<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Itai on Medium]]></title>
        <description><![CDATA[Stories by Itai on Medium]]></description>
        <link>https://medium.com/@itaiboublil?source=rss-cd936f90d430------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*97TrTqHNp-tv63R3LnBLwA.jpeg</url>
            <title>Stories by Itai on Medium</title>
            <link>https://medium.com/@itaiboublil?source=rss-cd936f90d430------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Thu, 19 Nov 2026 02:44:22 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@itaiboublil/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[The Six Enemies of Every Founder]]></title>
            <link>https://medium.com/@itaiboublil/the-six-enemies-of-every-founder-61a65b5c7433?source=rss-cd936f90d430------2</link>
            <guid isPermaLink="false">https://medium.com/p/61a65b5c7433</guid>
            <category><![CDATA[founders]]></category>
            <category><![CDATA[startup-founders]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[startupş]]></category>
            <dc:creator><![CDATA[Itai]]></dc:creator>
            <pubDate>Sun, 15 Mar 2026 09:05:01 GMT</pubDate>
            <atom:updated>2026-03-15T09:05:01.992Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/612/1*ivcRJF0tqu88U9hgOu40BQ.jpeg" /></figure><p>Here is what nobody is telling you. Opening your eyes at three in the morning is a “thing!”</p><p>I was in my late forties when I made the decision to become a founder. That age matters because it means I was not a twenty-six-year-old with nothing to lose and a runway of cheap rent and parental patience. I had built something across two decades in SaaS. I had a real salary, a real title, and a professional identity with weight and history behind it.</p><p>I left all of that on the table, deliberately, because I had a conviction I could not talk myself out of. I wanted to change something that no one else had managed to change. And what I discovered, without any cushioning whatsoever, was that <em>entrepreneurship does not attract a certain kind of person</em>. It simply removes everyone who cannot survive what it actually is.</p><p>The filtering mechanism is psychological. It is a sequence of forces, each arriving in its own time, each capable of ending the venture entirely. Most founders never name them, so allow me to share with you my first-hand experience.</p><h4>The First Two: Doubt and Fear</h4><p>These two show up first. They show up together, and they catch you by surprise.</p><p>Doubt does not argue with you. That would be easier to handle, believe me. Instead, it just asks questions at the worst possible moment.</p><p>Was that customer a repeat or a one-off? Is the differentiation you keep pitching actually real, or have you rehearsed the story so many times that you have started believing your own slide deck? Did you pick this architecture because it was right, or because you had already announced it?</p><p>These questions do not need answers, oh no, that is not what they are there for. They just need to land right before you need to be decisive, and they do, with no fail, every single time.</p><p>Fear is doubt’s operational partner. If we think of doubt as “something” that is questioning the past, <em>fear is a force that colonizes the future</em>. Every outcome you’re facing starts looking like a worst-case scenario before it has had a chance to develop. You soften the ask when you should hold firm. You add a caveat to something that didn’t need one. You wait for more data when the data is already sufficient. It is less an emotional damage and more an operational one. Your decisions get slower, softer, and smaller than the moment requires.</p><p>There have been more instances than I can count where I sat across from someone, managing both the conversation and my own internal noise at the same time, and made a call I did not feel certain about. The only way to get capable at it is to keep doing it until the noise loses its veto power.</p><h4>The Third Enemy: Loneliness</h4><p>Nobody talks about this one honestly, so let me.</p><p>Everyone in the building is looking at you to know whether they should be worried. Your investors want to see a founder in control. Your team wants to believe the vision holds. Your early customers are watching to see if you are still standing. So you perform the confidence, all of it, every day, and then the meeting ends, and you are back to the actual situation, which nobody in that room was fully informed about.</p><p>That gap, between the face you present and the reality you carry, is where the loneliness actually lives.</p><p>Entrepreneurs rate <a href="https://lifehackmethod.com/blog/entrepreneur-mental-health-statistics/">their loneliness at 7.6 out of 10,</a> and many report spending significantly less time with friends, a spouse, and their children as a direct consequence of the hours this life demands. That number does not surprise me. What surprises me is that it is not higher.</p><h4>The Fourth Enemy: Impostor Syndrome</h4><p>Ok, I am going to be very straight here. Imposter syndrome does not retire when the results improve, so please get it out of your head/fantasy. You close a round, and the voice does not disappear. It just finds new material. You got lucky with that customer. The investor does not yet fully understand the risk. This version of the product will expose something you have been papering over. The argument updates itself continuously, always using whatever just happened as evidence for the same conclusion: you are not the person everyone in that room thinks you are.</p><p>I have been in those rooms more times than I can count. Delivering the pitch with full conviction on the surface, while somewhere in the background, my mind is already running the scenario where they ask the question I have no clean answer to, and the whole thing unravels right there. Learning to let that parallel track run without giving it the wheel is one of the more underrated skills in this business.</p><h4>The Fifth Enemy: Identity Collapse</h4><p>This one arrives later and is the most dangerous because it shows up when you are already depleted.</p><p>In my corporate career, my identity was distributed across multiple layers. It lived in my title, my team, my organization, and my compensation structure. If one thing failed, the others held. When you build a company from scratch, the distribution collapses. The company becomes your identity. Its failures become your failures. Its setbacks feel like personal verdicts rather than ordinary turbulence.</p><p>I depleted my savings. I found myself in a rabbit hole so deep it took weeks to climb out, and the climb was not just financial. The money problem was actually the simpler part. The harder part was recovering a sense of self that existed independently of whether the current month’s progress was pointing in the right direction.</p><p>When the company and the person become one and the same, a business setback ceases to be just a strategic problem. <em>It becomes a question about who you are</em>. The recovery, when it comes, has to address that layer first. You cannot make good strategic decisions while you are also conducting an identity crisis. Getting clear on where the company ends and you begin is not a therapy concept. It is an operational prerequisite.</p><h4>The Sixth Enemy: The Feedback Asymmetry</h4><p>Here is the thing about markets: they do not hurry. You can be wrong for a long time before the signal is unambiguous enough to act on. In that window, almost everything you hear is negative and self-generated.</p><p>The pilot dragged on, going nowhere. The partnership conversation that stalled at the third meeting never recovered. The hire that seemed right, but wasn’t. The product decision looked defensible until it became clear it was not.</p><p>The positive signal you eagerly wait for, when it comes, arrives in pieces. It can be a customer call that goes genuinely well, or an investor who follows up unprompted, or even a user who tells you the product changed how their team operates. These things happen, and they are real, but they, and that is a very important point, do not arrive on <em>your</em> schedule, and they do not cancel out the accumulated weight of everything that has been going sideways simultaneously (because chaos is the nature of startups). You develop a tolerance for operating in that imbalance. You have to, and if you cannot, then this journey is not for you.</p><h4>How to Handle It</h4><p>Every consultant, coach, and podcast guest has a framework for this. Three steps, five principles, a morning routine, pet a dog, raise a cat, whatever. I do not have one of those (though I do love dogs). What I do have is a short list of indicators that the needle is actually moving. These indicators are based on direct experience rather than pattern-matching to someone else’s situation.</p><p>The first thing is naming the force when it arrives. When you can name it precisely, it loses reach. I don’t focus on the fear that everything is collapsing, but rather on the fear of one specific conversation with one specific person next Wednesday.</p><p>I don’t dwell on the doubt about the whole thesis; instead, I focus on whether the positioning on slide seven actually holds up against what the market is telling you this month. The more specific you get, the smaller the thing becomes, and the smaller it is, the more manageable it gets.</p><p>The second is finding people who are in it the same way you are. I am not talking about advisors or investors, both of whom have structural reasons to maintain optimism about you. I am talking about other founders. People at a similar stage, with similar problems, who have no incentive to perform.</p><p>The conversation you have with another founder at 11pm about what is actually happening is worth more than most board meetings.</p><p>The third is separating yourself from your company’s current performance or stage. This is genuinely hard when you are the company’s most visible face, its chief salesperson, its primary decision-maker, and its emotional anchor. But it is necessary. Founders who embed their self-worth entirely in their business outcomes are more susceptible to burnout and less able to make clear decisions under pressure. The company needs you to be functional. Functional means knowing <em>where you end </em>and<em> where the company begins.</em></p><h4>A Happy Ending Is Not Guaranteed</h4><p>Persistence is key. I have seen startups fail because of decisions made by their founders. Other startups fail because of conditions the founder never had any control over. The latter is more common than you might think. It can be, for example, that the market did not move as the thesis required, or that the capital environment turned hostile at exactly the wrong moment, or, in some cases, that a regulatory change rewrote the landscape six months after the product launched. <em>Effort does not cancel all of those variables</em>.</p><p>What I can tell you is that the people who stay, the ones who climb out of the sixth rabbit hole having already climbed out of the previous five, <em>are not optimists</em>. They are <em>not particularly fearless</em>. Instead, they developed, through repetition and through pain, an almost irrational refusal to interpret a “no” as permanent. They learned to separate the business being wrong in this iteration from the thesis being wrong entirely. They got very good at absorbing failure without letting it become identity.</p><p>Are we a little crazy? Probably. Aristotle observed that no great mind has ever existed without a touch of madness. I am not claiming greatness. I am just saying that the people who do this work, and keep doing it, and keep showing up after the things that should have stopped them, are running on something that cannot be fully explained by rational expected-value calculation.</p><p>We do it because we cannot <em>not</em> do it.</p><p>And we are very, very good at <em>not</em> accepting a no.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=61a65b5c7433" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The North Star You Never Look At]]></title>
