Today, we continue my annual tradition of making ten predictions for the coming year while reviewing my ten predictions for the prior one.
As always, I’ll remind you of some disclaimers: (a) don’t view this as investment advice or anything resembling it, (b) while I generally avoid politics on my blog, I make an exception in these posts because I like to start macro and then zoom in, (c) remember that I do this for fun and entertainment, so please take it as such, and (d) my self-scoring for the previous year is always generous [1].
This is my twelfth annual predictions post. Let’s begin.
Review of 2025 Predictions
1. America gets what we deserve. Hit. I predicted a more brazen, more interest conflicted, and less constrained Trump — and scored a hat trick. The more interesting question today, given that Trump has increasingly diverged from his official platform (if less so from his unofficial one), is whether people are getting what they think they voted for and if they are happy with the results? Some visible MAGA types, from MTG to the QAnon Shaman, are saying no. Either way, we voted for Trump and we got him. I believe that will one day serve as an enduring reminder that character matters in leader selection.
2. The broligarchs enjoy their 15 minutes of fame. Hit. I almost scored this a miss because some of the tech bros have enjoyed more than their 15 minutes. But I’ll credit myself with the hit because Elon Musk was the poster child for this group and, with the DOGE fiasco, he exemplified what I was predicting.
Many Silicon Valley leaders appear to have aligned themselves around four priorities: (a) a more favorable exit environment, including a revival of big-tech M&A after an FTC posture that had largely stymied it; (b) crypto, which — despite substantial effort to convince myself otherwise — I continue to view as suspect [2]; (c) a light-touch regulatory framework for AI; and (d) pro-growth energy policy, given AI’s heavy dependence on power.
I foresaw much of this thanks to this December 2023 quote from Ben Horowitz [3]:
We are non-Partisan, one issue voters: If a candidate supports an optimistic technology-enabled future, we are for them. If they want to choke off important technologies, we are against them.
Thus far, the administration has largely delivered, so the digerati may well feel vindicated, though I could do without the sane-washing [4]. In the coming year, we will learn more about precisely what they traded in this Faustian bargain.
3. The startup ecosystem purge continues. Hit. I got lucky here because I meant to say “accelerates,” but wrote “continues” — which is more indicative of what happened. In 2023, 769 startups shut down; in 2024, that figure rose to 966; and Carta’s commentary suggests the 2025 total will fall between 1,000 and 1,100. Interestingly, per SimpleClosure, startups shut down later in their lifecycle compared to 2024 [5].
But an ecosystem purge is also about M&A, not just shutdowns. Per Berkery Noyes, 2025 M&A volume was up 66% from $253B to $420B, with deal count up 8% from 1,992 to 2,152.
The point was that many startups were going to hunker down in 2025, extending cash runways to buy time until the return of a healthier exit environment. That return did not happen in 2025, so many of them will need to keep hunkering. Exit multiples were flat to down in 2025, though I am hopeful they will bounce up in 2026.
4. Attention is the new oil. Hit. This one is an accidental double entendre because of the popularization of LLMs (enabled by the transformer architecture published in the famous 2017 paper Attention Is All You Need). But my intent here was not to talk about transformers but communications, inspired by Chris Hayes’ book, The Sirens’ Call.
Anyone who’s lived with addictive, dopamine-fueled doomscrolling (i.e., anyone who’s ever used social media) knows that this is true. It’s a clickbait world, often a “rage bait” one (Oxford’s 2025 word of the year), and we have enormous trouble resisting. Click here to learn the nine ways to use clickbait tactics in your marketing strategy [6].
The point of this prediction was to say that we’re leaving the age of information and entering the age of attention. Information used to be scarce. That drove content marketing strategies and the use of gated assets to drive leads. But a few things changed:
- The algorithms have taken over social media to the point where the word “follow” is meaningless. It’s no longer enough to have a large follower count; you need to write posts that the algorithm likes or no one will see them [7].
- Thanks to AI, content generation has become free. Whereas information was once scarce, information is now plentiful and cheap. Unfortunately, it’s often AI slop.
- Thus, the challenge is no longer producing and optimizing gated assets, but instead trying to break through the noise with your point of view.
- All while the gatekeepers are changing from search engines to generative AI engines, requiring new and different optimization strategies.
There are two strategies in this environment: (a) play the game by fighting for attention in the media and on social platforms [8] and (b) build your own audience via newsletters, podcasts, blogs, and educational content. Companies should do both, subject to the constraint of not playing the game so well that you damage your credibility. Trust matters. See prediction six for 2026.
