people who teach, the train the trainer role. They sit in specific parts of the organization, usually L&D or an AI COE. Anthropic recommends about 10 to 20 of them per organization.
- The champions themselves, spread across every unit and team. Anthropic recommends one champion per 100 employees (meaning users who work with AI).
They mark this distinction as very significant, and as something that can cause the program to fail if you skip it.
I want to add another distinction, specific to champions. Over the past year I've helped organizations build communities like these, both in R&D and across the business units (cross company), and I've seen 2 different types here too:
- Bottom-Up champions, who can also be ICs. They're expected to embed AI in the different product units, with a specific focus on their own team.
- Top-Down champions, meaning VPs, directors and similar roles. They're expected to build cross-functional solutions to significant challenges that affect the entire domain, and to actually implement them.
When you decide to build a group of champions (or simply to train some), it's essential to understand the bigger goal: who you need to lead the change, and at what scale (team or group/domain). There's no way to skip this step, and everything else will be shaped by this decision.
How do you recruit champions?
Anthropic suggests two ways to recruit: ask managers for recommendations, or find the most active users in your AI usage data. They stress choosing people who love teaching others, and preferring them over seniority.
All true. Here are a few more things I put weight on:
- Commitment: what capacity do we expect from a champion, in time and attention?
- Motivation: are they ready and willing to take part in the mission of spreading the knowledge?
- Capabilities: what skills does the champion have? Are they a good communicator, sociable, well positioned in the organization, well connected? Do they have the ability to lead?
- Do we have good coverage across all the business and product units where we want to embed AI?
- In what sense do we want to embed AI: to drive innovation in the product itself, or internally, in everyday work?
The answers to these questions help us focus our programs and set the right recruiting targets.
One more thing worth understanding: a champion leads AI adoption and, beyond that, acts as a real change agent, so this is change management. Anyone who has ever been involved in a project like this knows it's a much bigger, more emotional and more complex challenge than embedding AI in a use case. Remember that behind every AI rollout are people who worry AI will replace them, and champions carry that too, both as employees themselves and as the people driving it forward in the organization.
Community or Network? The difference that decides everything
Anthropic prefers the word Network over Community (community only appears where they mention opening a channel for champions). The word choice matters. A network connects people with people and information with information. It can function even when the people in it don't know each other well, don't feel they belong to a group, and aren't committed to the mission beyond the role they were assigned. A community is built on shared identity. People in a community perform a role and are also part of something bigger than themselves.
And when the hard moment comes, when the boss applies pressure, when the sprint is packed and there's no time, when the energy of the launch has faded, that's when a network empties out and a community holds. The sense of belonging is stronger than everything else, and the commitment is real.
This difference is exactly the missing piece in a question Anthropic doesn't really answer: what makes champions stay? Equipping them with materials and access is one part. The other part is making them want to be there even after the buzz fades.
For me the answer is clear: identity. When a champion feels part of a group that shares a language, a perspective and a mission, they don't drop out because "I don't have time this week." They drop out only when there's really no choice (and that happens too, of course).
How do you sustain the activity over time?
Anthropic suggests focusing on 3 core formats:
- Sharing use cases, blockers and updates in a Monthly sync.
- Presenting demos at a Show & tell.
- Dedicated time to ask questions, get advice and get help with technical issues in Office Hours.
In my experience these are good anchors, but in reality not all of them hold up over time, and you need to be creative and respond to engagement as you go.
For example, Office Hours are nice, but people want immediate solutions and won't wait for specific hours in the week. They'd rather find the most active channel and ask there, or message whoever handles the topic directly because they know them personally.
A monthly update is too little given everything happening in AI, especially in organizations embedding it at scale. If we want champions to be the most up to date, the most connected and the most advanced, they need to sync much more often.
Show & tell sessions are incredibly important, and I think they deserve a more prominent place of honor. They need to be treated as strategic, meaningful events that involve key stakeholders, organizational visibility and more.
Beyond that, nowhere in the document did I find a reference to what really matters for sustaining the activity: measuring engagement. With the endless workload people in organizations are carrying these days, we need to build personal connection and commitment to the mission at the highest level possible. To make that happen, we need to create human connections between the champions, send the message that it's OK to ask anything, keep connecting them with leaders in the organization who give them the recognition and legitimacy to invest in the program, and give them first access to new releases, features and internal tools. Make them the organization's go-to for everything AI.
Remember that people join programs like these mainly for their own personal gain. Even if it ultimately shows up in their teams and in the mission they commit to, we need to keep making sure they want it and understand why it matters to them.