            <link>https://medium.com/@itaiboublil/the-north-star-you-never-look-at-7b7380c144bb?source=rss-cd936f90d430------2</link>
            <guid isPermaLink="false">https://medium.com/p/7b7380c144bb</guid>
            <category><![CDATA[gtm-strategy]]></category>
            <category><![CDATA[founders]]></category>
            <category><![CDATA[gtm]]></category>
            <category><![CDATA[startup]]></category>
            <dc:creator><![CDATA[Itai]]></dc:creator>
            <pubDate>Mon, 09 Mar 2026 02:48:01 GMT</pubDate>
            <atom:updated>2026-03-09T02:48:01.983Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*O7p4vKeyfmzjLsH73asIlQ.jpeg" /></figure><p><em>Why most startups scale wrong, what a leading indicator actually fixes, and why in a world of fast-copying AI and nimble competitors, the framework in Mark Roberge’s Science of Scaling is not just good advice. It is survival.</em></p><p>Almost every founder I have talked to knows a version of the same story, even if they have never said it plainly. Things are working. The Series A closes, you spend two months building out a sales floor, and somewhere around the third week of January, you tell the team that this is the year the business finally starts running the way it was always supposed to. Then it is October, and the board deck has a slide on churn that you hope nobody asks too many questions about.</p><p>I want to be precise about what that story is. It is not primarily a story about execution failure, bad hiring, or a product that did not work. The mistake underneath it is something more fundamental: founders scaling a business before they had genuinely understood it. Revenue, at the early stage, is largely sales force output. When someone signs a contract, what you learn is that your salesperson was persuasive enough to get them to take a chance, which is different from learning whether the product delivered what it promised once the new customer relationship wears off.</p><p>That distinction between what the sales team produced and what the product delivered is what Shikhar Ghosh’s research at Harvard Business School kept finding beneath failed venture-backed companies. Approximately 75 percent of venture-backed companies never return investors’ capital.[1] That is not mostly bad luck. It is mostly people scaling before they understand what is working.</p><p>Mark Roberge spent years as CRO at HubSpot watching this happen, and the <a href="https://www.stage2.capital/science-of-scaling">framework he put forward in Science of Scaling is an attempt to solve it</a>. It is not a consultant’s framework with six pillars and a diagram. Instead, it is a set of instruments. The kind of instruments you can use in a Tuesday morning meeting to determine whether you are actually ready to do what your investors are asking you to do.</p><p><strong>THE INDUSTRY’S COLLECTIVE DELUSION ABOUT PMF</strong></p><p><strong>Everyone uses the term. Nobody agrees on what it means.</strong></p><p>Product-market fit has accumulated the kind of authority that phrases get when they circulate long enough without being tested. Marc Andreessen, who popularized the concept in a 2007 essay called “The Only Thing That Matters,”[2] described it as being in a good market with a product that can satisfy that market. Which is true, and also entirely unfalsifiable. You cannot look at a business and know whether it satisfies its market until you have watched what the market does over time.</p><p>Sean Ellis tried to make it measurable. In 2009, after benchmarking nearly a hundred startups,[3] he landed on a single survey question: <em>how would you feel if you could no longer use this product?</em> If forty percent or more said they would be very disappointed, Ellis called it fit. That is a sharper instrument than Andreessen’s description, and it has the virtue of being something you can actually administer. The trouble, however, is structural. Survey respondents give you what feels true at the moment of answering, and what feels true in the moment of answering a survey about a product you just bought is often more generous than the judgment a person makes eleven months later when renewal paperwork arrives, and they have had time to find out whether the thing actually solved their problem.</p><p>Both definitions share the same limitation: they are snapshots of a moment rather than measurements of a rate. Roberge’s argument is that product-market fit is not a state you arrive at and declare; it is the consistent generation of customer success across successive cohorts. That reframe matters enormously when you are trying to decide whether to add twenty salespeople, because a snapshot can look healthy while the underlying trend is quietly deteriorating.</p><p>The proxy Roberge uses for customer success is <em>retention</em>. Annual retention above ninety percent in SaaS, or the equivalent behavioral signal in whatever business model you are running. Not because ninety percent is a magic number, but because a customer who renews has made a considered decision with money behind it. They had twelve months to reconsider, to look at alternatives, to decide whether the product was earning its cost. The renewal is a signal of a different quality than anything you can get from a survey or an early usage metric.</p><p>The practical problem with using retention as your primary instrument is that it arrives too late to be a useful real-time decision tool. A customer who signs in January will not tell you whether you actually delivered on the value proposition until January of the following year, and by that point, you have already spent twelve months making hiring decisions, pricing decisions, and go-to-market investments based on incomplete information. Roberge’s leading indicator framework is designed to compress that feedback loop into something you can actually act on in a timely manner.</p><p><strong>THE MECHANICS OF A LEADING INDICATOR</strong></p><p><strong>The ‘aha moment’ is not a metaphor. It is an engineering problem.</strong></p><p>The Customer Success Leading Indicator is a defined behavioral event that predicts long-term retention with statistical reliability. The formula is deliberately spare: <strong><em>P</em></strong><em> percent of customers achieve </em><strong><em>E</em></strong><em> event within </em><strong><em>T</em></strong><em> days</em>. You define E based on what you believe represents genuine value extraction from your product, run the correlation analysis to confirm it actually predicts retention, and then track it through cohort charts as a two-month proxy for what would otherwise take a year to know.</p><p><em>P percent of customers achieve E event(s) within T days.</em></p><p>Slack’s number was 70 percent of customers sending 2,000 or more team messages in the first 30 days. Dropbox’s was 85 percent uploading a single file into a folder on a single device within the first hour. HubSpot’s was 80 percent of customers using five or more features within 60 days.</p><p>Each of these events was chosen because it encoded the company’s core value proposition in behavioral terms. For HubSpot, an all-in-one platform competing against specialist tools, one-feature usage was a strategic liability. A customer using a single HubSpot feature could be replaced tomorrow by any point solution. A customer using five was inside an integrated system that was genuinely costly to leave. The leading indicator was not just a retention predictor. It was the behavioral definition of the moat.</p><p>Once you have your P, E, and T, cohort analysis does the work. You organize customers by acquisition month, monitor the percentage of each cohort that meets the leading indicator, and track whether that percentage is improving over time. When the chart turns consistently green across several successive cohorts, you have permission to begin scaling. When it stays flat or moves erratically, that is the data telling you to stay in the learning phase and keep iterating on your target segment, your onboarding sequence, your pricing, your sales motion, until you understand the pattern well enough to replicate it deliberately.</p><p>What the framework requires, and what makes it genuinely difficult to follow in practice, is refusing to advance to the next stage until the data actually supports it, rather than until the investors are asking why you have not moved faster. Those two timelines are almost never the same.</p><p><strong>WHY PEOPLE PUSH BACK, AND WHY THEY ARE WRONG</strong></p><p><strong>The counterarguments are there. They are also largely excuses dressed as strategy.</strong></p><p>The pushback I encounter most often in VC meetings is about competitive pressure. If I slow down to confirm retention before scaling, someone else takes the market. I have heard this in enough VC and leadership meetings to know it sounds more airtight than it actually is. Roughly 35 percent of startups that raise Series A never make it to a Series B,[4] and the ones that fail are not typically the ones that moved too carefully. They scaled a motion they had not yet genuinely proven, and the cracks that were invisible at twenty customers became structural problems at two hundred.</p><p>The competitor you are worried about is almost certainly not doing what you imagine. They are not using your restraint to quietly build an unassailable position. They are probably hiring salespeople they cannot feed, generating revenue from customers who will churn in year two, and building exactly the leaky bucket you are trying to avoid. The race you are imagining, where someone captures the market while you think carefully, is mostly fictional. The real race is about whose economics hold. Theirs usually do not.</p><p>A second objection arises from variations in the business model. The framework works for SaaS, the argument goes, but breaks down for hardware, or marketplaces, or consumer products. This is partially right as a caveat and wrong as a conclusion. Every business has a proxy for retention. Repeat purchase rate. Engagement frequency. Net Promoter trend over time. The specific behavioral event that best predicts value extraction will differ, but the underlying logic is that you need to confirm you are consistently generating customer success before you try to scale that process, which is not SaaS-specific.</p><p>The more sophisticated objection is that leading indicators can be taught to. If you optimize your go-to-market team around hitting a specific behavioral threshold, you will get that threshold, but perhaps through coaching rather than genuine value delivery. This is worth taking seriously. It is also why Roberge does not recommend using leading indicators without first running a correlation analysis. In the TeleMed example from the paper, customers who achieved the leading indicator retained at 93 percent annually; those who did not retained at 39 percent. If you run that analysis and find a gap that large, the indicator is real. If the correlation is weak, you have the wrong indicator. The method is designed to be challenged. That is not a limitation. That is the whole point.