5. The worldwide web, as we knew it, is dead. Hit. The Economist called this one for me.
Here’s what they had to say:
Artificial intelligence is transforming the way that people navigate the web. As users pose their queries to chatbots rather than conventional search engines, they are given answers, rather than links to follow. The result is that “content” publishers, from news providers and online forums to reference sites such as Wikipedia, are seeing alarming drops in their traffic.
As AI changes how people browse, it is altering the economic bargain at the heart of the internet. Human traffic has long been monetized using online advertising; now that traffic is drying up. Content producers are urgently trying to find new ways to make AI companies pay them for information. If they cannot, the open web may evolve into something very different.
There’s no better example of the effects of AI front-running on web traffic than Stack Overflow (chart posted by a top-ten, all-time contributor) as you can read about in this post by Gergely Orosz.
6. Working for the algo. Hit. This was a contrarian prediction inspired by an odd fact. The Turing test asked whether a computer could convince a human that it was human. Today we’re living through the inverse: humans increasingly have to convince computers that we are human. If that sounds strange, think about CAPTCHAs — every time you solve one, that’s exactly what you’re doing.
More generally, “working for the algo” means we are working to please an algorithm as opposed to it working to please you. Every time you optimize a social media post to drive amplification. Every time you jiggle your mouse to show you are still working from home. Every time an Amazon driver pees in a bottle. Every time you two-factor authenticate to prove that you are you. Every time you change a web page for SEO or AEO. Every time you modify your spending behavior to boost your credit score. Every time you do any of these things, you are working for the algo.
7. The death of SaaS is greatly exaggerated. Partial. This prediction was playing off two things: (a) a popular rant by Satya Nadella about how SaaS apps were just UIs atop a CRUD database, and (b) the widely-distributed September, 2024 story of Klarna’s plans to replace Salesforce with in-house AI applications.
Since then, the storm around the Satya rant quieted quickly and the Klarna CEO went on record saying he was terribly embarrassed about the whole situation.
While the reality is that enterprise software platforms are very hard to extract (“you inject them into your corporate veins” as my friend Paul Wiefels likes to say) and ergo at relatively low risk from generative AI replacement, there are nevertheless seeds of truth in the “death of SaaS” argument.
It doesn’t start with ripping out Salesforce or Workday. It starts with a huge variety of new applications focused either functionally (e.g., Vic.ai in accounts payable [9]) or vertically (e.g., EvenUp in personal injury law), continues into enterprise AI application platforms such as Writer [10] and Glean, and a new generation of low-code / no-code tools for building bespoke applications from the red-hot vibe coding companies (e.g., Lovable, Replit) to new offerings from established enterprise vendors like OutSytems [11].
SaaS is sick, but it isn’t dying. Though, if you’re not actively incorporating AI into your traditional SaaS product as your top product priority, then your company might be. My most-viewed article of the year was the newsletter I wrote for Topline broadly on this topic.
8. An unlikely revival of branding. Hit. “Brand versus demand” was the marketing watch-phrase of 2025. Brand is not only gaining relative steam in this dichotomy, but marketers would like it to gain even more.
The idea is simple: as traditional demandgen and “performance marketing” channels become more expensive and less effective, marketers increasingly want to allocate budget to more brand-oriented programs. I think this is a good idea because (a) brand and demand spend feed each other (you’re more likely to register for a webinar from a company that you know) and (b) over the past five years we, as an industry, have over-rotated to highly measurable marketing into a myopic position where boards and CEOs are now reluctant to invest unless you can “show me the money” quickly and directly. But that’s not how marketing works. It’s not how people work. And while measuring brand program effectiveness requires stepping outside the world of traditional CRM reporting, it is by no means impossible. That said, brand measurement is something of a lost art that marketers will need to relearn in 2026. (For listeners of The Metrics Brothers, Ray and I are scheduling an episode on this topic for early in the year.)
9. PR is the new SEO. Miss. While I think people would acknowledge the underlying point – that great PR hits can have a strong positive effect on answer engine inclusion — I don’t think this realization sparked the PR renaissance that I was hoping for in 2025. First, because few companies can generate such a hit regardless of their PR investment – i.e., even a great PR flack needs an articulate and charismatic founder, a product with demonstrable customer impact and momentum, and a story that maps pretty directly to mainstream business. Hypergrowth and an enormous fundraising round don’t hurt either.
More importantly, it’s because PR has changed. It’s a lot more trench warfare across dozens of bloggers, podcasters, and influencers than it is swinging for home run stories. If any spending shot up in 2025 it was on lobbying, not public relations. Though the two fields are less far apart than one may think.
But no sane marketer is going to propose a big PR budget in order to boost their AEO program; there are so many other things they can do first. So, hopefully that was thought-provoking. But a miss.
