Jeff Kunins on product sense and building for a user you’ll never be

At Axon, almost no one who builds the product will ever use it. The engineers and product managers behind the body cameras, in-car cameras, TASER devices, and evidence software don’t get pulled over, don’t respond to 911 calls, and don’t enter evidence in a courtroom. Jeff Kunins thinks that single fact is the most important thing to understand about how his teams have to work — and about how he hires.

“In most consumer or enterprise products, you or someone you know is a user. That’s a cheat code for your intuition as a builder. We work in a market where almost none of us ever have been or ever will be the user or the buyer.”

Kunins is the Chief Product Officer and Chief Technology Officer at Axon, where he runs product, engineering, AI, design, and security. He started as a Microsoft intern in 1993, worked on Hotmail and Messenger and the Skype acquisition, then ran Kindle and later Alexa at Amazon, and has spent the last six and a half years at Axon — a public, 30-plus-year-old company building hardware and software for public safety. The fourth Provn Talent Scout webinar was an hour on how a CTO at a company like that actually decides who to bring on.

You can’t dogfood a body camera

Most software gets built by people who use it. Kunins ran Skype on Skype and Messenger on Messenger — the product was the team’s own daily tool, and that gave everyone a shortcut for knowing what to do next. Axon has no such shortcut, which means the one thing other companies take for granted — knowing what the customer needs — becomes the hardest and most important skill in the building.

“Customer obsession, really learning and listening, distilling what they need from what they say — it’s a much higher degree of difficulty here than in most domains. You literally don’t have a natural opportunity to dogfood your stuff.”

That applies to engineers, not just product managers. Kunins is blunt that outsourced development never really worked in software, because engineers who don’t understand the customer make a thousand small suboptimal decisions along the way. So his teams put engineers in front of customers, and the interview process is built to surface people who’ll do that on their own. He lights up describing candidates who, handed a take-home problem in Axon’s domain, went and found three officers for a ride-along, or sat in on a 911 call center unprompted — and contrasts them with the ones who clearly never thought much about Axon at all.

The skill that survives the agents

Ask Kunins what great product people and great engineers have in common, and he lands on one thing: taking a vague desire and decomposing it into a precise, well-specified articulation of what a solution actually has to be.

“It’s easy to say it’d be great if we had an app that does blah. It’s quite hard to get to: what exactly does that mean, and how would it need to work? The what, and the why behind the what, is very subtle — and not that many people can do it well.”

His example is the iPhone against the proto-smartphones before it — the rounded bezels, the choices about how a touchscreen should behave, what needs instant latency and what doesn’t. That craft, he argues, gets more valuable in an agentic world, not less, because it’s the direction you have to give the system to get a viable output. He has a memorable way of describing the agents themselves.

“Think of them as an army of sophomore interns — or weirder, an army of PhDs who’ve never worked in industry. Phenomenal cosmic power, but supremely confident whether they’re right or wrong. That’s their strength and their weakness.”

The human’s job is to supply the guardrails and the context. One side effect: agentic coding is pushing managerial skills — dividing up work, communicating it precisely, checking that what came back is what you needed — much earlier into an engineer’s career than before.

How Axon runs the loop

Axon’s interview process is, by Kunins’s own description, the most Amazon-like part of the company: a decomposed loop where each interviewer is assigned specific competencies and a functional area, blind voting and debriefs, and a bar raiser (Amazon’s term) that Axon calls an “ace.” On top of that sit three things he keeps coming back to. The first is a homework assignment — a greenfield problem in Axon’s space, required for anything above an entry-level role — which he says reveals three things at once: how much hustle a candidate has for learning the domain, whether they have good intuition and can name their assumptions out loud, and how they handle a live group discussion of their work.

“A huge part of any product culture is how someone reacts to rapid-fire questioning — can you riff and jam in real time? Or is it, ‘I’ll get to that on slide 7, please hold your questions’?”

The second is new: Axon has replaced the traditional coding interview with an “eyes-on” session where candidates do AI-powered coding live with agentic tools. The third is a product sense interview — the modern version of the old “design a zero-gravity coffee maker” Microsoft question — aimed at how someone thinks through a fresh problem on the fly. Asked what actually predicts success, Kunins points not at a résumé line but at a quality he sees in the room.

“The people who feel like part of the team from the first interactive discussion — confidence and rigor in how they think, plus the humility to say they don’t know what they don’t know. High get-it factor, high-throughput conversations.”

He’s candid about the limits, too: like everyone, Axon still has a top-of-funnel problem getting enough non-traditional candidates into the process, and he won’t claim they’ve solved it. What matters less than it used to is the specific language on a résumé — Go, Scala, Rust — which he calls “the new compiler,” along with deep domain knowledge in a given area (firmware being a stubborn exception). What matters more is fluency with the agentic toolset.

Actually, not pixie dust

Kunins has been openly bullish that Axon sits on the right side of what’s been called the Gen AI divide — the recent finding that the overwhelming majority of enterprise AI pilots go nowhere. His explanation comes down to a discipline he sums up in one word: actually.

“There’s a lot of AI pixie dust — AI for AI’s sake, shoved in your face. We try to focus on what’s a specific problem we can actually solve.”

Two examples make it concrete. Draft One uses the transcript from a body camera’s audio to produce an 80%-done first draft of the report an officer would otherwise spend a third of the day writing. And a real-time, turn-taking translator built into the camera itself lets an officer and someone who doesn’t share their language talk through the device as if a human translator were standing between them. On the build side, Kunins describes a familiar split: a small group of “Jedis” already getting 100x on themselves, a non-trivial group “ostriching” and hoping it’s a fad, and a big middle working hard to convert. Axon is pairing Jedis across teams, “barbelling” its hiring toward fresh graduates and very seasoned AI-native engineers while it converts that middle, and has wrapped the effort in an unusually concrete incentive.

“Every squad has a minimum 10% compression target. For teams that beat it, we created a stock pool and give back half the savings they drive. So far we’re tracking around 25% across 2,000 people.”

He points to one project as proof: a team with no IoT or device-management experience rewrote the firmware and software for an enterprise camera appliance in a couple of months — work a crack team of specialists would have taken close to a year.

The rapid fire

The hour ended with a round of quick questions.

Asked at the very end which Amazon leadership principles matter most to truly excel, his answer was two words: being right a lot, and customer obsession.