Recognition
Here Anthropic offers something that I, or anyone other than them, can't offer: direct access to the Anthropic team and early access to new features. Drop the mic. By the way, it's unclear whether this is only for the core team or for all the champions, and whether the people available to champions are technical staff or CX, but let's set that aside.
Beyond that, they suggest focusing on things that are very standard in the internal world of organizations: swag, a community channel, visibility with leadership, a stage to present use cases. All of these are good, and you need to take them to the extreme for this to really succeed. For example:
- Give champions a place of honor at all hands, beyond the community's own sessions
- Let champions lead learning sessions at company days or hackathons
- Bring leadership to open sessions where champions can ask anything
- Look for external opportunities, such as conferences and big events, where champions can present
- Build platforms to showcase their work to the whole company: internal platforms, a podcast, a newsletter, a dedicated channel and more
And that's just the tip of the iceberg. The more significant it is, the stronger the feeling of "we're being seen."
How do you measure success?
Anthropic defines 4 metrics:
- AI usage in the champions' teams: Adoption
- The number of use cases documented and shared
- The number of sessions champions ran themselves: Enablement
- How users feel about Claude in their workflows: Sentiment
First, about sentiment: I'm really not sure we care that specifically about Claude (I understand why Anthropic does). What matters in programs like these is whether our confidence in using AI is growing, and whether we feel the organization gives us what we need to succeed.
Beyond that, I think these targets are too soft and too general. They're good and they point in the right direction, but I find that the best way to get things moving and make them routine is to set clear targets for each champion individually.
For example, in one program I built, we decided that every champion had to complete a use case in each community cycle, which we set at 6 weeks. It's specific, focused, and long enough to allow success. In stages, rolling over between cycles, or 2 or 3 use cases in that time: all good, as long as there's a clear time frame.
In another program I built, we defined which deliverables had to be shared within the teams, with how many people, and over what period. Everyone knew exactly what was expected of them and got all the support they needed to make it happen.
Focusing on specific targets, set per person and beyond the group as a whole, helps everyone create momentum, and helps whoever leads the community measure engagement and make sure everyone is in. For me, champions need to produce deliverables and work in the field, and if they don't, there's no reason for them to be part of the community. Specific measurement lets us see this and decide how to act. When the program is strong enough and has a specific win-win, it's easier to know whether it's working and to recruit the right people.
Where can this fail?
Anthropic highlights a few specific areas:
- At Launch: before this stage, you must make sure everyone has access to the tools, connectors and so on.
- Combining the train the trainer role with the champion role. It's too much for one person.
- No ongoing recognition tied to a specific timeframe.
- No single place where all the shared materials are collected.
- No meaningful sponsorship at the leadership level.
I'd add to the list difficulty building connection between the champions, too wide a professional and technical gap between people in the community, and major organizational changes that reshuffle the deck. True, not everything is in our control, but these are the kinds of things you need to know will change the picture.
Not covered in the document at all, but can also affect success
1. Where does the program sit in the organization?
If it sits under the VP R&D, the CTO or the CEO Office, that's one thing. If it sits under L&D, IT or HR, that's another. In the end, all the pieces have to connect, and substantial organizational sponsorship is critical to success. Partnerships between different functions are great, just make sure the program has enough teeth at the organizational level.
2. Who actually leads the program?
The program lead needs to see the full, strategic picture, connect the program to every part of the organization (which means knowing the organization deeply), and have very strong people skills. Ideally, they also understand AI themselves and can guide champions on their use cases so they really get implemented. If that's not possible, they need to be able to recruit people around them for support, even when it isn't their job (which also requires the lead to have strong standing in the organization to get them on board).
3. How do you keep the train the trainer role up to date and relevant?
The trainer's role is critical for people skills and equally for the ability to handle technical complexity that keeps changing and is shaped by many elements: FinOps, Governance, Guardrails, Context layer, Legal and more. If the trainer doesn't cover all of these, we need to make sure someone else provides support when needed, especially when implementing use cases full of infrastructure and organizational challenges.
To sum up
There's no doubt Anthropic has the name, the positioning and the capabilities to set up a champions and train the trainer program in your organization that will get a lot of buzz and attract the right people.
What's missing from their offering is attention to what happens to the champion during this process. Nothing about growth, identity, or the sense of belonging to a group that shares a mission. They're building a network, and what makes programs like these last over time is a community. These are two completely different things.
Remember that in the end you're building a program that should last and go far beyond the buzz. And like everything in AI: today it's Anthropic, tomorrow it's Codex, and who knows what comes after that.
What remains is people (and agents, but that's for another article).