10. LinkedIn enters the social media death cycle. Partial. Followers will know I’ve been trying to manifest LinkedIn improvement because I’ve had troubles remaining on X. And this prediction was kind of a caution, an anti-manifestation (as if anyone was listening) to say if LinkedIn doesn’t get better, it might get a lot worse. But neither happened. It didn’t get better, but it didn’t really get worse. It’s still the same vanilla, namby-pamby, excessively self-promoting content as it was in 2024. But I do see an increasing number of posts from second-hand recyclers of ideas. Think: “Steve Jobs said this” or “Revolut did that ” from people who were not first-hand involved in the company or its story. Which, when they are, is the ultimate cool thing about social media. So, while I think LinkedIn has thus far avoided enshittification and the social media death cycle, it’s still a site that I’m rarely excited to visit.
So, an 8 out of 10, overall. Let’s move on to the coming year.
Predictions 2026
1. Trump, but even more so. With checks and balances significantly weakened, expect an even less constrained Trump presidency. His leadership and communication style — modeled more off a 1970s mafia Don than a democratic statesman — will likely become even more pronounced. That means more indecency, corruption, lip service to the rule of law, revisionism, power plays, and “Donroe” doctrine foreign policy. Given his age and health, Trump may view 2026 as a culminating chapter, raising the odds of bolder, riskier moves aimed at both leaving his mark and cementing his legacy.
Speaking of legacy, prediction market Kalshi currently assigns a 16% chance that Trump leaves office in 2026 [12]. Thus, the possibility of a Vance presidency is becoming more plausible — if not in 2026, then perhaps in 2027. Maybe, among other things, we’ll learn to stop electing septuagenarians.
Meantime, if you’re interested in learning how people can perceive reality so differently, try reading Good Reasonable People by Keith Payne. It’s insightful, if not particularly optimistic.
2. It’s a bubble, but this time it’s different. Make no mistake, today’s AI market has many of the hallmarks of a classic bubble (e.g., technology catalyst, euphoria, high valuations, debt accumulation). And, of course, part of the euphoria is everyone saying it isn’t a bubble (i.e., “this time is different”).
But I think there are some key differences between today’s situation and the Internet bubble that will affect how this one unwinds. Usually bubbles pop; I think this one is going to leak – i.e., the deflation will run in slow motion relative to traditional bubbles. Why? Three things: (a) this is a private market bubble where most of the players are not public companies, (b) everything happens in slow motion in VC/PE land where funds last a decade and trading (i.e., valuation determination) is infrequent, and (c) there is opacity in the private markets that can make it hard or impossible to see when the bubble is deflating.
For example, consider a native-AI company growing at 200% that recently raised money at 50x $20M in ARR, so a $1B valuation. Now, let’s say growth slows well below 100%, but they grind it for 2-3 years to double revenues. For a flat round, they’d need a 25x valuation, but with things deteriorating that’s not going to happen. So, how can you obscure a down-round if you want to? Structure. You can put a 2x liquidation preference on the new money to preserve the headline $1B valuation, but ceteris aren’t paribus. This is a down-round in a wig and lipstick. But since terms are not disclosed, the world will basically never know. And multiple liquidation preferences are just one type of structure you can use in this situation.
This can’t happen when companies are public. Thus, I believe that private bubbles end differently than public ones. They deflate, as opposed to pop. The process happens more slowly and with less hysteria. But it happens. Some will argue it’s already happened to the last generation of SaaS unicorns. In 2026, it will start to happen to many of the AI ones, including a few of the big ones.
3. IPOs are back, baby. Many folks thought 2025 would be a banner year for tech IPOs but, while they did increase over 2024, the damn didn’t break open due to tariff shock in the spring and the government shutdown in the fall.
I think 2026 will see a big uptick relative to 2025, both in the number of transactions and the deal sizes. Why? First, because there is a backlog of amazing companies waiting in line. Second, because while companies can stay private longer – sometimes indefinitely – that’s becoming the luxury of the ultra-elite firms, so all the merely elite ones will still need to go public at some point to get proper liquidity. Third, because a better exit environment is one of the top priorities Trump-supporting VCs wanted, and I believe both Sand Hill Road and Wall Street will apply significant pressure to get what they bargained for. Finally, because there is a liquidity crisis in parts of VC/PE and if the public markets can successfully absorb the elite companies, then lower-tier companies will be in a hurry to go public while the IPO window remains open.
And that’s not to mention the potential gale-force tailwind if the administration ends up setting interest rates — at least before the subsequent sugar high would likely lead to 1970s-level inflation.
4. AI takes jobs, but not as we think. There’s no doubt that AI will replace jobs and plenty of them. The questions are (a) will they be missed and (b) will it drive folks up or out?