</p><p><strong>SPEED OF MARKET AND THE AI ACCELERANT</strong></p><p><strong>The argument for this framework just got exponentially stronger.</strong></p><p>When Roberge was building HubSpot, the competitive timeline for software was manageable. A well-funded competitor needed twelve to eighteen months to build something credible enough to threaten your market position. That window was long enough to establish brand, distribution, and customer relationships that were genuinely hard to dislodge.</p><p>That window is gone. AI coding tools now allow small teams to ship in days what once required months. GitHub’s research on developers using AI assistants found task completion rates up to 55 percent faster.[5] Low-code and no-code AI platforms have cut time-to-market by as much as 70% compared to traditional development cycles.[6] What this means practically is that a team of five engineers with access to foundation models can build a working version of your product in weeks. Features that used to be defensible for a year are now a six-week sprint for a motivated competitor with decent funding.</p><p>The venture community has been slow to update on what this means for how they evaluate companies. ARR growth rate, logo count, and<em> </em>sales efficiency, the metrics that mattered most in 2018, are all gameable now in ways that were not really possible before AI-accelerated outbound and aggressive free-tier strategies lowered the cost of acquiring a customer who does not actually intend to stay. You can build a revenue line that looks healthy while the product never establishes genuine stickiness, and you can do it faster than ever. Cohort analysis is one of the few instruments that cuts through that particular illusion, because it is not asking whether customers started, it is asking whether they stayed and whether their staying was correlated with something the product was actually doing.</p><p>When you define your leading indicator before scaling, you also define exactly what you are defending when the competition arrives. You know the specific behaviors your GTM system was built to create. When a new entrant shows up with a shinier interface and a lower price, your entire customer success architecture is already oriented around the behaviors that predict loyalty. Companies that scaled without that clarity have no such anchor. They do not know what to defend because they never precisely defined what was working.</p><p><strong>THE MATRIX AND THE BOARD MEETING</strong></p><p><strong>The most underrated part of the framework is also the most practically powerful.</strong></p><p>By the time a company reaches Roberge’s Growth and Moat phase, the PMF question has been answered, the GTM playbook has been codified, and unit economics have been validated. The question has shifted from whether to scale to where and how fast. Roberge’s answer is a segmentation matrix across three dimensions: products, markets, and channels. Each cell gets evaluated on the same two criteria that governed the earlier stages, customer retention and unit economics. Cells where both are strong get scaled. Cells with thin or mixed evidence get a structured experiment. Cells where neither is present are shelved until one of the others is proven sufficient to finance the investigation.</p><p>The reason this matters in practice has more to do with boardroom dynamics than with strategic planning theory. Conversations about resource allocation, which markets to pursue, which channels to fund, and which product bets to make tend to drift toward whoever makes their case most forcefully rather than toward whoever has the most accurate read of what the data shows. The matrix does not eliminate that dynamic entirely, but it provides a structure that is harder to override by a narrative. When the question on the table is which cells are green and which are not, confidence is less useful than evidence.</p><p><em>When an investor pushes you to scale faster, the matrix lets you respond with evidence rather than anxiety.</em></p><p>Roberge’s pace prescription follows the same logic. The prescription is to add two salespeople per month for six months, monitor leading indicators, and adjust velocity based on whether the speedometer is holding. Over 65 percent of senior GTM hires at pre-Series B companies leave within 18 months,[7] and the most common reason is that they were brought on before the playbook had been proven sufficient to support them. Scaling a sales team before you have a reliable GTM motion is one of the fastest ways to consume a Series A without building anything that compounds.</p><p>The approach also gives you a defensible answer when the pressure to move faster comes, and it will come. You walk into the board meeting with your cohort retention trend, your LTV to CAC ratio, and your leading indicator data, and you lay out the pace at which you can grow without breaking the underlying economics. That is an argument built on evidence rather than ambition, which is the only kind of argument worth making to a board that will be around when the evidence becomes undeniable one way or the other.</p><p><strong>WHAT THIS REQUIRES OF YOU</strong></p><p><strong>The tools are not the hard part.</strong></p><p>The tools themselves are not complicated. The cohort chart lives in a spreadsheet. The unit economics are algebraic. The segmentation matrix has three dimensions and a decision rule. Anybody who can run a business can implement these in an afternoon.</p><p>What is genuinely hard is the honesty required to use them. Staying in the learning phase when your investors want to see top-line growth. Resisting the temptation to call two good cohort months proof of fit. Accepting that your retention numbers might be telling you something you do not want to hear and that the right response is to stay in the room with that information rather than hiring past it.</p><p>The segmentation matrix is particularly unforgiving on this front. Aggregate metrics let you obscure which parts of your business are actually working and which are being subsidized by the ones that are. Segment-level data does not. Precision is uncomfortable when some of the cells are red.</p><p>The alternative, though, is not faster growth. It is a faster failure. The companies that skip the learning phase do not outcompete those that do. They outspend them for a few quarters, then collapse under the weight of churn, CAC inflation, and a sales team trained on a playbook that never actually worked.</p><p>The startups still playing in year ten, with retention above ninety percent and a GTM motion that compounds without requiring ever-increasing capital, are almost uniformly the ones that were willing to stay in the learning phase long enough to learn something real. The five-year failure rate for startups is roughly 50 percent.[8] The businesses that make it past that threshold are not the ones that scaled fastest. They were the ones who understood, before scaling, what they were actually building.</p><p><strong>SOURCES</strong></p><p><strong>1 </strong><em>Shikhar Ghosh, Harvard Business School, cited in The Wall Street Journal: approximately 75 percent of venture-backed companies never return cash to investors.</em></p><p><strong>2 </strong><em>Marc Andreessen, “The Only Thing That Matters,” pmarchive.com, June 2007.</em></p><p><strong>3 </strong><em>Sean Ellis, “Using Product/Market Fit to Drive Sustainable Growth,” Medium / GrowthHackers, 2019. Ellis first proposed the 40 percent benchmark in 2009, derived from benchmarking nearly 100 startups.</em></p><p><strong>4 </strong><em>Maddyness UK / Smedvig Ventures: approximately 35 percent of startups that raise Series A fail to reach Series B. Post-Series B failure rate falls to around 1 percent.</em></p><p><strong>5 </strong><em>GitHub internal research on Copilot usage: developers completed tasks up to 55 percent faster with AI assistance.</em></p><p><strong>6 </strong><em>ClassicInformatics, “AI Development Statistics 2025”: low-code and no-code AI platforms reduced time-to-market by as much as 70 percent.</em></p><p><strong>7 </strong><em>ScaleWise, “Venture Graduation Crisis” (2025): over 65 percent of senior go-to-market hires at pre-Series B companies leave within 18 months.</em></p><p><strong>8 </strong><em>U.S. Bureau of Labor Statistics / Growthlist.co (2026): approximately 48 percent of private sector businesses fail within five years.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7b7380c144bb" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Disappearing Moat]]></title>
            <link>https://medium.com/@itaiboublil/the-disappearing-moat-ee26dd37a2df?source=rss-cd936f90d430------2</link>
            <guid isPermaLink="false">https://medium.com/p/ee26dd37a2df</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[vc]]></category>
            <category><![CDATA[startup]]></category>
            <dc:creator><![CDATA[Itai]]></dc:creator>
            <pubDate>Sun, 08 Mar 2026 08:26:58 GMT</pubDate>
            <atom:updated>2026-03-08T08:26:58.065Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/612/1*eZqdwqx8Rx4Uzu7DI2pUgw.jpeg" /></figure><p><strong>The Disappearing Moat</strong></p><p><em>In the age of large language models, a startup’s competitive advantage can vanish overnight — not because a rival outcompeted them, but because the platform they depended on absorbed them whole.</em></p><p>I.</p><h3>The Castle and the Contractor</h3><p>Warren Buffett made the concept of the economic moat famous through decades of straightforward metaphor. A great business, he argued, resembles a medieval castle: imposing, durable, surrounded by water wide enough to repel invaders and deep enough to drown them. He was describing companies like Coca-Cola, which commands the distribution relationships of a century and the psychological imprint of a brand taught to billions as synonymous with refreshment. He was describing toll bridges, where the geography eliminates competition entirely. He was describing, in essence, the ownership of something that cannot simply be replicated by anyone with capital and ambition.</p><p>The artificial intelligence startup economy of the past three years has produced a peculiar inversion of this logic. Thousands of companies have raised hundreds of millions of dollars to build what might generously be called a room inside someone else’s castle. They have constructed their products on top of models they do not own, using APIs that can change terms overnight, toward customers who are increasingly aware that the same capabilities can now be accessed directly, for less, from the same foundation that powers the startup itself.</p><p>This is not a niche observation. In February 2026, Darren Mowry, who leads Google’s global startup organization across Cloud, DeepMind, and Alphabet, publicly identified two categories of AI startups he believes are existentially at risk: LLM wrappers and AI aggregators. His warning was blunt. If your product is “really just counting on the back-end model to do all the work” and you are “almost white-labeling that model,” the industry has, as he put it, very little patience left for that. [1]</p><p><em>“What was a product in 2023 became a feature in 2024 and became an assumption in 2025.”</em></p><p>The question this essay sets out to answer is not whether those companies are at risk. That much has become obvious. The more interesting questions are: what does genuine defensibility actually look like in the age of large language models, what kinds of startups are building it, and, perhaps most urgently, where might venture capital be making bets that the market will not ultimately validate.</p><p>II.