# How AI Broke Your Strategy

Martin Eriksson, Just Product 2026, 09:45. Automatic transcript of the livestream (Parakeet ASR, unedited, expect misheard names). Timestamps are positions in the talk video.

[0:01] So the first speaker is Martin.
[0:04] And let me tell you a bit of a personal story about Martin.
[0:09] When I came to Munich 11 years ago, I was still a developer and I desperately wanted to transition into product.
[0:16] And then a friend of mine, Rasmus, he dragged me to a product tank.
[0:19] And I didn't know Martin at the time.
[0:21] I also didn't know that Martin actually founded Mind the Product and the product tank.
[0:28] But I have to say that this for me in my personal product journey, that was an absolute game changer because there was like a super nice community.
[0:37] And I I have to tell you, I hope the developers are not too mad at me, but developer meetups are very, very different.
[0:46] Different.
[0:47] And the product community was just so nice.
[0:50] And so if he wants it or not, Martin is one of the puzzle pieces that allowed me to transition into product.
[0:57] And I'm pretty sure that for some of you in the audience, this is the same.
[1:02] So please give him a roaring applause.
[1:05] Please welcome on stage Martin Ericsson.
[1:12] Thank you for that.
[1:17] Well, that's that's quite an introduction.
[1:19] Hard to live up to.
[1:45] And of course, the board wants to know what are we doing about AI.
[1:51] So it's not so strange that we're all freaking out a little bit.
[1:55] But every one of those headlines is a prediction.
[1:58] A roadmap is a prediction, and the board's question is a prediction.
[2:04] The problem with that is that we are terrible at predicting the future.
[2:29] That within five to ten years, AI would replace them.
[2:34] He went on to say that radiologists are like the coyote who's already run over the cliff and hasn't realized that there's no ground underneath them.
[2:44] Well, that sounds terrifying.
[2:48] And again, this is not some random pundit.
[2:51] This is not your doom scrolling headline in the morning.
[2:54] This is the godfather of deep learning.
[2:57] But it's been 10 years.
[2:59] So what actually happened?
[3:02] Well, AI really did get remarkably good at reading medical images.
[3:08] It's faster, it's more accurate, and it's cheaper than radiologists.
[3:14] But it turns out radiologists are actually doing better than ever.
[3:19] In the US alone, there's 16% more radiologists than there were 10 years ago.
[3:24] And they're earning 44% more on average.
[3:31] So AI didn't replace radiologists.
[3:34] It just made imaging so much faster and so much cheaper that we started doing vastly more of it.
[3:41] We started doing preventative screening at scale.
[3:45] More scans means more findings, means more follow-up, means more demand for radiologists, not less.
[3:53] Hinton gave them five years, and instead they got a renaissance.
[4:09] James Watt's steam engine had made coal dramatically more efficient.
[4:14] But in the 90 years since that invention, coal consumption had gone up every single year, not down.
[4:23] And so this is now known as the Jevons paradox.
[4:26] That when you make something cheaper, faster, more efficient, you don't just do the same amount of it for less money, you do a lot more of it.
[4:37] And this is playing out everywhere AI touches.
[4:43] So how did a future Nobel laureate get the future wrong?
[4:48] He made an assumption about demand.
[4:50] He assumed that it was finite.
[4:53] But his prediction didn't fail on technology, it failed on that assumption.
[5:02] It's broken at least one of your assumptions.
[5:04] The question is, do you know which one?
[5:10] So how do we find it?
[5:12] Well, first of all, you have to find where it's hiding.
[5:17] So you heard a bit from Sebastian, I've been doing this for over 30 years.
[5:21] I co-authored the book called Product Leadership, mainly as an excuse to interview hundreds of product leaders.
[5:25] And for the last 10 years or so, I've advised nearly 150 companies on product strategy, scaling, org design, and more.
[5:33] And what I've learned in that time is that every product, every team, every company has to be able to answer five simple questions.
[5:43] Where are we going?
[5:45] How are we going to get there?
[5:47] What matters right now on that journey?
[5:50] What are the actions we're going to take to move forward?
[5:53] And how do we choose between those actions?
[5:58] Now these seem delightfully simple, but I promise you, if you start going through your organization and asking them, they become harder and harder to answer.