I don’t think anyone misses elevator operators, a job that peaked in the 1950s and all but extinct today.
The same is true of switchboard operators, pin setters, lamplighters, linotype setters, icemen, typists, shorthand secretaries, and slubber doffers. Yet, all these jobs were automated away. While I’m sure transition was rough for any given elevator operator, as a society we don’t miss them. Most coal miners work hard so their children won’t have to be coal miners.
Let’s remember where the word luddite comes from. Today, we use the word to describe people opposed to new technology. But the Luddites were part of a labor movement; they didn’t destroy machines because they were technophobes, they destroyed them to protect jobs. So, the transitions can be rough – and come with social unrest — when jobs are automated away. But we still don’t miss the jobs they fought for.
And sometimes, when we think a job is going to be automated away, we get it wrong. In 2016, the godfather of AI, Geoffrey Hinton, said that we should stop training new radiologists because AI could read images better than people. And he was right. AI has outperformed people for nearly a decade at image reading. But today we have more radiologists than ever, with an average salary of $520K. It turns out that reading images isn’t all that radiologists do.
This is an example of AI driving a job up, not out. But this doesn’t always happen, either. I watched Jensen Huang trying to apply this exact argument to Uber drivers [13]. Let’s just say that it didn’t work [14]. For some jobs, such as drivers, they just drive. Those jobs are more driven out than up.
So, are we all condemned to become unthinking LLeMmings? I’m not as negative as the “we won’t be missed” crowd (thoughtful rebuttal here). Nor do I believe that the path to AGI is through LLMs. But, with admittedly little background in the space, I am nevertheless drawn to the idea of world models, which act as state prediction machines as opposed to word prediction machines.
In short, AI will displace a lot of jobs, but many will be pushed up, not out. While the transition costs will be painful for those affected, society overall will benefit. In the end, a lot of bygone jobs won’t be missed. And I’m quite optimistic about our ability to make interesting, new ones.
5. The polymaths aren’t polymathing. Since money isn’t the flex it used to be in Silicon Valley, something was bound to replace it. Now you can’t just be well educated, you can’t just be rich. You can’t just be smart about technology. You need something more. You need to be smart about everything. That is, you need to be a polymath.
So, the first part of this prediction is simple: I think being a polymath will be the ultimate flex in Silicon Valley in 2026 and thus we’ll see an increasing number of people working to attain, and demonstrating the flex of, polymath status. If you want to join the wave, start by reading these lists of books recommended by Marc Andreessen. There’s some great stuff in there.
Don’t get me wrong. I’m all for well-rounded individuals who don’t specialize too early and end up one-dimensional as a result. Many of my friends are super interesting because, in addition to high proficiency in tech, they’re students of business, languages, philosophy, or history. This is all cool. Great stuff.
What’s less cool is when Silicon Valley leaders decide that they are smarter than everybody about everything. Sure, in the spirit of the bear joke, you might be smarter about history than the other MIT engineer, but are you smarter than the person whose lifelong passion is history? Do you know more about international relations than the person who went to SAIS and then to the state department? Do you know more about improving government efficiency than people who’ve spent their entire careers working in government?
Moreover, were the people who built the data structure behind tweets the best people to predict the societal impact of Twitter? Are the people who build GPUs or foundation models really the best people to predict the societal impact of AI? We tend to confuse the ability to build something with being in the best position to understand its impact. Oppenheimer knew how to build the bomb, but later said, quoting the Bhagavad Gita, “now I am become death, the destroyer of worlds.” If you want to know how to build a bomb, ask a physicist. If you want to know its impact, ask a general.
Finally, there’s a difference between being well rounded and being a dilettante. When the new Silicon Valley overlords – whose primary competency is making money by creating technology businesses [15] [16] – start thinking they should be in charge of government and society, blind to their own ambitions and agendas, we’re going to get more messes (e.g., DOGE), more inane statements, and more crazy interviews. The reputation of Silicon Valley will continue to suffer and eventually – probably in 2027 or 2028 – we’ll see backlash-driven policies that actually hurt the ecosystem. As bizarre as the Thiel interview was (and as hilarious its parodies), if you wanted to drum up popular support for a billionaire tax, it’s hard to find a better tool.
If they were really dialed in, the new overlords would take a lesson from the robber barons, such as Andrew Carnegie — and build libraries and write essays like The Gospel of Wealth. They’d also take fewer dubious actions and obviously self-interested policy positions.
6. Trust is the antidote. Across a web of AI-generated, algorithm-juiced content, people don’t know who they can trust anymore. What news can they trust? What companies can they believe?