</p><h3>A Brief History of Being Eaten</h3><p>The pattern is now well established enough to have a graveyard. Consider the arc of the AI writing tools market. Jasper, founded in 2021 as Conversion.ai, reached $90 million in annualized recurring revenue within two years. It raised $131 million at a valuation that reflected genuine market enthusiasm. It was, for a moment, a rocket ship. Then OpenAI shipped ChatGPT, and suddenly the core value proposition — a well-prompted interface to a language model for marketing copy — was available directly from the source, for free, to anyone with an email address. Jasper’s valuation was internally marked down by roughly twenty percent within months. Its founders departed. The company pivoted toward enterprise, toward brand voice, toward features that ChatGPT does not prioritize. Whether that pivot is sufficient remains genuinely uncertain. [2][3]</p><p>Jasper is the most prominent casualty, but the pattern has repeated across categories with a consistency that suggests it is structural rather than incidental. Tome, which built AI-powered slide decks and attracted millions of users, watched its differentiation disappear when Microsoft embedded Copilot directly into PowerPoint. Meeting transcription tools — Otter.ai, Fireflies, Granola — collectively occupied a market they created, and then watched OpenAI bundle a Record Mode for meetings, with transcription and task extraction, directly into the ChatGPT desktop application. Companies building enterprise document search, like Glean and Guru, woke up to find that ChatGPT Business had begun indexing SharePoint, Dropbox, and Google Drive natively. [4]</p><p><strong>The Founder’s Letter</strong></p><p>One founder posted an open letter to OpenAI in September 2025 that has since become something of a canonical text in this discourse. He had built on the platform twice, he wrote, and twice a single product decision by OpenAI had erased his business overnight. The experience of having your startup eliminated by a feature release has a quality that does not map cleanly onto any prior category of business failure. You were not outcompeted. You were incorporated. [5]</p><p>This is the mechanism worth understanding clearly. When OpenAI or Anthropic releases a new capability, they are not launching a competing product. They are raising the floor of what the base model does. Every application built on that base must now differentiate against a higher baseline — one that includes the capability they previously supplied. The effect is cumulative. Each release cycle raises the floor. Each movement further compresses the viable space for thin applications. [6]</p><p>III.</p><h3>Why the VC Playbook May Be Wrong Here</h3><p>Venture capital has developed a coherent theory of why AI startups deserve large checks. The theory is plausible on its face. AI represents a platform shift comparable to mobile or cloud. Platform shifts produce large, durable companies. The winners of the mobile era were not the handset manufacturers but the application layer companies. Instagram, Uber, and Airbnb were all built on top of infrastructure they did not own and became genuinely valuable businesses. Therefore, building on top of large language models today is analogous to building on iOS in 2009.</p><p>The analogy contains a flaw worth examining. When Instagram was built on top of iOS, Apple was not in the photo-sharing business. The iPhone was the distribution mechanism, not a competitor to the applications running on it. Apple’s interests aligned with creating an ecosystem that made the device indispensable. The richer the app store, the more compelling the hardware. Apple had no reason to ship its own Instagram.</p><p>OpenAI and Anthropic occupy a structurally different position. They are simultaneously the infrastructure provider, the platform, and an increasingly aggressive product company competing in many of the verticals where their API customers operate. When OpenAI ships a meeting transcription feature, it is not a neutral infrastructure decision. It is a product decision that directly eliminates a category of API customers. The incentive structure is inverted relative to the mobile analogy.</p><p>A company that raised a Series A in 2023 on the basis of “we use GPT-4 better than anyone else” in a horizontal category has, by now, almost certainly seen its core differentiation commoditized. The more interesting question — which many investors appear to be asking later than ideal — is whether those companies built anything proprietary during the time that their initial feature advantage held. Did they accumulate data that only exists because their platform exists? Did they develop workflow integrations with switching costs? Did they enter regulatory environments that create structural barriers? [7]</p><p>According to late 2025 reports from Crunchbase and PitchBook, approximately fifty percent of all venture capital invested flowed into AI companies, with nearly seventy percent of that going into mega-rounds above $100 million. The winner-take-all dynamic at the top of the stack is being replicated in the investment strategy even for companies that are structurally unable to sustain winner-take-all outcomes. [8]</p><p>IV.</p><h3>What an Actual Moat Looks Like in 2026</h3><p>The failure mode is clear. The prescription is less so, partly because “build proprietary data” and “go vertical” have become as reflexive and underexamined as the wrapper strategies they are meant to replace. Both are correct as general directions, but neither is sufficient without a more precise understanding of what makes any particular instance of them defensible.</p><p>Andreessen Horowitz published a now-frequently-cited essay titled “The Empty Promise of Data Moats” that complicates the easy version of the data strategy. Their central observation is that data advantages erode in ways that are counterintuitive for founders to anticipate. The cost of adding unique data to a corpus tends to increase as scale increases, while the marginal value of incremental data decreases. First-mover data-collection advantages in most categories last between twelve and eighteen months before well-funded competitors catch up. The conclusion is not that data is irrelevant, but that data alone is rarely sufficient. [9]</p><p><strong>The Right Question</strong></p><p>The question is not whether you have data. It is whether competitors would need to replicate your entire business model to access the same data. Tesla’s self-driving dataset is defensible not because Tesla decided to build a good dataset, but because generating it requires deploying millions of vehicles with a specific sensor configuration over years of real-world driving. A competitor cannot purchase it. They cannot scrape it. To replicate it, they must become Tesla. [10]</p><p>The more durable version of defensibility involves what researchers describe as workflow-integrated data generation: building products where usage itself generates data that improves the product, in ways structurally inseparable from the specific business model. The feedback loop compounds over time. The company that entered a clinical workflow two years ago and has accumulated two years of physician corrections, edge case annotations, and outcome data possesses something that cannot simply be purchased by a later entrant, even if that entrant has access to better foundation models. [11]</p><p>Regulatory environments create a related but distinct category of defensibility. In financial services, healthcare, legal services, and defense, the barrier to deploying AI is not primarily technical. It is procedural and legal. Obtaining the certifications, clearances, and compliance frameworks required to operate in these environments takes time that cannot be compressed with capital. The regulatory stack becomes the moat, and the technical capability sits behind it.</p><p>V.</p><h3>The Fintech Case: When Regulation Becomes the Moat</h3><p>Consider the architecture of a company building AI tools for financial compliance and analysis. The surface-level description — “AI that helps analysts understand financial documents” — is exactly the kind of description that invites dismissal as a wrapper. The foundation models can summarize financial documents. The enterprise tools can process 10-Ks. Why would a standalone product survive?</p><p>The answer, for companies that have built correctly in this space, lies several layers below the feature description. Financial compliance work generates a specific category of data that has no public equivalent: the correction layer. Every instance where a compliance officer overrode the AI output, annotated an error, flagged a hallucination, or identified a misapplication of regulatory guidance. This correction layer is proprietary to the platform that generated it. It cannot be scraped. It cannot be purchased from a data broker. It exists because the platform was deployed in production in a regulated environment where errors have consequences and every output is reviewed by domain experts who are paid to find mistakes.</p><p>After eighteen months of deployment across a meaningful number of financial institutions, that correction layer begins to function as a qualitative differentiator that compounds with each additional institution. The model has seen the edge cases that only exist in production regulatory workflows. It has been corrected by compliance officers who know both the regulatory framework and the business context — whose corrections are most valuable precisely because they are so specific. [12]</p><p>The regulatory dimension adds a second layer. A financial institution that has integrated an AI compliance tool into its workflows has typically gone through a vendor procurement process that takes six to twelve months. The AI outputs may be subject to model risk management requirements under banking regulations, meaning the institution’s model risk officers have reviewed and signed off on the system. Replacing that system requires initiating another procurement cycle, another model risk review, <em>and </em>another period of parallel testing. The switching cost is not primarily emotional or contractual. It is procedural, embedded in compliance obligations that the institution cannot simply waive.</p><p>This architecture — where proprietary feedback data compounds over time within a regulatory framework that makes replacement structurally costly — is meaningfully different from a product differentiated primarily by interface design or prompt engineering. It cannot be replicated by a feature release from a foundation model provider, because the foundation model provider would need to operate in regulated financial environments as a compliance tool for years to accumulate the same dataset. That is not a product launch. That is a business model change.</p><p>VI.</p><h3>Where Startups Should Focus — and Where They Probably Should Not</h3><p>The most defensible startup categories share several characteristics. They operate in industries with genuine regulatory friction, where the barrier to deployment is procedural and legal rather than technical. They generate proprietary operational data that is inseparable from their specific workflows. They are building toward workflow integration that creates switching costs measured in compliance obligations rather than convenience. [13]</p><p>The categories that appear most exposed are roughly the inverse. Horizontal productivity tools in categories where the foundation models are actively expanding. Meeting intelligence, document generation, code assistance, and general business writing all sit in areas where the base capability is improving rapidly and where no regulatory friction protects the incumbent from platform encroachment. Companies in these categories that have not built proprietary data loops or deep workflow integrations during the window when their initial feature advantage held are now competing against a baseline that has risen to meet them.