Your AI champions are the key to engineering adoption

Estimated reading time: 7 minutes

Your AI power users, not mandates, are the skeleton key to broader and more effective AI adoption. Here’s how to unleash them.

For software engineers, the rise of AI-assisted coding represents much more than the adoption of a new tool; it’s a fundamental shift in how they work. While many engineers have embraced this shift, others are  justifiably skeptical.

When executives hand down strict adoption mandates for developers or push to achieve rigid productivity metrics, it only seems to build resentment. 

Engineers who say their teams took a more grassroots approach, experimenting together and sharing their learnings as peers, report a more positive experience. This culture can also give rise to early adopters who are eager to uncover real impact and champion the role AI can play in engineering. When these AI champions showcase results and practical tips they discovered, it drives organic interest among their fellow engineers in a way that actually resonates. 

“We knew that you have to make people want it,” said Tyler McConnell, staff software engineer at equity management software company Carta. “You can’t tell them a better way; you need to show them. And the person showing them needs to be an existing trusted peer.” 

The AI flywheel 

When Neel Sundaresan, IBM’s general manager of automation and AI, set out to lead the build of the firm’s own agentic coding tool, he knew the first step was to make sure he was creating something IBM’s tens of thousands of developers would want to use themselves. The second step was to get them excited about it.

It turned out that 40% of features in the IDE, called Bob, could be built by Bob, he said, which meant the company’s engineers were using the product to build the product. This provided an immediate confidence boost, and word started spreading. Soon he was receiving requests from engineers across the organization who wanted to try it.

For the larger rollout, Sundaresan strategically started with the teams he felt were the most “AI ready.” These teams already had good testing discipline, and he hoped, would generate some early champions who could show measurable impact quickly and teach others how to work with the agent. He said one team cut a four‑week data lake task – building a more robust probability engine for the company’s data security platform Guardium – to a single day. Another reduced a 30‑day FedRAMP compliance mapping exercise to two days.

“Those aren’t vanity metrics. They’re the kind of outcomes that turn skeptics into vocal champions,” said Sundaresan.

Just as he’d hoped, adoption spread organically, with several engineers becoming “very vocal internal advocates.” Overall, Sundaresan believes training for using new AI tools effectively and responsibly has to be peer‑to‑peer, citing how enthusiasm can be contagious. 

“Many users report that, true to its design, using Bob can be similar to pair programming alongside a mentor or peer,” he said. “That kind of authentic testimony travels faster inside an engineering culture than any slide deck I could make.”

Show, don’t tell

At Carta, power users emerged early, sharing their learnings, and creating a cycle of sharing that brought more people into the mix.

In some cases, people organically became promoters and mentors. The company also actively sought out power users, using its dashboard to determine who was using the most tokens – and thus who was experimenting the most – to learn from what they were finding and give them opportunities to share.

As a member of Carta’s team focused on operationalizing AI, McConnell sees being an evangelist for the technology as a core part of his role. He’s helped organize vibe coding sessions, presented on AI best practices, and showcased how to tackle specific tasks with AI. 

He believes real-world demonstrations are what can best convince colleagues, because watching a peer accomplish tasks similar to your own is directly transferable. For example, in a presentation focused on spec-driven development, he demonstrated how providing the AI with a rigorous technical specification upfront dramatically improves the output. 

“It was a perfect example of ‘show, don’t tell,’” he said. “I watched the room shift from passive interest to excitement because I wasn’t just showing them a theoretical playbook or a magic trick. I was giving them a repeatable framework to get work done.”

The company also has weekly office hours and targeted sessions where engineers in the same domain do a combination of “show-and-tell” and pair programming. They also have dedicated Slack channels for specific tools like Claude Code and Windsurf, plus broader channels for AI coding news. 

Recently, Carta’s engineering organization ran a refactoring hackathon where the explicit goal was to utilize AI coding tools to clear technical debt. McConnell sees it as a prime example of how specific challenges can familiarize a team with new paradigms much faster than documentation ever could. 

“By creating a space for communal experimentation, we generated immediate buy-in and excitement for the new way of working,” he said. 

The power of demos

Marina Wyss, an engineer at Twitch who has championed AI coding within the company, has also seen the power of hands-on demos. 

She said her role on her team has often been to learn about new tools and techniques, identify opportunities, and share them. In her spare time, she also creates educational content for developers on YouTube. So experimenting with AI tools to find which would be best for her team was a natural move for her. 

Overall, she thought about her team’s specific needs and how to balance usefulness with complexity. She also wrote docs, held training sessions, and was considerate of the AI anxiety some of her colleagues were feeling, being sure to address both what it can and cannot do. 

In one training with the data scientists on her team, for example, she demonstrated a complete workflow including ideation, queries, data analysis, model building, and writing a final report. 

“I think demoing the process and how each step can be made simpler with the use of AI assistance made it less intimidating,” said Wyss.

This pattern of communal learning and AI champions driving strategic adoption plays out even at companies building AI coding products. At coding agent company Kilo, using AI to code is a given. Still, engineer Brian Carlson believes collective collaboration and constant communication has been the key to driving progress – and keeping up in this fast-paced domain. 

He said the engineering organization is constantly experimenting with different models, ways to prompt, discussing which models work best for specific tasks, and sharing outcomes. He also shows up with this mindset outside of Kilo, mentoring a very cost-conscious non-profit engineering team on how to employ AI, as well as junior engineers he knows who are still wary of leaning too hard into AI assisted coding. 

“Things are changing so fast, and there’s no way to learn it all on your own,” he said. “Sharing the learnings means you get to learn what you learn, plus what your coworkers learn.”

AI champions > AI mandates

To get the ball rolling on AI coding, some companies have taken a different approach: handing down mandates to use the tools, sometimes tracking AI usage without any regard for impact and what’s actually helping. 

Sundaresan said IBM didn’t mandate engineers use Bob because he doesn’t think mandates are the right lever for this kind of change. “If you force usage, you risk shallow adoption,” he said.

Engineers seem to agree. Carlson, for example, said that learning and experimenting with AI coding yourself offers a sense of ownership, creativity, and wonder that’s often lost when something is mandated. Engineers also naturally seek to emulate the techniques of their peers, suggested McConnell. Wyss said she thinks it’s fine for higher level folks to encourage the exploration of and use of AI coding tools, but she wouldn’t suggest mandates.

“Many of us are drawn to tech because we like autonomy in how we work, and being told what to do can backfire if it increases resistance and decreases curiosity,” she said. 

This doesn’t mean higher ups don’t play an important role. They can set the stage for supporting the type of collaborative experimentation and mentoring that engineers thrive with. At Carta, McConnell said the company’s CEO and CTO were the catalyst, establishing the very first vibe coding session and giving people permission and space to experiment.

“Without that top-down support, engineers are too entrenched in their day-to-day trying to complete work and move on to the next thing,” he said.

Sundaresan believes leaders have a big role in encouraging AI champions. He found that doing so successfully requires giving engineers a space to share their experiences, allowing room for experimentation, and spotlighting the wins.

“Keep it structured enough to spread good habits, but loose enough to let teams discover their own uniquely perfect use cases,” he said. “We want our engineers to enjoy using these tools – not only to be productive, but also to feel more capable than they did before.”

Get articles like this in your inbox

Choose your LeadDev newsletters to subscribe to.