The SaaS Apocalypse: Who Lives & Who Dies | Insight Partners Co-Founder, Jerry Murdock

Summary

• The AI tsunami is specifically autonomous agents, not AI in general — and we’re in the anticipatory period, which is why the “SaaSacre or the SaaS apocalypse” is hitting prices before the wave lands. Murdock’s AI-native portfolio (likely E2B, Eventual, Lotus AI, Get Dynasty, Aven) has used OpenClaw, NanoClaw, or homemade agents to write code for just 2–6 weeks, and most of them reportedly view $27–30B Cursor bluntly: “Cursor is obsolete. That’s where the product is today.” He thinks the team has money and time to pivot, but “you can’t be thinking about yesterday.”
• The stack call: a LAMP-style open-source “claw stack” for agents, with an orchestration layer triaging workflows between expensive Claude tokens and open-source models like DeepSeek or Llama 3 — driving the rise of open-source models and then an ASIC explosion with the model on the chip, “cheaper, a lot more tunable… than an expensive chip from Jensen.” His read on the Groq deal: Nvidia bought it to make sure CUDA is viable for the ASICs it knows are coming, while Meta “had the balls to say no to Jensen” precisely because it’s betting on ASICs. Nvidia’s fate: “depends on execution.”
• Agents become the software buyer — an employee with credentials and identity that you review (“What did you buy? What did you spend? What did you accomplish?”) — pricing goes consumption-based (Docker is moving dramatically toward it as it moves into AI), and software not built for agents is “severely challenged” in “maybe 6 months, maybe a year, maybe 18 months.” Infrastructure gets rebuilt to agent speed: humans notice 400ms delays, likely E2B’s sandboxes respond in 80ms, and “the agents notice it and that’s what matters.”
• Systems of record become valueless databases or infinitely more valuable — execution decides. Carta is “infinitely more valuable” if stock tokenization routes through its cap table, worth little if a new system of record bypasses it; Salesforce is “like Mount Everest… it’s not going to melt overnight,” and the tell is the health of the dozens of companies built on top of it.
• The Cloudflare/CrowdStrike ~10% drop on Anthropic’s security release wasn’t panic — “it’d be down a hell of a lot more” — but sidelined buyers waiting to learn winners from losers. The rhyme is March 2000: tech down 30–40% across the board, miss a quarter and you’re down 50–60%, and the tsunami “took out the dotcoms and then took out all software.” Nothing is safe now: 80% of his investments returned less than 1.3x, and some PE firms “will end up like Forstmann Little” after its all-in telecom bet died with 9/11.
• Labor displacement could decide the next presidential election 2.5 years out, with “minimal viable income” a plausible ballot question; the first casualty is the next hire — the junior developer, EA, marketing person never recruited — with SMBs adopting first and enterprise last. The company Harry identifies as likely Klarna going from 7,000 to under 2,000 by 2030 is a culture question, and billion-dollar single-person companies: “Yeah, absolutely.”
• “Absolutely the best time” ever to start a fund — timing is the single correlation across VC vintages (an ‘05/‘06 fund caught mobile; a 2009 fund missed Twitter, Facebook, Uber early), and this sea change gives fresh entrants a huge advantage over successful incumbents: “hard to get old dogs to move fast.” VCs and founders will be judged on the same axis — “how well do we use autonomous agents in our job?” — while the one moat agents lack is intuition, “not anytime soon.”
• Quickfire: he would take both OpenAI “at 500” and Anthropic “at 380”, favoring OpenAI because of 800 million users — “until you get to a billion users, I’m suspicious of any consumer technology” — though Google’s Gmail-plus-agents are better long-term assets, and converting ChatGPT into a personal autonomous agent (his read on what Peter Steinberger will do) is “an insanely great business.”

Deep dive

Autonomous Agents Are the Tsunami

• Murdock’s opening frame: a tsunami “is harmless when it’s out at sea. It’s only dangerous when it hits the beach” — and it’s messy, with warning earthquakes and pre- and post-peak waves. The event itself is “autonomous agents… not just AI in general.”
• We’re in the anticipatory period, which is why the “SaaSacre or the SaaS apocalypse” is showing up in prices before the wave arrives. He refuses the doomsayer label but not the urgency: “change is coming fast and you need to anticipate that.”
• On bolt-on AI: “It’s possibly true, you can maybe get an exit” — but AI-native thinking makes a better company. His Olympics verdict on late pivots: “you can try all you want, but you better be world class at what you’re trying to do if you’re going to get the medal.”