</p><p><strong>The Unsexy Compliance Stack</strong></p><p>Financial crime detection, healthcare prior authorization, regulatory filing preparation, insurance underwriting documentation: these categories are high-stakes enough that errors are unacceptable, specialized enough that foundation models are inadequate without domain fine-tuning, governed by regulatory frameworks that create procurement friction, and chronically underserved by technology investment because the sales cycles are long and the work is unglamorous. [14]</p><p>The companies building in these categories often cannot tell a compelling demo story. The product does not look impressive in a fifteen-minute pitch. The value is visible only to someone who has spent years inside the workflow and understands what was broken before and what the cost of errors actually is. This makes them less attractive to a certain kind of venture investor and more attractive to investors who understand that defensibility and demo quality are not positively correlated.</p><p>VII.</p><h3>What Is Still Uncertain</h3><p>The analysis above rests on several assumptions that deserve explicit acknowledgment, because intellectual honesty about what is not known is more useful than false confidence in a framework.</p><p>The first is the assumption that foundation model providers will continue to expand into application categories. This is currently the observed behavior, but it is not guaranteed to continue. OpenAI and Anthropic both face tension between their roles as infrastructure providers and product companies. Their enterprise API customers are also their primary source of near-term revenue. Aggressively eliminating those customers through feature bundling carries economic risk. There is a version of this market in which the foundation model providers, under pressure from enterprise customers and competitive dynamics, pull back from application-layer expansion and focus on infrastructure.</p><p>The second is the assumption that data moats are more durable than model quality advantages. This is currently well-supported by the evidence, but the emergence of synthetic data at scale complicates it. The research suggests this effect is limited by what is called the model collapse constraint, where recursive training on synthetic data degrades model quality, which preserves the value of authentic production data. But this is an active area of research, and the limits are not fully established. [15]</p><p>The third is the question of what happens to the venture capital already deployed into wrapper companies. A meaningful portion of the AI startup portfolio from 2022 through 2025 will likely not produce the returns that justified its valuations. Whether this produces a correction that is orderly or disruptive depends partly on how quickly limited partners and general partners recognize the structural issues, and partly on how many of these companies are able to execute pivots toward genuine defensibility during the time they have remaining.</p><p>VIII.</p><h3>The Question the Investors Should Be Asking</h3><p>The framework for evaluating AI startup defensibility in 2026 is, in essence, not complicated. It requires asking a single hard question: if OpenAI or Anthropic releases the relevant capability as a native feature of their platform next quarter, what remains of this business?</p><p>If the answer is “not much,” then the follow-up question is whether the company has been building proprietary assets — data, integrations, regulatory clearances, institutional relationships — during the window when its initial feature advantage held. If those proprietary assets are meaningful, the company may survive the platform encroachment with a defensible residual. If they are not, the company is in the process of being incorporated, and the remaining question is only about timing.</p><p>If the answer is “quite a lot,” then the analysis requires understanding precisely why. The reasons that hold up under scrutiny tend to share the properties described above: workflow-integrated data generation, regulatory friction, and switching costs that are procedural rather than purely contractual. The reasons that do not hold up tend to involve claims about brand, user experience, or superior prompt engineering — none of which constitute durable barriers to entry when the platform itself is the competitor.</p><p>Warren Buffett’s castle metaphor holds. What has changed is the address. The question is not whether the room is well-appointed. The question is what happens when the landlord decides to build the same room himself.</p><p><strong>Sources</strong></p><ol><li><em>Google VP warns that two types of AI startups may not survive</em> — TechCrunch, Feb 2026<br><a href="https://techcrunch.com/2026/02/21/google-vp-warns-that-two-types-of-ai-startups-may-not-survive/">techcrunch.com</a></li><li><em>Jasper Business Breakdown &amp; Founding Story</em> — Contrary Research, 2024<br><a href="https://research.contrary.com/company/jasper">research.contrary.com</a></li><li><em>The Coming ‘Collapse’ of the AI Startup Bubble</em> — Medium, May 2025<br><a href="https://optimaljon.medium.com/the-coming-collapse-of-the-ai-startup-bubble-703287bf8f42">optimaljon.medium.com</a></li><li><em>Will OpenAI Really “Kill” Start-Ups? A Deeper Look</em> — First AI Movers, Jun 2025<br><a href="https://www.firstaimovers.com/p/openai-kill-start-ups">firstaimovers.com</a></li><li><em>OpenAI, why are you killing my startup — for the second time?</em> — OpenAI Developer Community, Sep 2025<br><a href="https://community.openai.com/t/openai-why-are-you-killing-my-startup-for-the-second-time/1355491">community.openai.com</a></li><li><em>The AI Wrapper Problem: Why 80% of “AI Startups” Will Disappear by 2026</em> — Medium, Nov 2025<br><a href="https://medium.com/@Binoykumarbalan/the-ai-wrapper-problem-why-80-of-ai-startups-will-disappear-by-2026-6b4a873b0ad3">medium.com</a></li><li><em>Are LLM Wrapper Startups Really Unfundable?</em> — 1Mby1M, Oct 2025<br><a href="https://1m1m.sramanamitra.com/virtual-accelerator/no-equity/the-myth-of-unfundable-llm-wrapper-startups/">1m1m.sramanamitra.com</a></li><li><em>The Essential AI Startup Funding Guide 2025</em> — DealMaker, 2025<br><a href="https://www.dealmaker.tech/content/the-essential-ai-startup-funding-guide-2025-strategies-for-success">dealmaker.tech</a></li><li><em>The Empty Promise of Data Moats</em> — Andreessen Horowitz, Apr 2024<br><a href="https://a16z.com/the-empty-promise-of-data-moats/">a16z.com</a></li><li><em>Data Moats in the AI Era: What Actually Survives Foundation Model Disruption</em> — Ferguson Analytics, Jan 2025<br><a href="https://fergusonanalytics.com/blog/ai-data-moats/">fergusonanalytics.com</a></li><li><em>Castles, Moats, and Quicksand: The AI-Native Paradox</em> — DataDrivenInvestor, Nov 2025<br><a href="https://medium.datadriveninvestor.com/castles-moats-and-quicksand-the-ai-native-paradox-021ec89ec270">medium.datadriveninvestor.com</a></li><li><em>AI is Eating Enterprise SaaS</em> — Medium, Feb 2026<br><a href="https://medium.com/@rsaker/ai-is-eating-enterprise-saas-1259d352f193">medium.com</a></li><li><em>Defensibility in the Age of AI: Five Attributes Every Vertical Startup Needs</em> — Medium, Sep 2025<br><a href="https://medium.com/@elvia_perez/defensibility-in-the-age-of-ai-five-attributes-every-vertical-startup-needs-14f0d99eff60">medium.com</a></li><li><em>2024: The Year of AI Infrastructure Startups — Avoiding the Commoditization Trap</em> — Flybridge, Oct 2024<br><a href="https://www.flybridge.com/ideas/2024-the-year-of-ai-infrastructure-startups-avoiding-the-commoditization-trap-m5y6d">flybridge.com</a></li><li><em>Are Data Moats Dead in the Age of AI?</em> — V7 Labs, 2025<br><a href="https://www.v7labs.com/blog/data-moats-a-guide">v7labs.com</a></li><li><em>99% of AI Startups Will Be Dead by 2026 — Here’s Why</em> — Medium, Mar 2026<br><a href="https://skooloflife.medium.com/99-of-ai-startups-will-be-dead-by-2026-heres-why-bfc974edd968">skooloflife.medium.com</a></li><li><em>OpenAI 2024–2025: The Company That Won Everything and Lost Its Way</em> — DigiDai, Mar 2026<br><a href="https://digidai.github.io/2026/03/06/openai-2024-2025-valuation-products-organization-full-review/">digidai.github.io</a></li><li><em>AI Wrapper Startups Face Extinction: Google VP Warning</em> — ByteIota, 2026<br><a href="https://byteiota.com/ai-wrapper-startups-extinction-google-vp-warning/">byteiota.com</a></li><li><em>5 Ways Startups Avoid Getting Killed by OpenAI (and Succeed)</em> — Product Market Fit, Oct 2025<br><a href="https://www.productmarketfit.tech/p/5-ways-startups-avoid-getting-killed">productmarketfit.tech</a></li><li><em>There Is No AI Revolution</em> — Where’s Your Ed At, Feb 2025<br><a href="https://www.wheresyoured.at/wheres-the-money/">wheresyoured.at</a></li></ol><p>This essay raises structural questions about AI startup defensibility and venture capital allocation. It does not determine that specific firms have made bad bets or that any individual company is overvalued.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ee26dd37a2df" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Before You Let AI Touch Your Financial Books]]></title>
            <link>https://medium.com/@itaiboublil/before-you-let-ai-touch-your-financial-books-31faf49aa929?source=rss-cd936f90d430------2</link>
            <guid isPermaLink="false">https://medium.com/p/31faf49aa929</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[finance]]></category>
            <dc:creator><![CDATA[Itai]]></dc:creator>
            <pubDate>Mon, 17 Nov 2025 09:57:16 GMT</pubDate>
            <atom:updated>2025-11-17T10:52:52.421Z</atom:updated>
