Henrik Nyberg spent years working with product teams at Spotify and Lego before co-founding Abundly, which builds autonomous AI agents. On the latest episode of Puzzel Talks, he joined Puzzel’s Anna Dunn to discuss why the biggest mistakes in AI adoption look a lot like the ones he saw during the shift to agile.
Most leaders have heard some version of the same advice on AI: move fast, or you'll be left behind. What's harder to find is someone who has actually lived through a comparable shift before and come out the other side with a clear view of what worked.
That's what makes Henrik Nyberg's perspective useful. Before Abundly, he spent years embedded in agile and lean transformations at two of the world's best-known product companies. His conclusion, discussed on the latest episode of Puzzel Talks: "at the end of the day, humans are humans, and we're not changing very much."
The technology might be different, but many of the organisational patterns that made agile transformation succeed or fail are playing out again with AI.
The most successful transformations Henrik has seen, whether agile or AI, share a common structure.
Leaders create the right conditions by providing support, access to technology and room to experiment. But the momentum has to come from the teams themselves: experimenting, learning and sharing what works.
The opposite approach can quickly become counterproductive.
Henrik has seen agile rollouts collapse when leaders tried to force everyone into a rigid template. Today, he sees the same thing happening with AI mandates: a fixed monthly token quota, a ban on writing code by hand.
"All these kinds of mandates I find don't really work. It's better to just encourage people to try stuff and find the best way to use it."
In other words, adoption shouldn’t be about hitting an AI usage target. It should be about helping people find where the technology genuinely makes their work better.
There is, however, one important difference between AI and previous transformations: risk.
AI is more powerful than many of the technologies organisations were adopting during the shift to agile, and the consequences of using it badly can be greater. But Henrik’s answer isn’t to lock it down.
Instead, he compares it to the way organisations learned to approach the internet: something with real risks, but risks that can be managed through training, sensible guardrails and clear expectations.
The goal is to make experimentation safe, not stop it altogether.
Henrik thinks of well-built AI agents as teammates - up to a point.
When an agent operates in the same Slack channel as the rest of the team, pushes back on a plan, or acts on its own initiative, it behaves like one. But it isn't human, and Henrik is clear that it never will be treated as one. His preferred description: an alien. Brilliant at some things, surprisingly limited at others, and worth learning to work with on those terms.
That framing has a practical payoff. Henrik has found that treating a misbehaving agent like a struggling new hire, giving it direct feedback rather than rewriting its code, produces better results than treating it as a pure engineering problem.
Puzzel has tested this internally. Through Abundly’s Early Builders programme, Anna built a competitive intelligence agent that now supports teams across sales, marketing and product.
"I do consider the agent that I built to be a bit of a teammate," she said. "They take a lot of initiative." That instinct, Henrik argues, tends to correlate with good agent design: when a tool starts to feel like a colleague, it's usually because it's actually delivering value.
Asked what he'd tell a leader unsure where to begin, Henrik's answer had two parts.
First, be wary of building an AI strategy around headcount reduction.
There’s an obvious cultural challenge. If employees see colleagues losing their jobs as AI adoption increases, they have very little incentive to experiment with the technology, share what works or become better at using it themselves.
But it can also mean missing the bigger opportunity.
Instead of asking how many people AI can replace, Henrik suggests thinking about how much more your existing teams could achieve with AI alongside them.
In a contact centre, that could mean helping agents find the right information faster, removing repetitive admin, summarising interactions automatically or giving people more time to focus on complex customer conversations.
Second, don't wait for a perfect strategy before starting.
"Don't worry about it. You're not gonna be able to read everything or catch up or make a perfect strategy. Just get your hands dirty."
Give teams access to good tools, encourage experimentation, and expect some early attempts to go nowhere.
The strongest use cases often emerge from that process.
People closest to the work can see where time is being wasted, where processes break down and where AI could genuinely help. Giving them room to explore those opportunities can reveal more than a strategy developed entirely from the top down
Henrik's central point isn't really about AI
It's that organisations already know, from agile and lean transformations, what makes a shift like this stick: leadership that enables rather than mandates, teams that are trusted to experiment, and enough training and guardrails to make that experimentation safe.
AI changes the scale of what's possible.
But it doesn't fundamentally change what makes adoption work.
Listen to the full conversation between Henrik Nyberg and Anna Dunn on Puzzel Talks.