Can they trust the answers to seemingly basic questions, like “how much does this hotel cost?” Ah, the mandatory resort fee. Or “how much does this ticket cost?” Ah, convenience fees or seat-reservation fees. And, with generative AI, can they even trust what they see with their own eyes? Was that video real? How about this photo?
People can’t trust the algorithms to show them content they want to read. Slide into the wrong thread and X instantly turns into a bot-driven cesspool of ad hominem attacks. LinkedIn has become a tedious sea of corporate news, thinly-veiled pitches, humblebrags, and ghost-written content. Reviews sites, despite consumer protection rules, are often gamed.
Soon, if not already, we’ll be using AI both to generate and summarize our communications. And, as Scott Brinker says in his predictions, “AI inbox gatekeepers will turn email marketing into earned media.” Ponder that for a second.
So, what’s a marketer to do? One thing. Focus on trust. Only trust will get people to open your emails. Only trust will get them to sign up for your newsletters and subscribe to your podcasts. Only trust will allow them to believe the reviews and testimonials about your product. Only trust will get them to listen to what you have to say.
Trust that you produce great content. Trust that you won’t lie to them. Trust that you won’t waste their time. Trust that you won’t sell their contact information to someone else. Trust that you won’t speedbag them with SDR calls. Trust that if someone unsubscribes, you’ll stop. Trust that your product does what the marketing says it does. Trust that customers get the outcomes you promise.
I’ve always felt that branding was simply about trust. Trust that you’ll be you [17]. Trust that you’ll look like you (visual). Trust that you’ll sound like you (voice). Trust that you’ll act like you (values, operating principles). Trust in your mission (vision).
It’s funny. The further we got into the Internet era the more we heard that people didn’t have time anymore. And that long-form content was dead. So, everything needed to be bite-sized morsels. There’s no time. Make it shorter. Make it a short video. Make it an infographic.
I always believed David Ogilvy: “long copy sells.”
You know when people don’t have time for your content? When it’s bad. And there’s so much bad content out there that we started to believe that lazy people were the problem. No, they aren’t. Bad content is the problem. And now it’s easier than ever to generate it. So, do we win the war by out-slopping our competition? Or do we try something else?
You know when people do have time for your content? When it’s good and they’re interested in the subject (you know, like when people are betting a part of their career on your software package). Ask the guys at Acquired and observe evolution of their average episode length:
- Season 1: ~45 minutes.
- Seasons 2–4: ~1 hour 15 minutes.
- Seasons 5–8: ~2 hours 30 minutes.
- Seasons 9–11: ~3 hours 30 minutes.
- Seasons 12–14: ~4 hours+
There is literally no better example that when people are interested and you produce good content, that they will make the time for it. I liked the three-hour episode on Costco so much, I listened to it twice.
So, overall, what would I do? Make good content. Make long content, get rid of artificial and outdated best practices and constraints. Take the time to tell the story you need to tell. Build your own (“first party”) audience. Get access to, and respect the rules of, private communities (e.g., The SaaS CFO, Exit5) that are increasingly forming to provide trusted spaces.
And don’t be afraid to place marketing chips on brand over demand in 2026.
7. Fee culture changes VC. As with several of my predictions, this one is already in progress, but I do see it accelerating in 2026. The trend in VC is towards increasingly larger funds, e.g., a16z’s recent announcement of over $15B across about six funds, representing 18% of all VC dollars raised in the USA in 2025. (Read The Metamorphosis of a16z for a fascinating, if somewhat conspiratorial, take.)
While many smart people are writing about how VC is changing in terms of expected outcomes, fund performance, concentration, kingmaking (and how founders can deal with it), risk, value creation, and scalability, there’s one thing that few people are talking about: fee culture, which is enabled by these ever-increasing fund sizes.
Back in the day, using several assumptions [18], a successful VC running four funds in parallel could make around $3M/year in fee income. With that, they could live a very comfortable Silicon Valley life: a house in Atherton, a place in Tahoe, tuition at the Menlo School, and all the rest. But to get really rich – fly your PJ to your Montana ranch rich – then you needed carry.
What’s carry? The second half of the 2 and 20 fee structure common in VC. The 2% is the annual fee on committed capital. The 20% is the carry, or cut, that VCs take after the fund returns its initial capital to investors. For example, if a $250M fund returns 3x to investors, the VCs would make $100M in carry (20% of $500M) over the fund’s roughly ten-year lifecycle [19]. Split that $100M across five general partners, and after one fund you’ve got the Montana ranch and, after another, the PJ to fly to it [20].
But 2% of a huge number is a huge number. So today, with a $5B fund, a VC can make more money off fees alone than they previously made off fees and carry [21]. By my rough math, more than double. On top of this, the carry, split across 15 GPs, would be $133M each [22].