[6:08] I call this the decision stack, and I use vision, strategy, objectives, opportunities, and principles to answer those questions.
[6:18] But the point here is that you can use anything else.
[6:20] If you don't like OKRs, that's fine, not everyone does.
[6:23] But you have to have an answer to those questions.
[6:25] And even more importantly, those questions or those answers have to connect.
[6:29] Because from the top down, you're answering the question how.
[6:33] Here's our vision, how are we gonna achieve that?
[6:35] Here's our strategy, how are we gonna achieve that?
[6:36] Here's our objectives, and so on.
[6:40] And from the bottom up, as you look at any individual piece of work, you should be able to ask why.
[6:46] Why is this important?
[6:47] Well, here's our objectives, why is that important?
[6:49] Here's our strategy, and so on.
[6:53] It's called a decision stack because every level is a decision.
[6:57] Your vision is a decision about where you're going and what you're not going to do.
[7:01] Your strategy is then a decision about how to achieve that and what you're not going to do, and so on.
[7:05] What you decide not to do is just as important as what you decide to do.
[7:12] Now every organization has a decision stack, whether you call it that or not.
[7:15] It's just that most look something like this.
[7:19] You have a vision from a few years ago.
[7:21] It might be on a poster on the wall of the lobby of the building that nobody actually goes to anymore.
[7:27] You have a strategy that's stuck in a PowerPoint slide on the file server somewhere.
[7:32] You have OKRs in a spreadsheet that nobody's looked at since the last OKRC meeting.
[7:38] And then we start sprinkling in AI strategy.
[7:41] And we wonder why everyone gets confused.
[7:46] To understand the impact of AI, you have to understand what your stack is sitting on.
[7:53] Because every decision is based on assumptions.
[7:57] So every layer of your stack has a hidden set of assumptions underneath.
[8:03] They might be assumptions about the future you're describing, assumptions about how you're gonna win and where you're going to play.
[8:09] Assumptions about whether you have measuring the right outcome.
[8:15] Again, AI just pulled one of those assumptions and broke your stack.
[8:19] The question is, do you know which one?
[8:22] The GIF wasn't working.
[8:25] It's also worth knowing why it broke, because the ground is moving faster than any of us can see.
[8:32] Today's AI is the worst AI you are ever going to work with.
[8:37] Every model you use from here on out is going to get better and faster.
[8:43] Adoption is faster than any technology that we've had before, and model costs are collapsing 10x year on year.
[8:53] Might not feel like it when you get your anthropic bill at the end of the month, but that's because you're using so much more of it.
[9:00] And we've seen this before as well.
[9:03] Google made the same bet.
[9:05] When Larry and Sergey launched Gmail in 2004, they launched it with one gigabyte of storage.
[9:12] At a time when competitors were offering two to four megabytes.
[9:17] They launched on April 1st, and people literally thought this was an April Fool's joke.
[9:22] But Google wasn't betting on the storage cost of the day, they were betting on the future.
[9:27] When they started planning this in 2000, storage cost about 12 dollars a gigabyte.
[9:32] When they launched in 2004, it was already down to a dollar a gigabyte.
[9:36] And today it's a fraction of a penny.
[9:40] So we have to ask ourselves the same question about AI.
[9:44] What if models become commodities?
[9:46] What if the cost of a token goes to zero?
[9:49] What if the cost of software goes to zero?
[9:54] But those are some pretty big assumptions.
[9:57] Tokens going to zero rest on other assumptions about energy costs, the cost of chips and data centers.
[10:03] Hinton got it wrong.
[10:04] What if I'm wrong?
[10:05] What if the opposite happens?
[10:08] What if models become MOS?
[10:10] What if the cost of a token goes up 10x?
[10:16] I don't know which of those is coming, and neither do you.
[10:20] But that makes it more important, not less, to know which assumption that you're standing on.
[10:25] And most companies are looking in the wrong place.
[10:30] Most companies are only looking at the opportunities layer, and they're basically treating AI as a feature.
[10:38] Which means that we're sprinkling AI on everything.
[10:41] Every product, every pitch deck, every company update now leads with AI powered.
[10:46] And most of it is incremental at best.
[10:50] AI as a feature basically assumes that the customer's problem is a lack of AI.