            <content:encoded><![CDATA[<h4>Why finance teams need to think twice before connecting their data to generic AI platforms.</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*FxEtibw6Kw58QdoNQMrtpw.jpeg" /><figcaption>Before You Let AI Touch Your Financial Books</figcaption></figure><p>Before you let AI touch your financial books, take a moment to understand what’s happening under the hood. <a href="https://www.tenable.com/blog/hackedgpt-novel-ai-vulnerabilities-open-the-door-for-private-data-leakage">Recent research has revealed</a> that some of the most popular AI models — the same ones that power thousands of “AI for finance” tools — carry vulnerabilities that could quietly expose your company’s most confidential data.</p><h3>Let us give a bit of background first</h3><p>When we founded <a href="http://www.finnx.ai/">FINNX</a>, everyone around us was moving fast (and they still do). It was the height of the AI wave. Startups were spinning up products in weeks, plugging into large language model APIs, and launching new “AI accountants” almost overnight.</p><p>Investors loved it, and the market rewarded speed by paying for these tools. So far, so good.</p><p>But finance isn’t marketing, customer service, Sales, or HR. It’s the beating heart of a company; the bloodstream that carries its most sensitive information: revenues, payrolls, forecasts, supplier terms, tax positions, and audit notes.</p><h3>The moment the risk became visible.</h3><p><a href="https://www.tenable.com/blog/hackedgpt-novel-ai-vulnerabilities-open-the-door-for-private-data-leakage">Tenable’s investigation</a> revealed a quiet truth about the modern AI ecosystem: that these systems, powerful as they are, can also be manipulated in ways few people outside the security community (or inside the security community!) fully grasp.</p><p>The report detailed a series of experiments showing that large language models could be tricked into leaking private data through what researchers called “prompt injection.”</p><p>The idea is deceptively simple. By embedding hidden instructions into text or code that an AI later reads, an attacker can make the model do things its creators never intended. Things like revealing information from a previous conversation or sending internal data to an external site.</p><p>Even more worrying was the discovery that the attack didn’t require users to click on anything. In some cases, a model that could browse the web or fetch documents might stumble upon a malicious webpage, ingest a line of hidden text, and execute those instructions automatically.</p><p>The report demonstrated how an attacker could embed instructions deep inside documents, hide them in code blocks or comments, and cause the model to “remember” them even after the original session ended. A seemingly harmless PDF or a standard vendor invoice could contain a digital Trojan horse. And once the model’s memory was compromised, the manipulation could persist indefinitely.</p><p>It was chilling to read not because we didn’t expect vulnerabilities to exist, but because it confirmed a thesis we had held all along: general-purpose AI systems were never designed for the kind of data finance teams handle.</p><p>A second problem concerns the online training and maintenance of these 3rd-party LLM models. Part of the process is ‘tokenization’, the breaking down of text into pieces. Unless the service provider has followed heroic levels of isolation, the system will record strings such as account numbers, contract numbers, and corporation names, even though the text itself isn’t being trained on them. This opens the door for prompts such as “<em>Have you ever heard of [Stealth Project XYZ], Answer with Yes or No</em> over an unrelated model and still get competitive intelligence out of the system!</p><h3>What finance leaders need to understand</h3><p>For finance executives, the implications are profound. Most Finance leaders today are being asked the same question: <em>“When are we bringing AI into finance?</em>” The pressure is real, and we at FINNX understand it more than anyone. There is a need to modernize and keep up with the evolving technology landscape. But there’s a difference between modernizing processes and outsourcing judgment.</p><p>Finance teams deal in verifiable, reconcilable, and traceable truth. That means every transaction, every number, and every data point must be auditable. The moment you introduce an AI model that cannot be fully audited, that “truth” becomes vulnerable.</p><p>In most of today’s AI landscape, the architecture looks like this: your company’s data (perhaps uploaded invoices, P&amp;L statements, or budget forecasts) flows through a model hosted on someone else’s servers, trained on data you’ll never see, interacting with external sources you’ll never control.</p><p>The model might even browse the web or retain memory across sessions to “improve user experience.” On paper, this sounds convenient. In practice, it means your data is being exposed to the very kind of vulnerabilities <a href="https://www.tenable.com/blog/hackedgpt-novel-ai-vulnerabilities-open-the-door-for-private-data-leakage">Tenable’s team demonstrated.</a></p><p>Finance data is not just another dataset. It is, arguably, the single most valuable category of data an organization possesses. It reveals your cost structures, revenue dependencies, cash positions, and future assumptions; all the information competitors would pay to see and regulators expect you to protect.</p><p>When Tenable’s researchers concluded that “prompt injection is a known issue with the way large language models work, and it will probably not be fixed systematically in the near future,” it became clear that for finance, relying on these shared systems was not just risky. It was reckless.</p><h3>Building trust from first principles</h3><p>So we decided to build differently. We started from the premise that financial data must remain sovereign. In our world, that means every company’s environment is isolated — their data, their model instance, their processing pipeline. There are no shared memory pools across clients, no cross-tenant interactions, and no browsing of external content.</p><p>We designed our models so they can’t wander off the digital equivalent of the balance sheet. They don’t fetch live URLs or browse websites. They operate within controlled boundaries, with every input and output traceable, every document scanned before ingestion, and every memory event logged for audit.</p><p>Building this kind of system means saying no to shortcuts. It means we can look a CFO in the eye and say: yes, this system can analyze your invoices, your ledgers, your statements, and do your accounting — and we can prove exactly how it did it.</p><h3>What Tenable’s research taught us about the future of AI</h3><p>Reading through the “HackedGPT” report, what stood out most wasn’t just the technical depth of the vulnerabilities. It was how familiar the pattern felt.</p><p>Every industry has its version of this story. In finance, we’ve seen it before: in spreadsheets shared without controls, in ERP migrations rushed to meet deadlines, and in cloud systems adopted without proper governance.</p><p>AI is repeating that pattern on a global scale. The drive to adopt is outpacing the ability to secure. Tenable’s research simply held up a mirror. It showed how quickly a well-meaning AI integration could turn into a data liability.</p><p>The attack vectors they described are not science fiction. Imagine a financial controller asking an AI tool to summarize a supplier contract, only for the model to pull a malicious web snippet embedded with instructions to send parts of that contract text to an external server.</p><p>Imagine an analyst uploading a CSV of transactions that contains hidden code the model reads as an instruction to expose prior answers. Imagine a memory-enabled AI that remembers those instructions the next time someone from your company logs in.</p><p>These are not theoretical scenarios. They are the reason we took a different path.</p><h3>The anatomy of a safer system</h3><p>Building secure AI for finance requires more than encryption and access control. It requires designing a system that assumes bad actors exist, that vulnerabilities will be found, and that trust must be continuously verified.</p><p>In the FINNX architecture, every client instance exists in isolation. Data is processed within a sealed environment; outputs are monitored, contextual memory is limited to what’s operationally necessary, and external calls are restricted. We removed open browsing entirely, because the cost of one poisoned web source isn’t worth the convenience.</p><p>When a company uploads a document, our system doesn’t just read it; it inspects it. We built filters that detect hidden prompts or instructions, the same technique Tenable’s research showed could be used to compromise models. Any content that looks suspicious is quarantined.</p><p>Our memory systems are governed by explicit policy. Nothing persists without user consent, and every retention event is logged for review. In practice, this means that what a finance team tells our system today doesn’t bleed into what another team or client might see tomorrow.</p><p>Perhaps most importantly, our models aren’t trained to know everything. They’re trained to understand the specific language and logic of finance. It doesn’t need to know who won the Super Bowl or how to write a sonnet; it needs to know what deferred revenue means, how to trace a journal entry, and when an exception in an invoice requires human review.</p><h3>What this means for CFOs right now</h3><p>If you are a CFO or finance leader evaluating AI tools, <a href="https://www.tenable.com/blog/hackedgpt-novel-ai-vulnerabilities-open-the-door-for-private-data-leakage">Tenable’s research</a> should give you pause, not fear, but pause. The question isn’t whether AI will enter finance; it already has. The question is under what conditions, and with what safeguards.</p><p>The temptation is always to move fast. The board wants efficiency, and the CEO wants innovation. But finance has never been the place for haste. It’s the department of reconciliation, validation, and accountability. The same principles must apply to AI adoption.</p><p>When you evaluate vendors, ask them not just about what their AI can do, but how it does it.</p><ul><li>Can they tell you where your data goes?</li><li>Can they show you how memory is managed?</li><li>Can they prove that their model won’t browse external content or execute hidden instructions?</li><li>Can they isolate your data from other clients’?</li></ul><p>If the answer to any of those is vague, it’s worth reconsidering the partnership. Because in the end, the future of AI in finance won’t belong to whoever moves first. It will belong to whoever builds trust that can’t be hacked.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=31faf49aa929" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Antisemitism and Anti-Israel Sentiment: Divergence, Conflation, and Escalation]]></title>
            <link>https://medium.com/@itaiboublil/antisemitism-and-anti-israel-sentiment-divergence-conflation-and-escalation-5c87b1b2d670?source=rss-cd936f90d430------2</link>
            <guid isPermaLink="false">https://medium.com/p/5c87b1b2d670</guid>
            <category><![CDATA[gaza]]></category>
            <category><![CDATA[israel]]></category>
            <category><![CDATA[israel-palestine-conflict]]></category>
            <dc:creator><![CDATA[Itai]]></dc:creator>
            <pubDate>Thu, 07 Aug 2025 08:29:11 GMT</pubDate>
            <atom:updated>2025-08-07T08:29:11.442Z</atom:updated>