While I’ve spent my career in and around VC-backed startups, I don’t consider myself an expert on VC, per se. But I think this new mega-fund reality changes things. Remember the Charlie Munger quote: “Show me the incentive and I’ll show you the outcome.”
While today’s mega-fund VCs are certainly incented to earn the aforementioned “substantial” amount of carry, they can live better — on fees alone — than the VCs of yore did on total compensation.
So, what do I think this means?
- People managing large amounts of money need to deploy it. You can’t raise $5B, collect $100M/year in fees, and then not invest the money.
- They will generally prefer to deploy it in bigger chunks. It’s easier to deploy $5B with $250M to $500M checks than $25M to $50M. It’s the difference between 20 investments and 200 investments. There’s a fixed cost – and a relatively high one – in pre-investment diligence and investment commitment presentations.
- Kingmaking should increase. If I’m sure, for whatever reason, that founder X will be the winner, I can load them up with money to help ensure that outcome happens. While more VC gets deployed, it ends up being a have and have-not financing environment.
- There will be a temptation to “foie gras” startups. Unlike the geese, this requires founders as willing accomplices, but if you follow the simple rule of “raise money when you can on good terms,” most will say yes. Hopefully they then won’t piss it away, which is how you get in real trouble.
- This should create a damn-the-torpedoes attitude. If you need to deploy $5B in order to make $20M/year in fee income, then I’m pretty sure you’re going to find a way to do so. Moreover, the game becomes “tails I win big, heads I win five times bigger,” then who doesn’t want to play? Put differently, it’s not hard to be a techno-optimist when your firm is collecting $300M/year in fee income off your last $15B round of funds. Heck, I’d be one, too.
- Mega-VCs will become more investors and less business partners. I remember when Sequoia’s tagline was “the entrepreneurs behind the entrepreneurs.” Around that time, I also remember being shocked – and this is going to sound incredibly stupid – when I heard someone refer to VC as part of “financial services.” “What? VCs aren’t Wall Street types,” I thought. “They’re tech people.” Of course, I was wrong at the time, but part of me nevertheless continues to see this as “the financialization of VC,” or more aptly perhaps, “the PE-ization of VC.”
8. A growth retention apocalypse. I’ll start this one with an excerpt from The Grapes of Wrath.
Then the grapes [ripen] – we can’t make good wine. People can’t buy good wine. Rip the grapes from the vines, good grapes, rotten grapes, wasp-stung grapes. Press stems, press dirt and rot.
But there’s mildew and formic acid in the vats.
Add Sulphur and tannic acid.
The smell from the ferment is not the rich odor of wine, but the smell of decay and chemicals.
That’s the way I feel about ARR today. Press trials. Press onboarding. Press professional services. Press overages and hardware. Take anything, multiply it by 12 and then press that. The smell from the ferment is not the oh-so-sweet smell of annual recurring revenue, but that of bygone customers, tire kickers, experiments, and decay.
In other words, I see what Cassie Young calls a growth retention apocalypse. I think we should measure it with a metric that I call retention spread.
Retention spread = NRR – GRR
That is, the gap, in percentage points, between NRR and GRR [23]. When the apocalypse hits, you might see it in NRR alone, but to the extent that spectacular growth can hide heinous retention, you might not. That’s why we need to look at retention spread [24].
9. Context graphs storm the market. Sing with me: “on the negative-third day of Christmas, my X feed sent to me: a billion posts on so-called context graphs.”
The first post I saw was this paper by Jaya Gupta and Ashu Garg of Foundation Capital. I’ve rarely seen any enterprise idea hit harder, take off faster, and get more engagement. Over the next few days, here are some of folks who joined the conversation: Dharmesh Shah (Hubspot) Aaron Levie (Box), Nick Mehta (Gainsight emeritus), Arvind Jain (Glean), Satyen Sangani (Alation), Anshu Sharma (Skyflow), Colin Treseler (Supernormal, Radiant), Animesh Koratana (PlayerZero), Diego Lomanto (Writer CMO, who made a particularly compelling case for marketing), Kirk Marple (Graphlit), Anshul Gupta (Actively, and whose sister was a co-author of the original article), and Tony Seale (The Knowledge Graph Guy).
This is not a new academic paper from a research lab, but an article from a VC firm on what they see as a trillion dollar opportunity. Why did it resonate?
- First, they either got profoundly lucky or did a very well-coordinated launch that leveraged their considerable social network [25]. Or perhaps a bit of both.
- It hits on a sensitive issue within the enterprise: the predicted imminent death of systems of record made by AI zealots who may understand AI well, but who lack an understanding of the enterprise. Even if you believe in an upcoming wave of AI agents – as I do – the question remains: where will “the truth” live in enterprise systems?