Cursor Faces Obsolescence

• His true AI-native companies — likely E2B, Eventual, Lotus AI, Get Dynasty, Aven — are all running OpenClaw, NanoClaw, or homemade autonomous agents to write code, only 2–6 weeks in, a phenomenon “about 2 months old” and “not obvious yet in the marketplace.”
• Most of their reported view on the $27–30B Cursor, as they’ve told him, is: “Cursor is obsolete. That’s where the product is today.” His own nuance — smart team, lots of money and customers, a shot to pivot into agents — but “in the AI business, you’ve got to be going where things are going to be. You can’t be thinking about yesterday.”
• Why developers are so passionate about OpenClaw: it works on its own without review. “You’ve gone from an assistant to actual employee. Think about that.”

The Claw Stack Emerges

• The historical template is LAMP: post-9/11 in 2003–04 nobody could afford Sun servers and Oracle databases, the Linux/Apache/MySQL/PHP stack arrived, websites and commerce exploded in 2004–05, and Google IPO’d in 2004 “and rode that wave brilliantly.” The open-source community — whose sheer integration numbers he weighs against OpenAI’s and Anthropic’s concentrated talent — will produce the equivalent “claw stack” for agents.
• Today’s reasoning layer is dominated by Claude, Codex, and Gemini; next comes an orchestration layer that triages workflows: “for this part of the workflow, let’s use Claude. They’re expensive tokens, but they’re going to get the job done better,” and open-source models like DeepSeek or Llama 3 elsewhere. That drives the rise of open-source models — and then ASICs, with the model on the chip, “a lot cheaper, a lot more tunable for a specific workload than an expensive chip from Jensen.” “That’s a revolution.”
• His read on Nvidia buying Groq — “the thing that wasn’t mentioned”: Groq already puts memory on the chip, and the deal is about making sure “CUDA is viable for the ASIC explosion that’s coming. They absolutely know what’s coming.” Does Nvidia keep its value? “Depends on execution” — and note Meta “had the balls to say no to Jensen… because he’s betting on ASIC chips. No question about it.”
• Harry’s commoditization worry — models racing to the bottom and eating the application stack — gets a structural answer: that question “is going to be decided by the autonomous agent, not developers.” Agents are probabilistic; they’ll “get 10 Python libraries, run them in 10 different sandboxes… and see which one performs better.”

Agents Become Software Buyers

• The premise behind “triple-triple-double-double is dead”: today all software is eventually purchased by humans; tomorrow it’s purchased and used by agents. An autonomous agent becomes an employee — credentials, identity, reviewed like one: “What did you buy? What did you spend? What did you accomplish?”
• Pricing follows the buyer: consumption-based models — Docker is moving dramatically toward this as it moves into AI — a sandbox is just memory and compute, free to start, and the agent itself tells you when the limit is hit.
• The likely E2B specimen for agent-speed infrastructure: humans register delay around 400 milliseconds and most sandboxes build to that; E2B responds in 80 milliseconds. “No one would ever notice it… the agents notice it and that’s what matters” — when one agent “spins up literally a 100,000 sandboxes in seconds.”
• His timeline, hedges intact: enterprises “may take a year or longer” to enable agents, but if you’re not building software for agents today — “maybe it’s 6 months, maybe a year, maybe 18 months” — “you’re going to be severely challenged if you still think human beings are going to buy your software.”

Systems Of Record Diverge

• The Carta test: if tokenization of stocks routes through Carta, which holds the cap table, it becomes “infinitely more valuable”; if tokenization bypasses it and a new system of record gets built, “there’s not much to say that system of record’s going to be worth much.”
• Salesforce is “like a Mount Everest… an 8,000-meter peak. It’s not going to melt overnight.” The way to analyze it: watch the health of the dozens of companies built on top (likely nCino among them) — “if those guys start getting knocked off one by one,” the underlying value falls.
• To Harry’s crisis — revenue, growth, and margin have become “transient” as valuation anchors — Murdock’s answer: new technology “expands the market before it contracts,” fear reprices first, then business quality follows management adaptation. “Move to higher ground. Don’t get caught on the beach” when the damn thing hits the beach.