            <content:encoded><![CDATA[<p>The distinction between antisemitism and criticism of the Israeli government has become one of the most contested and consequential debates in global politics. While both forms of discourse have long histories, their increasing overlap in recent years has raised urgent questions about political speech, civil rights, international policy, and Jewish safety worldwide.</p><p>This article explores the definitional boundaries between antisemitism and anti-Israel sentiment, the evidence for recent escalation, Israel’s potential responses, and the downstream consequences for Israeli society and the global Jewish community.</p><h4><strong>Defining the Lines: Antisemitism versus Anti-Israel Sentiment</strong></h4><p>Antisemitism refers to hostility, prejudice, or discrimination against Jews as a religious, ethnic, or cultural group. It has manifested historically in theological condemnation, racial pseudoscience, conspiracy theories, violent pogroms, and state-sponsored genocide. The International Holocaust Remembrance Alliance (IHRA) definition, adopted by dozens of governments and institutions, includes examples of modern antisemitism such as Holocaust denial and holding Jews collectively responsible for actions of the State of Israel.</p><p>In contrast, anti-Israel sentiment refers to criticism of the policies, actions, or foundational ideology of the State of Israel. Such criticism can range from opposition to settlement expansion or military tactics to wholesale rejection of Zionism as a national movement. Anti-Israel positions can emerge from various ideological roots — human rights frameworks, postcolonial theory, or Islamist politics — and are not inherently antisemitic.</p><p>Yet in practice, the line between the two is frequently blurred. Critics of Israel sometimes use antisemitic tropes or target Jewish individuals or institutions unrelated to Israeli policy. Conversely, supporters of Israel often label even harsh but legitimate criticism as antisemitic, creating a chilling effect on discourse and conflating political dissent with racial hatred.</p><h4><strong>Escalation: Perception and Reality</strong></h4><p>In the past two years, accusations of antisemitism have surged globally, particularly in the context of escalated conflict between Israel and Palestinian groups, including the <a href="https://en.wikipedia.org/wiki/Sheikh_Jarrah_controversy">2021 Sheikh Jarrah protests</a>, the <a href="https://en.wikipedia.org/wiki/2021_Israel%E2%80%93Palestine_crisis">May 2021 Gaza crisis,</a> and most dramatically, the Hamas attacks of October 7, 2023, and Israel’s military response in Gaza. These events have acted as catalysts, igniting protests, boycotts, and political ruptures worldwide.</p><p>Empirical evidence of escalation is complex. Organizations such as the Anti-Defamation League (ADL), the Community Security Trust (CST), and governmental bodies in France, Germany, and the United States report a significant rise in antisemitic incidents, particularly in the wake of the 2023 conflict. These include physical assaults, synagogue vandalism, and harassment of visibly Jewish individuals. University campuses in the U.S. and Europe have become flashpoints, with pro-Palestinian demonstrations occasionally accompanied by antisemitic slogans or threats, while Jewish students report rising fear and isolation.</p><p>At the same time, governments and institutions have responded with increasing severity to perceived anti-Israel speech, often under the guise of combating antisemitism. In some cases, legitimate protest has been silenced or criminalized, especially when framed as anti-Zionist. Thus, escalation is occurring on both fronts: a rise in actual antisemitic acts, and an expansion of the definitional scope of antisemitism to encompass more and more anti-Israel critique.</p><h4><strong>What Can Israel Do? And Can It Actually Do It?</strong></h4><p>Israel faces an unenviable balancing act. It seeks to ensure the safety and legitimacy of its Jewish citizens worldwide, protect its international alliances, and maintain domestic security. However, its military policies — especially in Gaza — and its approach to the Palestinian question more broadly, are central drivers of global criticism.</p><p>Israel has two broad strategic paths ahead:</p><h4><strong>Defensive Posture with Global Outreach:</strong></h4><p>In this path, Israel would continue its current military and political strategies while doubling down on global advocacy. This includes promoting the IHRA definition of antisemitism, strengthening ties with Western governments, investing in diaspora outreach, and branding critics as antisemites to delegitimize opposition.</p><p>This path may yield short-term diplomatic protection and solidarity from allies. However, it risks alienating liberal Jewish communities, exacerbating academic and cultural boycotts, and increasing polarization. Over time, it may also entrench Israel’s image as a state above reproach, which could backfire if geopolitical tides shift.</p><h4><strong>Strategic Reassessment and Political Reform:</strong></h4><p>In this alternative path, Israel would undertake meaningful political reform — revisiting settlement policies, advancing viable peace frameworks, and more actively differentiating itself from the far-right elements within its own government. It would also participate in nuanced international discussions about antisemitism without equating it wholesale with anti-Zionism.</p><p>This path could rehabilitate Israel’s global image and reintegrate alienated diaspora voices. However, it is politically perilous domestically, where hardline coalitions dominate, and any concession is framed as weakness. It also depends on the feasibility of re-engagement with Palestinian leadership and the willingness to accept painful compromises.</p><p>Whether Israel <em>can</em> choose the second path depends not only on external pressures but also on internal political will — something currently in short supply.</p><h4><strong>Impact on the Average Citizen</strong></h4><p>The consequences of these two paths are not abstract — they directly affect the daily lives of both Israelis and Jews around the world.</p><p>If Israel continues the first path, Jewish communities abroad may find themselves more vulnerable to scapegoating, particularly in polarized societies. Israeli citizens may enjoy a sense of nationalist pride and security, but may also become more isolated diplomatically and economically over time. Civil discourse inside Israel may become more constrained, with dissenters increasingly branded as traitors or “self-hating Jews.”</p><p>On the other hand, a reformist path could reduce international hostility and revive diplomatic partnerships, but would likely trigger domestic upheaval. Israelis may feel insecure during transitional periods, and diaspora Jews could find themselves caught between pride in reform and fear of backlash.</p><p>For both paths, the average citizen bears the psychological and material cost: fear, uncertainty, and a hardening of identity politics. Jewish students abroad may be asked to answer for Israeli actions they have no control over. Arab citizens of Israel and Palestinians in the territories may continue to suffer under occupation or internal discrimination. Israelis themselves may experience increased violence or societal fragmentation, depending on the decisions made by their leadership.</p><h4><strong>A Tipping Point or a Loop?</strong></h4><p>The last two years have seen a volatile convergence of rising antisemitism, surging anti-Israel activism, and reactive policymaking. The world is at a discursive and ethical tipping point: one where clarity of terms, openness to dialogue, and principled political courage are essential.</p><p>Israel has the agency to influence this trajectory, but not unilaterally. Whether it leans into defensive isolation or opens itself to reform, each path brings consequences that ripple far beyond Jerusalem. For the average citizen — Jewish or Palestinian, Israeli or American, ally or critic — the stakes are deeply personal and increasingly urgent.</p><p>What remains to be seen is whether the global community can hold space for both legitimate criticism of a state and the fierce defense of a people without collapsing the two into a binary of blame.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5c87b1b2d670" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Kahanism as the End of Zionism: An Ideological Merger and Israel’s Moral Collapse After October 7]]></title>
            <link>https://medium.com/@itaiboublil/kahanism-as-the-end-of-zionism-an-ideological-merger-and-israels-moral-collapse-after-october-7-b83b2f9ba36f?source=rss-cd936f90d430------2</link>
            <guid isPermaLink="false">https://medium.com/p/b83b2f9ba36f</guid>
            <category><![CDATA[israel-palestine-conflict]]></category>
            <dc:creator><![CDATA[Itai]]></dc:creator>
            <pubDate>Tue, 05 Aug 2025 04:04:43 GMT</pubDate>
            <atom:updated>2025-08-05T04:04:43.411Z</atom:updated>
            <content:encoded><![CDATA[<p>No political shift happens in a vacuum. Every ethical deterioration is the result of public discourse, ideological choices, and often the deliberate or unconscious distortion of collective awareness.</p><p>Since its founding, the State of Israel has maintained a fragile balance between democratic Zionist ideals and opposing internal currents that, until recently, were relegated to the margins.</p><p>But in recent years, and more sharply since October 7, that balance has collapsed.</p><p>Kahanism, a racist and violent ideology once banned from Israeli politics, has not only returned but fused with the Zionist institution.</p><p>The result is a transformation of Zionism into something entirely different, something that in structure and tone resembles fascist regimes of the past.</p><p>Zionism, born as a national liberation movement, sought to provide the Jewish people with sovereignty, protection, and a homeland.</p><p>It was not monolithic in ideology, but it carried with it universal principles: a commitment to justice, a vision of a just society built by a persecuted people returning home.</p><p>Kahanism, by contrast, is rooted in the rejection of those very ideals. It offers no tolerance for the other, denies equal rights to non-Jews, and preaches Jewish supremacy; not as a lesson from history, but as a dangerous internalization of it.</p><p>What has happened in Israel since October 7 is not just a security crisis, it is also a moral one.</p><p>In the wake of national trauma, when a nation grieved and feared for its future, its leadership saw an opportunity.</p><p>A prime minister, already abandoned by the public, clung to power not through unity or vision, but through division, incitement, and survival at any cost.</p><p>Hostages were deprioritized. Democracy was weakened. And the Kahanist ethos crept in, masked as patriotism, but rooted in vengeance and hatred.</p><p>From there, the ideological merger was inevitable. When Kahanism fuses with Zionism, the result is not a stronger national movement. It is, instead, a moral distortion.</p><p>A national idea held hostage by dehumanization and religious fanaticism ceases to be a movement of liberation. It becomes a vehicle of oppression.</p><p>What once was called Zionism is now something entirely different. And yes, it bears uncomfortable resemblance to ideological patterns we know from history. Not in deeds, not in genocide, but in its architecture: the erosion of rights, ultra-nationalism, moral collapse, and the use of fear as a tool of governance.</p><p>This is not a comparison to the Holocaust. Let us be clear. But it is a warning, just as relevant today as it was then.</p><p>Nazi Germany, in its final days, had to acknowledge that the idea it had built itself upon had led to moral and physical ruin. Only then could the German people begin to rebuild something new.</p><p>Israel is not yet at that point. It continues to sink deeper, refusing to admit its mistake, clinging to a version of Zionism that has already decayed from within. Every attempt to “fix“ the situation without admitting its failure has only pulled the country further down.</p><p>And the world sees it. In an era where there is no buffer between camera and conscience, no country has the privilege of hiding its actions.</p><p>The destruction in Gaza is documented. The statements of Israeli politicians — calls for revenge, expressions of ethnic superiority — are broadcast and archived.</p><p>The suppression of dissent, the assault on democratic institutions, the normalization of hate speech: all of these have stripped Israel of the international legitimacy it once enjoyed. Kahanism didn’t just return, it went mainstream. And the world is rightly rejecting it.</p><p>Where does this leave Israel?</p><p>The country stands at a critical crossroads. The current government, deeply entrenched in its Kahanist mindset, will not step down voluntarily.</p><p>It no longer sees governance as a public service but as ownership. But the international community is beginning to respond. Sanctions. Isolation. Civil boycotts. Economic collapse. These are no longer distant possibilities. They are the likely consequences of moral failure. Perhaps only through collapse can Israel begin to rebuild.</p><p>That may be the price. The cost of ideological fusion. Of denial. Of a political gamble that failed, and took with it the last remnants of integrity, leadership, vision, and international trust.</p><p>And yet, even now, there is hope. Not for the rehabilitation of Zionism as we knew it, but for something new.</p><p>Democratic. Equal. Grounded not in fear or supremacy, but in shared humanity.</p><p>That future depends on us, and on our willingness to begin again.</p><p>(Photo: channel 12, Israel, interview., June 2024)</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/605/1*8WhLi-XlwFOKr2IxDJqWMw@2x.jpeg" /></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b83b2f9ba36f" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Part 5: #EBC2017 — Climbing to Namche Bazaar and Getting Hypothermia]]></title>