[10:59] And we think that because we live in a tech bubble.
[11:03] Each dot on this chart represents about 3 million people around the world.
[11:08] Despite the fastest adoption of any technology in history, four out of five people on Earth have still never touched AI.
[11:18] That little red dot, that's us.
[11:31] We have to remember that most of the world doesn't want AI.
[11:36] The AI panic, data center controversy, and the general inshidification at the feature layer means that AI powered is now a liability.
[11:46] A recent Washington State University study showed that the term artificial intelligence in product descriptions actually reduces purchase intent.
[11:58] As April Dunford reminds us, customers don't buy AI, they buy the elimination of a pain point.
[12:07] So what does it look like when you make that assumption?
[12:11] Before AI, the Cheg value proposition was something like this.
[12:15] A student would Google a homework problem, they would find a Cheg article, they would sign up and paycheck a subscription for answers to all of their homework problems.
[12:26] Before AI, Cheg had a near monopoly on homework help.
[12:32] In May 2023, their CEO did their quarterly update.
[12:38] And this is a few months after ChatGPT dropped, and said that in the first part of the year we saw no noticeable impact from Chat GPT.
[12:46] He went on to say, our first big step is the introduction of Chegmate, which we recently announced in cooperation with OpenAI.
[12:58] So when AI came along, Cheg chose to focus on features.
[13:02] And they built their own bot called Chegmate.
[13:06] This is the same assumption.
[13:07] They thought that the student's problem was a lack of AI-flavored Cheg.
[13:13] The assumption they didn't look at was much further up the stack.
[13:18] Whether students would keep paying for answers in a world where they can get them for free from ChatGPT, Gemini, Perplexity, and everyone else.
[13:28] And that assumption cost them dearly.
[13:32] While he was on that call, the stock dropped by half.
[13:40] And their revenue has dropped 75% from pre-AI highs.
[13:45] They stopped reporting subscriber numbers at the end of 2024.
[13:49] And in July of this year, they got warned by the New York Stock Exchange that they might get kicked off the stock exchange.
[13:57] They never asked which assumption changed.
[14:02] Turns out Cheg's customers didn't really want Cheg.
[14:05] They definitely didn't want AI.
[14:06] They just wanted their homework done.
[14:14] Too many companies are looking at the objective level and see AI as cost cutting.
[14:20] This is basically the same assumption that Hinton made: that demand and value are finite.
[14:28] And it leads too many organizations to ask: how do we use AI to do what we already do, but cheaper?
[14:35] To me, that's using once in a generation technology to shrink yourself.
[14:42] Imagine this 10 years ago, if you had woken up, your engineering team had been 10 times more productive, would you have fired 90% of them?
[14:51] Of course not.
[14:52] Nobody fires that many engineers.
[14:54] They ship more products, they take more swings, they enter new markets.
[14:58] But this is exactly the conversation we're having about most other roles today.
[15:06] When IKEA started rolling out AI customer service, they quickly noticed 50% of their tickets were being resolved automatically by the AI agent.
[15:17] The obvious move, the move almost any other company would have made, would have been to lay off 50% of the help desk staff, bank the cost saving, and move on.
[15:28] Instead, IKEA thought differently.
[15:30] They looked at the remaining 50% of tickets and realized that they were almost all design questions.
[15:36] People asking whether the curtains matched their sofa, whether this table would go well with these chairs.
[15:44] And they realized there was an opportunity there.
[15:46] So they went back to their assumptions and realized that their assumption of what a help desk was for didn't hold.
[15:54] But their strategy did.
[15:56] So they rebuilt from there and they retrained their help desk as remote interior designers.
[16:07] As Ulrika Bissert says, people have been at the heart of IKEA for over 80 years, and that's exactly where they'll stay.
[16:15] The best part is this is not just a feel good story.
[16:19] Because that team now drives meaningful new revenue.
[16:23] In their first year alone, they delivered 1.4 billion dollars in net new revenue.
[16:30] Same humans, completely different value.
[16:35] So AI is the biggest growth lever your business has ever had, and treating it as a cost lever is a failure of imagination.
[16:47] They asked the question, and they found the assumption one level up.
[16:53] So we have to stop treating AI as a feature, we have to stop treating it as efficiency, and we have to start treating it as a multiplier.