- It raises the critical, and often forgotten, question of context. Not just the current state (which operational systems capture) and not just historical states (which data warehouses and analytic infrastructure capture), but the context for why it happened, which they individually call decision traces, whose accumulation results in a context graph.
As someone who started in operating systems and then moved into databases, I had to be dragged into caring about context. That’s not what we do here. We store stuff. We index it. We back it up. We ensure ACID transactions. We let you query it in flexible ways. But: How did it get here? Where did it come from? What does it mean?
As the Tom Lehrer lyric goes: “Once the rockets are up, who cares where they come down? ‘That’s not my department!’ says Wernher von Braun.”
But data warehouses taught me to care about technical and operational metadata. Data catalogs taught me to care about governance, collaboration, and quality metadata. Over time, I became a true believer in context around data [26]. Thirty years in business taught me to care about business metadata. Stuff that, until recently, most of us only dreamed of keeping in systems.
Ultimately, what’s exciting is that this is much more than just replacing existing systems or building value-added layers atop them. It’s about capturing a category of information – knowledge – in enterprise systems that has, for the past 5 decades, generally eluded us in all but the most specialized systems. That’s big. In the 1980s, I got to watch the industry transform by capturing data.
Now, I’m going to watch the sequel by capturing knowledge [27], not just in the sense of documents and text, but process [28]. Remember the old Lew Platt quote: “if HP knew what HP knows, we’d be three times more productive.”
I know that context graphs hit hard over the holidays. Will we still be talking about them next Christmas or will they set a new record for hype cycle traversal? I think the former.
10. The Rule of 60 replaces the Rule of 40 for traditional SaaS. As industries mature, so do their financial metrics.
And their multiples change. Consider this tweet: “looks like we’re paying 5x sales for software. Mr. Valentine has set the price.”
Let’s observe that 5x sales with 33% EBITDA margins means 15x EBITDA, which by PE standards is a healthy multiple for a software company. Also remember that 5x is about twice the average price/sales ratio of the S&P 500. So, software companies are still well valued; just not as well valued as they were before.
The Rule of 40 is a SaaS metric that was created to be a better predictor of a company’s enterprise value to revenue (EV/R) multiple than growth or EBITDA alone. The intent, coming off a growth-crazed period, was that the best companies should balance growth and profit. Hence the Rule of 40 was generally calculated as [29]:
Rule of 40 score = YoY revenue growth rate + FCF margin
Year later, Bessemer came along and correctly argued that a point of growth should be worth more than a point of profit, and created the unfortunately-named “Rule of X” which put a slow-varying multiple on growth relative to profit, typically around 2.3x. Years before, and less visibly, the Software Equity Group had created a 2x growth-weighted Rule of 40 which did roughly the same thing.
But today’s debate is not about growth weighting. It’s about the proper score. Increasingly you hear people, particularly in PE, say that are now pushing towards a Rule of 40 score of 60, or in their parlance, a Rule of 60.
Let’s look at a spreadsheet that compares two companies. They both start out in Year 1 at the same size with the same Rule of 40 (R40) score of 40, composed of (20%, 20%) [30]. The first company remains at (20%, 20%) while the second evolves from (20%, 20%) to (15%, 45%) over the four-year period. Let’s see what happens to valuation.
Over the four years, the first company ends up valued at $518M on a 15x FCF basis and $864M on a 5x sales basis. Since there’s a question whether either of those companies would get valued on a revenue multiple — which is usually associated with higher growth — let’s look primarily at the FCF multiples. The second company ends up worth $1.03B based on 15x FCF, whereas the first one ends up worth only about half that at $518M. Since PE investors usually see profit as “more of sure thing” (i.e., “more within management’s control”) than growth, I think the vast majority would push to become the company on the right.
And most do. Hence, why I predict – particularly in the traditional SaaS, non-native-AI part of the market – that most companies will be shooting for R40 scores of between 40 and 60 in 2026. The Rule of 40 is dead! Long live the Rule of 60!
(That is, unless you’re a native AI startup and growing like a weed.)
Conclusion
If you’ve read all the way through, let me offer a special thank you for sticking with me. I wish everyone a happy, healthy, and above-plan 2026. I’ll conclude with a hat tip to Bob Weir of the Grateful Dead who died twelve days ago. Here’s John Mayer playing a touching Ripple, on one of Bobby’s guitars, at his life celebration in San Francisco last weekend.
Notes
[1] Also see terms and conditions as well as other disclaimers in the Kellblog FAQ.