            <link>https://medium.com/@itaiboublil/part-5-ebc2017-climbing-to-namche-bazaar-and-getting-hypothermia-c254a73e9213?source=rss-cd936f90d430------2</link>
            <guid isPermaLink="false">https://medium.com/p/c254a73e9213</guid>
            <category><![CDATA[travel]]></category>
            <category><![CDATA[trekking]]></category>
            <category><![CDATA[everest]]></category>
            <dc:creator><![CDATA[Itai]]></dc:creator>
            <pubDate>Wed, 19 Apr 2017 00:00:00 GMT</pubDate>
            <atom:updated>2017-04-20T03:16:19.236Z</atom:updated>
            <content:encoded><![CDATA[<p>This is one in a series of posts about my recent Trek to Everest Basecamp (EBC). If you wish to read the posts you can access them at <a href="http://www.itaiboublil.com/category/ebc2017/">#EBC2017</a></p><p>March 21, I woke up after sleeping for almost 10 hours. My body was exhausted from yesterday’s trek, so you can certainly say I collapsed on my “bed.” I say “bed” because basically it was a wood board with a thin blanket on it that served as a “mattress.” The night was very cold around -6c degrees. I slept with my clothes on, including my gloves and wool hot, and I still found it hard to control my shaking. The walls in the room were made out of thin wood, and thus the temperature in the room was almost identical to the outside temperature; quite cold. The shower didn’t really work, and hot water wasn’t available — which made brushing my teeth quite uncomfortable (freezing water). Prior to my 10-hour sleep, I managed to rearrange my backpack a bit for better weight distribution. One of the mistakes I made on the first day is that I carried the heavier bag on me instead of giving it to the porter. I now managed to reduce the weight of the backpack from 12.5kg to about 6kg, which made the trekking much more comfortable.</p><p>I was up at 5:00am today. I went to sleep around 19:00, and thus was well rested when I woke up. After washing my face with ice-cold water, and eating a Cliff bar to get my body going, I stepped outside to get some fresh air. Breakfast was scheduled at 7:00am, so I had a few hours to “kill” before I could have my two eggs, potatoes, and two slices of bread (a common breakfast throughout the trek). The view was spectacular! It was cloudy the day before so I couldn’t really see the big icy mountains that revealed themselves in the morning. I stayed outside for almost an hour to watch the sun rising on these big mountains, and felt appreciative for the opportunity to be where I was at that moment.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Ap7U0d20iC3XRMTJ5eRwHA.jpeg" /><figcaption>Yaks marching to Namche Bazaar</figcaption></figure><p>Shortly after breakfast, we started our trek to <a href="https://en.wikipedia.org/wiki/Namche_Bazaar">Namche Bazaar</a>. I was warned by others that this will be the toughest climb of the entire trek, but nothing really could have prepared me for it. We climbed for almost 1,000 meters — about 6 hours of which ~3 hours of it was straight uphill in one of the most challenging activities I have ever completed in my entire life…but wait, I will get to it.</p><p>Coming back to “shortly after breakfast.” Shortly after breakfast we grabbed our gear and headed toward Namche. I felt a bit better walking with my light backpack, and that feeling put me in a better mood. The only thing I was struggling with was the breathing. I was still trying to figure out how to breathe at this altitude, which was very difficult. The best way I can describe it is this: imagine that you’re holding your breath to the maximum of your lungs capability and right when you feel that sharp burning pain in your lungs (before you die from lack of oxygen), you open your mouth to inhale as much oxygen as possible. THAT is the feeling I got every time I tried to walk a bit faster, or if I skipped half a breath. It is a very uncomfortable feeling that takes some time to get used to.</p><p>The first 3 hours of the trek were not as challenging as the last 3 hours, but were challenging enough for me to struggle with my breathing. We climbed up and down, crossed bridges, and witnessed the most heavenly views I have ever come across! My senses were overwhelmed with the view around me to the point I could barely speak. I kept walking with a big smile on my face, wondering where I was. The “easy” part ended after three hours or so, and I was now facing the toughest part of today’s trek — climbing up to<a href="https://en.wikipedia.org/wiki/Namche_Bazaar"> Namche Bazaar</a>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*WboDWVvXTqUqRnKlt1zkBQ.jpeg" /><figcaption>Walking to Namche Bazaar</figcaption></figure><p>The second half of the trek started with a steep uphill climb, followed by steep uneven stairs made out of uneven rocks, and continued this way for the next 3 hours or so. Each step was a big lunge climbing uphill that repeated itself thousands of times. Shortly into the climb, I started feeling the mental part taking over my body. I felt out of shape (even though I am not) as breathing at this altitude was something new to me. The trail was relatively busy with climbers, porters, and locals who were all taking the same route to Namche. Being the competitive person that I am, I tried to compete with them by keeping the same pace and soon discovered that it was a crucial mistake as I kept losing, which added to the demoralizing feeling I already struggled with. Clearly, things were not going as I planned, I thought to myself, and it was the first time I actually broke down. I sat down on a big rock facing the valley below in order to recollect my thoughts. Bad thoughts were coming to my mind, anywhere from “what am I doing here?” to “I cannot do this.” I felt frustrated, demoralized, and the thought of quitting made me want to cry (I don’t like quitting). It is at this moment I (re)discovered a new “muscle” I have forgotten I had — the mental muscle.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*lVv9ebtiJVMrmjUBQpXe9Q.jpeg" /><figcaption>Sitting on a rock and getting my thoughts together.</figcaption></figure><p>As I was sitting on the big rock facing the green valley below, strong cold wind blowing on my face, I realized that my struggle is not physically but mentally. I needed to come up with a plan to climb this monstrous challenge — a plan I later called “the short goals” plan. It was a simple plan, but it helped me go through the entire journey that was ahead of me. In a nutshell, I set myself short goals as I was climbing up. A short goal was usually a big rock 5 meters ahead of me that I needed to reach. Once I reached that goal, I would pick another short goal. I kept repeating this process hundred of times until I actually reached the final destination for the day. The second part of the plan was never to look forward, or up. Whenever I picked a goal, I put my head down and kept walking until the “goal” crossed my eyes. This is the point when I looked up to pick a new short goal. It was a simple plan, but it helped me keep my focus. It disconnected me from the negative voices inside my head (the “let’s quit this” voices) and helped me organized my mind. I felt at peace after coming up with this plan — though very tired physically — and knew then that I have discovered the secret that will take me all the way to Everest Base Camp.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*lGYLq-mwtf_W6MzPXQtsjw.jpeg" /><figcaption>Namche Bazaar</figcaption></figure><p>By now I have climbed for almost 3 hours. My legs were beyond tired, I was breathing heavily, my eyes were red because of the dried salty sweat, and overall I was completely depleted. When I finally entered the gates of Namche Bazaar, I felt as if I wanted to cry out of joy. I never thought I will reach there, but I did, and it felt super good! When I finally arrived at our hotel (<a href="https://www.tripadvisor.com.sg/Hotel_Review-g1203014-d8725001-Reviews-The_Nest_Hotel-Namche_Bazaar_Khumbu_Sagarmatha_Zone_Eastern_Region.html">The Nest</a>) and sat down at the table in the dining room, a strange thing happened to me that I have never experienced before. My body has started to shake uncontrollably, and I felt very cold. In addition, my heart rate dropped to about 45 bpm and I felt a bit confused. I knew immediately what was happening and I needed to act fast! I was getting a hypothermia. The temperatures went from hot to cold, and vice versa, during the climb (strong winds will drop the temperature instantly), which encouraged sweating (hence the hypothermia). In addition, I exhausted my body during the climb and forgot to replenish the carbohydrates (glycogen levels) I lost during the activity. My body was getting into a shock, and I needed to act super fast if I wanted to continue my adventure and not be flown back to Kathmandu in a helicopter.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*yhD6PexT8_cT0YnSMQyFaQ.jpeg" /><figcaption>Namche Bazaar Airfield</figcaption></figure><p>Shivering, weak, and disoriented, I immediately went back to my room and grabbed two Cliff bars which I shoved down my throat in seconds. I then took off my clothes and got into a very hot shower, followed by wearing all the hot clothes I could possibly wear. I went back to the dining room and ordered a big bowl of soup and some hot tea, and that seemed to help. My heart rate slowly climbed up to about 65 bpm, and the shaking slowly started to fade away. I was in the safe zone again and seemed to overcome this awful and scary experience. I kept sitting in the dining room for another two hours or so until I went back to my room and fell asleep for about 12 hours. I was tired after the long climb and the minor shock my body has gone through, so I think sleeping was the only thing my body could think of. I never slept 12 hours in my entire life, so that was a strange feeling in itself.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*mrOZ2JbIhbJn3Sy0WErUlA.jpeg" /><figcaption>Sunrise the following morning in Namche Bazaar</figcaption></figure><p>We stayed in Namche Bazaar for two days (acclimatization day), which was great because I needed the rest. On the second day, we climbed to Hotel Everest View to view Mt. Everest for the first time, and also prepared ourselves to the next part of the trip: climbing to Tengboche.</p><p><strong>To be continued…</strong></p><p>This is one in a series of posts about my recent Trek to Everest Basecamp (EBC). If you wish to read the posts you can access them at <a href="http://www.itaiboublil.com/category/ebc2017/">#EBC2017</a></p><p><em>Originally published at </em><a href="http://www.itaiboublil.com/part-5-ebc2017-climbing-to-namche-bazaar-and-getting-hypothermia/"><em>www.itaiboublil.com</em></a><em> on April 19, 2017.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c254a73e9213" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>