[17:05] Because the companies that will win in the next decade and on will be the ones that are rethinking their strategy.
[17:13] As Rita McGrath said, in a fast moving world, strategy is more critical than ever because organizations have to make many more decisions.
[17:23] And she wrote this 10 years ago.
[17:51] But what does the evidence actually show?
[17:56] Well, a recent McKinsey report shows that a third of companies are trying to replace at least one SaaS product.
[18:04] But MIT studies show that 66% of in-house builds fail.
[18:10] And enterprise software spend is going up regardless.
[18:17] We have to remember, most of the world doesn't want to build software.
[18:23] They want to do the thing that software enables.
[18:27] A restaurant manager is not gonna build their own point of sale system, their own payment stack, their own inventory management system, no matter how good AI tooling gets.
[18:36] They want software that's easy to use, customizable to their needs, and then gets out of the way so that they can focus on what's most important to them, their customers.
[18:46] And that's true of almost everybody that uses software today, and the billions who don't use software yet.
[18:55] Even Anthropic still uses Slack, Workday, Figma, and more, and talk about this quite publicly.
[19:02] They have thousands of the world's best engineers, and they definitely don't have to worry about the token budget.
[19:07] And yet they choose still to use software and not rebuild it.
[19:11] Let's face it, nobody uses workday out of choice.
[19:19] So what happens when you reject this assumption and look a little bit deeper?
[19:24] Well, my favorite story here is Intercom.
[19:27] If you don't know Intercom, they're one of the classic SaaS companies that came out kind of mid late 2000s, early teens.
[19:35] Founded in Dublin, they sell software to support teams.
[19:40] In 2022, the business had kind of plateaued, growth had stagnated, but there was still a you know, a billion-dollar company, they were doing really well.
[19:46] And then ChatGPT lands.
[19:54] As co-founder Des Trainer tells it, so much of our product was assuming that it was humans who sat in seats talking to customers.
[20:02] When AI happened, it became very clear that that was not gonna be true anymore.
[20:09] When they realized what this was going to do to customer service, and that not everyone was gonna be as scrupulous as IKEA, they could have slapped some AI features on their software and hoped for the best.
[20:21] But instead, they went up the stack.
[20:24] AI features, there's not really much point.
[20:27] Our business model about selling support equals more seats sold equals more revenue, that's probably broken.
[20:33] Our strategy assumption that people are how businesses do support, that one's broken too.
[20:38] All right, what do we have left?
[20:40] Businesses want to support their customers.
[20:43] That's probably still true.
[20:45] That's still valuable.
[20:47] So they rebuilt everything from there.
[20:50] ChatGPT dropped on a Thursday.
[20:54] By Monday, they had scrapped their entire roadmap, reorganized their company, and gone all in on AI.
[21:02] And they built Fin, an AI customer service agent that now resolves over a million tickets a week.
[21:10] But the real move wasn't the technology, it was the pricing model.
[21:14] They moved from per seat pricing to 99 cents per resolution.
[21:19] So customers only pay for Finn when it actually solves a customer problem.
[21:26] Think about what that means strategically.
[21:28] In their old model, more customer problems equals more customer service agents equals more seats sold, completely misaligned incentives.
[21:37] In their new model, better resolution equals more revenue.
[21:41] The product simply has to work.
[21:44] And Finalone went to over 100 million ARR in the first year.
[21:50] They've then taken that expertise and started expanding into e-commerce agents, sales agents, and more.
[21:56] It's so successful they've renamed the whole company Fin.
[22:01] And in their last investment round in 2018, they were valued at 1.3 billion dollars.
[22:08] But just a few months ago, they were sold to Salesforce for 3.6.
[22:14] That's the value of a strategy level rethink.
[22:18] And so much for this pocalypse.
[22:32] IKEA went up to strategy and found that held.
[22:34] Intercom went all the way to Vision, found that held, and then rebuilt from there.
[22:40] So they went up until they found the assumption that held.
[22:45] So how do you find yours?
[22:49] Well it starts with this reminder from IBM that a computer can never be held accountable, so must never make a management decision.
[22:57] Every layer of the stack is a decision, and every decision needs somebody who is accountable for it.
[23:03] And that's true of your assumptions as well.
[23:06] And that means all of you have to step up.
