Jan Pivovarnik

Jan PivovarnikApr 17, 2026

When a company says "AI layoffs," look at its size. Two layoffs lately, same AI-era vocabulary in the press - and they should be read completely differently. Oracle cut around 30,000 people at the end of March. Productboard - Czech-born, now global - let go roughly 30% yesterday. Both will get filed under the same 2026 narrative. They shouldn't be. At Productboard's scale, an AI-driven cut is actually believable. I don't know what's going on inside - nobody outside does. But when a ~600-person SaaS company looks at what Claude Code, and modern AI tooling do to engineering throughput, at what an AI SDR stack does to outbound, at what agentic support does to tier-one tickets - the math can genuinely close. Small teams get disproportionate leverage from these tools. If a product org of 150 can ship what 220 used to ship, that's not a story you need to manufacture. It's just what's happening on the ground. Oracle is a different animal. You don't fire 30,000 people because your AI tools got better. You fire 30,000 people because you increased headcount by 32,000 between FY2021 and FY2023 (Cerner alone brought 26k), took on $58B in new debt to build data centers, and now need to free up $8-10B in cash flow to service it. The "AI pivot" is what you tell Wall Street so the stock goes up 6% on the day you send 6am termination emails to 26-year veterans. It's what you say while you also file 3,100+ H-1B petitions in parallel. At that scale, "AI efficiency" isn't a cause - it's cover. The tell is this: at a small company, AI changes the unit economics of the team. At a giant, AI changes the narrative of the balance sheet. One is an operational reality. The other is investor relations. Worth remembering when the next headline hits.

Jan PivovarnikJun 16, 2026

Spent last week in London mentoring robotics and embodied AI founders at the Google DeepMind Accelerator. Easily one of the most interesting rooms I've been in this year. Here's the thing though - I come in from the other side of the table. I don't build robots. I run AI inside a large enterprise, which means I spend most of my time on the question founders tend to postpone: not "can we build it" but "will anyone actually absorb it." The teams were genuinely excellent. E-skin for humanoids, spatial intelligence for warehouses, robots testing digital interfaces and doing brain surgery (!!!). The technical bar was high enough that the 1-1s really pushed me out of my comfort zone few times. But the pattern I kept pointing at was the same one I watch play out in enterprise AI every week: the model is almost never the hard part. The hard part is trust. The workflow it has to slot into. The person on the buyer's side who has to put their name on "yes, deploy this in production." That is change management, and it is where most genuinely good technology quietly dies. So a lot of my advice had nothing to do with robotics or AI. Design for the adoption curve as deliberately as you design the hardware. Know exactly who has to say yes, what they're afraid of, and what evidence actually moves them. Then build those receipts before you need them, not after a buyer asks. The founders who internalise this don't win because they have fewer obstacles. They win because they see the real one early - and it's rarely the one in the demo. None of this happens by accident. A room that good is the result of months of quiet filtering, and the Google for Startups team carried it - Noa Havazelet, Onajite Emerhor, Naomi Finkelstein and Mateusz Szymański, thank you for building the space and letting us do our thing in it. And the mentor bench was half the fun as always. Thinking out loud alongside Daniel Gwerzman, Yorai Gabriel, Irina Sergeeva, Ofer Golan, Katharina Arnold Rokita, Julie Robles, Adi Mazor Kario and others made the hard conversations a lot easier. Brilliant people, brilliant few days. To the founders in the room: keep going. The real obstacle is usually closer than the demo suggests. #AcceleratedWithGoogle

Jan PivovarnikJun 30, 2026

You learn the most by teaching. Earlier this month I co-taught a module on AI and strategy in the iLead Institute Česko Be Future Ready Executive programme - two days with a room of CEOs and senior execs. When you work with something every day, you're allowed to stay fuzzy. You carry intuitions you never have to defend. But you can't teach a hunch. Stand in front of that room - no laptops, no slides to hide behind - and every loose idea has to become a sentence you can say out loud and fully own. Co-teaching with Pavlína Louženská and having Laureen Höllge as guest-speaker raised the bar even further. Three of them only got sharp because I had to teach them. 1) "AI doesn't take jobs. It takes tasks." Every role is a bundle of tasks. Most strategy conversations jump straight to "will this replace people?" - when the real work is unbundling the role first, then deciding task by task. 2) "AI can't do this yet" has an expiry date. What a model can handle on its own has been roughly doubling every few months. So any strategy built on today's ceiling is quietly answering last year's question. 3) And the one that landed hardest: most of the value isn't in the model. It sits in people and process - call it 70% of it. Yet that's where almost none of the money goes. The distance between where the value is and where the budget goes is the whole game. The future part was the same. You don't plan it by predicting which AI arrives when. You plan it by asking what won't change, and how ready your people are - because half of that you actually control. The moment that stayed with me came from one of the execs. His biggest takeaway, he said, was that he could finally wrap his head around what AI means for him and his company - and that maybe they don't have to be AI-native, but they definitely want to be AI-friendly. That's a whole reframe in one sentence. #ILeadInstitute #BeFutureReady #MIT #Forbes #AI