[2] In my view, the most consistently monetizable activity in the crypto ecosystem has been the promotion and issuance of new tokens. Given that such tokens usually lack intrinsic value, these dynamics can, in practice, start to resemble Ponzi structures. Moreover, crypto assets have also been associated with a range of questionable and/or illegal use-cases, including money laundering, ransomware, online fraud and investment scams, political influence, tax evasion, and payments for illegal goods (remember how Tim Draper came across his considerable bitcoin position). The most legitimate use cases, in my view, are speculation — which may be unwise but should not be unlawful — and stablecoins for use in countries lacking reliable local currencies.
[3] At the time I remember thinking two things: (a) how unusual, and arguably unwise, it was for VC firm to declare a political position, and (b) that as soon as you declare yourself a single-issue voter – on any issue — you absolve yourself from looking at the bigger picture and the very real complexities within it. In short, it’s a cop out.
[4] The start of this video talks at length about the president’s personality which I find uncompelling because virtually all con men are personable (it’s kind of a job requirement) and there is ample data suggesting his behavior speaks otherwise. Ergo a higher bar than pleasant conversation and good listening skills should be applied.
[5] That there exists a startup focused on helping founders close startups is itself indicative of something!
[6] Please tell me I didn’t get you with this one!
[7] On LinkedIn and X, I can write posts that reach 1% of my followers if I don’t pay attention to the rules.
[8] And playing the game works differently on different platforms. LinkedIn swims upstream on many of its choices, so what works elsewhere will tend not to work on LinkedIn, and conversely.
[9] Where I sit on the board.
[10] Where Balderton is an investor.
[11] With whom I also do some work.
[12] As of the day I wrote this (1/14/26). That includes only resignation or removal. Prediction markets don’t allow bets on deaths so this market will not resolve to yes in that event.
[13] The things I do for my readers! Heretofore, I’d never watched an episode of The Joe Rogan Experience.
[14] Saying roughly that drivers, other than driving, do lots of things like … uh, uh, uh … uh, uh, uh … security? Not exactly an example we can all relate to. While billionaires often use drivers for security, for most people – if they’re even using one at all — a driver just drives.
[15] Even if those businesses are not ultimately successful or profitable
[16] In some cases, it’s creating. In others, it’s more “walking in front of a parade.” For example, Benioff, Musk, Altman – and many others — all to a greater or lesser degree walked in front of a parade. This, of course, is also a skill, both in identifying the right parade and in finding a way to get in front of it. And then, of course, building the business once you’re there.
[17] And deciding who you want to be is a key part of the exercise.
[18] Assumptions: $250M funds, 5 GPs, 2+20 fees, 4 funds in parallel, $3M staff and admin expense.
[19] This ignores the carry that is distributed to staff within the firm. But all math here is approximate, just to demonstrate the higher-level point.
[20] Or, more realistically, a nice fraction of a G5.
[21] Assumptions: $5B fund, 15 GPs, 2+20 fees, 4 funds in parallel, $30M staff and admin expense.
[22] Assumptions: 20% of $10B returned in excess of the $5B fund divided across 15 GPs.
[23] The naming here gets tricky due to existing conventions. If I were to call it net expansion rate, it would be too easily confused with net revenue retention (NRR); they sound too similar. What I’m trying to get at is the mechanism through which you hit your NRR. You can hit any given NRR (e.g., 100%) in several ways. For example, expansion of 50 against shrinkage of 50, or expansion of 0 against shrinkage of 0. The first example has a retention spread of 50 points (100 minus 50), the second 0 points (100 minus 100).
[24] Kyle Poyar wrote a nice report, The AI Churn Wave, on this topic. While he doesn’t explicitly look at retention spreads and while his data doesn’t show what I’d expect (i.e., higher retention spreads in AI than B2B SaaS), we do both arrive at the same conclusion: there is a wave of churn in AI startups coming.
[25] But launching something on December 22? Was this the ultimate contrarian timing, or a launch that slipped a few weeks and they thought, “Oh, just do it anyway”?
[26] I no longer tell my erstwhile favorite joke: the only thing you can make with meta-data is meta-money.
[27] Much like how AI was “a thing” decades ago, but only became real and widespread recently, so had knowledge management been a thing for decades. But it’s also never been real and mainstream. Maybe now it finally will.
[28] Similarly, I know that process mining has been around for a while and Celonis has created a ~$13B (valuation) business doing it. I view this as validation for lighter, more automated, more robust process graphs. As the William Gibson quote goes, “the future is already here, it’s just unevenly distributed.”
[29] Due to its odd name there are some tongue-tying semantics around it – e.g., “my Rule of 40 is 30” or “we are a Rule of 30 company.” To avoid this, I insert the word “score” – i.e., our Rule of 40 score is 30.
[30] I’ll use (growth, profit) as my nomenclature for communicating the elements of the R40 score.