[23:10] You have to start by naming your assumptions.
[23:13] And you can use Roger Martin's question, what would have to be true?
[23:19] And then you can go up the stack until you find the assumption that still holds and rebuild from there.
[23:26] You can start asking yourself, can we solve this customer problem in a completely new way?
[23:31] Not incrementally better, fundamentally differently.
[23:34] The way AI powered imaging didn't just make radiology cheaper, it made preventative screening possible at scale.
[23:43] You have to start asking yourself, are we still measuring the right thing?
[23:47] Do our objectives still make sense?
[23:49] If we're measuring something like MAU in our app, does that mean anything in a world where customers are using agents and MCP servers?
[23:56] At the strategy level, ask yourself can your business model change completely?
[24:01] Like intercom moving from seats to resolutions?
[24:04] What does outcome based pricing look like in your world?
[24:09] You can also ask yourself what demand is currently locked up that AI could unlock.
[24:14] This is the Jevons paradox question.
[24:17] What would your customers do with your product if it was 10 times cheaper, 10 times faster, or 10 times more accessible?
[24:26] Fundamentally, we have to ask ourselves what can we now build that was impossible before.
[24:35] I already test assumptions, I hear some of you saying.
[24:38] Yeah, we do.
[24:39] As product people, we're getting really good at it.
[24:42] We do the research, we do the customer interviews, we do the prototyping, we do a fake door test, then we argue about sample sizes.
[24:50] We've gotten really good at testing assumptions at one level.
[24:57] We obsessively test these relatively cheap decisions, and we pass five-year strategies on vibes.
[25:06] It's time to take those product superpowers up the stack and test all of the assumptions.
[25:14] But to do that, there's a catch.
[25:16] Because every opportunity test we run is on historic data.
[25:20] So with existing customers, it's a data that you're using in your app, maybe you're talking to prospects.
[25:25] But the future doesn't have any data.
[25:28] You can't A B test 2028.
[25:32] So we have to use our toolkit slightly differently.
[25:47] So remember that scenario from earlier?
[25:49] What if tokens go to zero?
[25:51] What if they go to 10x?
[25:53] You basically have to plan for both.
[25:56] And then you can see which assumptions survive different futures.
[26:03] Cheg's assumption that students will keep paying for answers survives neither of those futures.
[26:09] But Intercom's assumption that customers will still pay to resolve customer service problems holds in both.
[26:18] And then the brave part.
[26:16] You have to make a bet.
[26:22] You should absolutely test as many assumptions as you can, get as much research and data as you can, but you have to be comfortable moving ahead with 70% confidence.
[26:35] As Annie Duke says, this language is really important because calling it a bet brings a risk out in the open, makes explicit what is already implicit, and gives us something to come back to.
[26:48] Because that's the most important part.
[26:50] You have to write it down.
[26:52] And you have to write down what would prove you wrong.
[26:56] Annie Duke calls these kill criteria.
[26:59] Intercoms was the resolution rate, IKEA's the revenue generated by the new team, Google's was the price of storage.
[27:06] All of these are assumptions that they can come back to and retest and make sure that they're still on track.
[27:14] So do you know which assumption is missing in your stack?
[27:17] You will tomorrow, because you're gonna go and find it.
[27:29] Because if you don't challenge it, who will?
[27:40] I've made about 20 predictions on stage today, and I'm definitely not a Nobel laureate.
[27:47] But just because we can't predict the future, but just because we can't predict the future doesn't mean we can't prepare for it.
[28:00] AI is moving fast, but we're still only in the typing commands into DOS era of AI.
[28:08] And in the DOS era, we assumed that that was the best way to interact with computers.
[28:16] But while the rest of the world was happily typing commands into a black screen, a handful of people at Xerox Park, Adele Goldberg and Alan Kay among them, had already decided that that wasn't good enough.
[28:27] And they invented the graphical user interface, they invented the mouse that you've used every day since.
[28:36] Alan Kay reminds us that the best way to predict the future is to invent it.
[28:49] So Hinton's assumption gave radiologists five years, and instead they got a renaissance.
[28:56] My question to you is what are you going to do with yours?
[29:02] Thank you very much.
[29:03] If you want to learn more, you can follow the QR code, get the first two chapters of the book for free.
[29:08] And thank you very much.
[29:19] Absolutely fantastic.
[29:21] So who has a question for Martin?
[29:27] And why you think?
[29:29] Because nobody wants to be the first.
[29:30] I go first.
[29:32] So when you talk to people, what would you say is the assumption that you hear most often that you think is absolutely wrong?
[29:44] I think it it's well, here's our first it depends of the day.
[29:48] Um but I think it depends on the company, right?
[29:50] And so for me it's really the assumption, the core assumption that most people get wrong is that none of their assumptions have changed, right?
[29:57] That the strategy is still fair, that they can still execute on that.
[30:00] And when I walk through this exercise with companies, and even in the workshop we did yesterday, you quickly realize that wait, there's some assumptions here about even our objectives or our strategy or our vision that might not be true anymore.
[30:10] And so we have to go have that conversation.
[30:12] So I think it's just everyone's reacting so much right now to AI and the speed of AI, but they forget to go back to the drawing board a little bit and see wait a minute, has this actually changed a bunch of the things in our stack?
[30:23] Brilliant.
[30:24] Thank you so much.
[30:25] So, questions.
[30:28] Can't wait to throw this.
[30:32] I mean I can try.
[30:33] Yeah.
[30:34] Who wants a question?
[30:35] Come on.
[30:36] Don't be shy.
[30:37] Give it a try.
[30:38] Everyone's stack is perfect.
[30:40] Right in the back.
[30:40] I don't know if I can throw it that far.
[30:42] Okay, so important.
[30:43] Everybody's looking in my direction.
[30:45] That is very
[30:46] Very important.
[30:47] We learned that in the first year.
[30:48] And Martin's gonna throw it as far as he can.
[30:50] Oh boy.
[30:51] And you pass it on, unless you can throw it until the very, very.
[30:56] Back.
[30:57] What's the best one?
[31:00] Whoa!
[31:02] That was really good.
[31:02] By halfway.
[31:04] Yeah.
[31:06] It's gentle throws.
[31:09] There we go.
[31:12] All right.
[31:13] So I'm just curious as we go up the stack.
[31:18] Would you see us using AI to kind of explore and forecast and predict?
[31:27] Or the upper stack layers are limited to human.
[31:35] I I think it's a great question.
[31:36] I think AI can absolutely be a thought partner in this process, right?
[31:40] Even going up your strategy and your vision and helping you do research and looking at different options.
[31:45] It's also fantastic at helping you know look for consistency and arguments you're making and things like that.
[31:50] But I do still think there's an element here of uh the human intuition and that kind of leap of faith that we almost have to do at a strategy level, right?
[31:58] AI is not very happy making bets.
[32:00] Uh so I think that still has to be the human role.
[32:03] And I think, especially as product people, our ability to like see across all these signals that we're getting and see all the threads of what's happening.
[32:10] And wait a minute, this customer said that, and then this you know, enterprise customer said that.
[32:14] I can connect those dots.
[32:16] That to me is still a uniquely human capability.
[32:19] But AI can absolutely help you, right?
[32:20] It's a fantastic tool.
[32:22] Uh, and we use it a lot when we're working with clients to like figure these things out and go do the research.
[32:26] But then you should also go validate that yourself, right?
[32:29] Because it can also hallucinate, as we know.
[32:31] So we still have to go check that with humans and go talk to humans.
[32:34] Um, all of this comes back to you know, the lean startup methodology, comes back to Steve Blank's customer development.
[32:40] Like we have to talk to people.
[32:42] Uh, and again, AI is can do some of that, can give you the insight, but it can't really do that validation, I think.
[32:48] So it has to be hand in hand.
[32:54] Anyone else?
[32:56] We have way down here.
[32:57] If you want to try to throw it all the way down, everyone watch so they don't get hit in the head.
[33:02] Okay, I can't see the, I can't see the hands.
[33:04] Let's start over there, maybe, and then we'll come down to the front.
[33:11] Hi Martin.
[33:12] Um would you have any recommendations for uh challenging vision and strategy in a founder-led organization that's been around for over 20 years and have very specific ideas of where the business should go.
[33:27] And on one hand, maybe resistant to change, but are leaning fully into this AI hype at the same time.
[33:34] Yeah, this is this is my everyday um.
[33:37] I think there's a couple of things that you can do, right?
[33:39] So one is as soon as you go at strategy and vision and start challenging them directly, it becomes a very personal thing, right?
[33:48] And it sounds like you're challenging, or they feel like you're challenging leadership, the founder, it can be very confrontational.
[33:54] So I think it's there's other ways you can do it.
[33:56] So one is definitely working on the assumptions, right?
[33:58] So if you challenge the assumptions, you're not challenging the person.
[34:01] You're you're basically saying our strategy seems to rest on the assumption that you know enterprise customers need XYZ.
[34:06] I've gone and talked to 10 customers, they don't need that anymore.
[34:10] What do we need to change, right?
[34:12] Um, so I think that's one thread, and I think that's why the assumptions behind your stack is such a powerful tool for product people to be able to challenge upwards.
[34:19] I think the other part of it uh is to always bring evidence to that, right?
[34:23] Because if it is a hip opinion, um then the highest paid person's opinion is gonna win.
[34:27] The hip hippo is gonna trample all over you.
[34:29] But if you bring evidence, you bring data, you bring customer stories, that changes the conversation.
[34:35] And my final hack I would say is people don't even always realize that their strategy's gone out of alignment, right?
[34:41] My favorite trick when I go into a new company is simply to go ask everyone on the leadership team individually what they think the strategy is.
[34:49] And it won't surprise you to hear that I get a different answer back from almost every single person.
[34:54] And playing that back to them is the best way to get that aha moment of like, oh, wait, something's broken here, or we need more alignment, we need to re revisit this conversation.
[35:04] Again, that's a very different thing than if I just came into the room and went, your strategy isn't working, we need to come back to that, right?
[35:10] Because that's confrontational, but if you actually replay their words to them, it's it's not.
[35:15] So those would be some hacks I would think about of how to approach that and really how to depersonalize that kind of challenge, um, and but still get the result that you want.
[35:26] One more question.
[35:27] We had a hand down at the front somewhere, but if you can get this.
[35:31] Alright, throw it.
[35:31] Watch it.
[35:32] Wait, wait, wait.
[35:33] Now you need to turn around, please.
[35:35] Yes.
[35:36] Watch your head.
[35:37] So we have insurance, but I don't want to use it today.
[35:41] Oh, or we have a runner.
[35:43] Actually, we might not have insurance for this exact situation.
[35:49] I'm not sure.
[35:50] Let's not try it.
[35:57] Mike check?
[35:59] So, my question is um I often faced with the challenge that um the human in the loop always should be part of the AI interaction because somehow we have to be responsible and can get fired in case something is not wrong, is not working accordingly.
[36:19] So do you truly believe this is an assumption which could be true, or is it more like a limitation?
[36:25] Because if I think if a plane is crashing, there also something, some automatism is already failing, and it's fine as well.
[36:33] I think uh it probably depends on what level you're doing it, right?
[36:36] I think for sure if you have a an agent that's helping you with you know checking in and out tickets or you know, even checking in and out code, maybe that's not as critical, and you can kind of let them go without that that human in the loop.
[36:48] Uh once you built in all the evals and all the testing and all the follow-up.
[36:52] But I think when it comes to these kind of questions, right, when you're talking about objectives, when you're talking about strategy, when you're talking about vision, I think a human definitely has to be in the loop.
[36:59] And that's not just about the accountability.
[37:01] I think it is again about that inductive leap part, right?
[37:04] We can't just take what AI gives us and go, yep, that's a cool strategy, let's go execute, right?
[37:09] We have to kind of judge it through our lens.
[37:11] I think humans and you know, in your industries, we have so much more experience still than even the biggest LLM models uh in understanding our industry, what our customer problem really is, what the kind of end-to-end workflow looks like, etc.
[37:25] That we can quickly spot AI coming up with some really stupid ideas sometimes, right?
[37:30] And that judgment is important.
[37:31] Um it means we find the good ideas, right?
[37:33] Because AI is also gonna come up with great ideas, but our judgment needs to be there to make the filter, I think.
[37:40] All right.
[37:41] So please give a warm and roaring at laws.
[37:44] Thank you very much, everyone.
[37:49] Thank you so much.
[37:50] Thank you.