Markets Pause Amid Uncertainty

• On Cloudflare and CrowdStrike dropping ~10% on an Anthropic security release: not panic selling — “it’d be down a hell of a lot more” — but cautious sidelined buyers: “I don’t see a deal here… let’s get more information about who’s going to be the winners and who are going to be the losers.”
• The precedent he lived: March 2000, tech down 30–40% across the board, a missed quarter meant down 50–60%, and 9/11 as “the coup de grâce.” Insight had called the bubble — “you can’t do commerce on dialup and there’s not enough fiber in the ground” — yet the tsunami “took out the dotcoms and then took out all software. We all went down.”
• His honest base rate on safety: “80% of the investments I’ve made have returned less than 1.3x” — the 20% made all the money, and money “is just a sort of batting score for how much impact you had.” So on which agent-layer startups are safe from an Anthropic product update: “there is nothing safe right now.”
• For tech PE (Harry’s Orlando Bravo hypothetical, 15–20% growth SaaS books): Teddy Forstmann went all-in on telecom in 2000 and was done after 9/11 — “there’ll definitely be PE firms that end up like Forstmann Little,” and others that made the right bets will be “bigger and better and more important than they were before.”

Labor Displacement Shapes Politics

• The first casualty isn’t the person with the job but the next person in line — the junior developer, executive assistant, or marketing hire that never gets made. Hiring slowdown in white-collar data-input roles is the first signal.
• Adoption sequence: consumer and small business first — one secretary can make a huge difference for a 2-to-4-person company, and Retro is already using agents to triage support email — enterprise last, “they may catch up in a year or two,” which is exactly where ChatGPT, Google, and Anthropic will be focused.
• The politics, 2.5 years from the next presidential election: labor “could decide the election.” No administration will preside over 10–15% unemployment, so expect a “minimal viable income” — guaranteed monthly money plus retraining, his example being Wyoming branches leased to veterans where technology lets “one or two people run a massive branch instead of eight.” His nearer tell: healthcare may affect the election before labor does — “we’ll find out this fall.”
• On the company Harry identifies as likely Klarna — 7,000 to under 2,000 by 2030 — it’s a culture question — Steve Jobs’ A-players logic means “you’re going to have a hell of a lot more agents than people.” Billion-dollar single-person companies: “Yeah, absolutely” — it’s how smart your agent is, how well you deploy it, and how willing you are to listen.

Timing Favors New Funds

• The one correlation across all VC vintages is timing: an ‘05/‘06 fund was positioned for mobile when Steve Jobs got 10 million subscribers from AT&T in 2008; a 2009 fund missed being early in Twitter, Facebook, and Uber.
• Now is “absolutely the best time” because it’s a sea change — humans stop being the decision-makers about software — and fresh entrants gain a huge advantage over successful incumbents who “don’t quite move fast enough cuz they’re already rich. Hard to get old dogs to move fast.”
• The new diligence: vet the white space with data, then evaluate “not just the person, but the quality of how they use autonomous agents. That’s going to be the deciding factor for venture capitalists and startups. We’re going to be on the same level playing field.”
• The moat agents lack: every decision has two components, logic and intuition, and agents won’t have intuition “not anytime soon” — “we need good intuition to make good decisions. Without that component, you’re not going to succeed.” His quickfire choice between the labs: he would take both OpenAI “at 500” and Anthropic “at 380,” favoring OpenAI on 800 million users — “until you get to a billion users, I’m suspicious of any consumer technology” — while conceding Google’s Gmail-plus-agents are better long-term assets.

Scar Tissue Shapes Investing

• His sharpest self-correction: “intuition was almost never wrong. What I was wrong was me thinking it was intuition. It was nothing but wishful thinking” — specifically about founders he liked as people who were “too comfortable… not crazy enough.” “Most of the people that I liked were the ones that let me down”; the winners are sharp-edged and “challenged socially” (Harry adds Peter Fenton’s line that the best founders make you feel uncomfortable).
• The Twitter 2009 bet: 30-something people, no revenue, partners saying “are you kidding me?” — but “the status update for the world” was an idea “far better than anybody on the team’s ability to execute against it.” He put his entire reputation on the line and closed in under 30 days; he also thinks VC politics and firing Jack Dorsey early left the company “pretty messed up from that trauma.”
• The firm’s breakout wasn’t a deal — it was surviving 9/11, when the ‘99 fund was challenged and peer firms became “zombies, walking dead, partners splitting up.” “We hung in there and survived. That’s the breakthrough moment.” His one strategic regret rhymes: Insight Europe, shut after six months — “a mistake and a half. We were not mature enough” — and remote decision-making remains “very difficult.”
• The signature closers: “Money does not come with instructions… Money is the equivalent of energy and you need to respect it.” And the difference between a used-car salesman and a software salesman: “the used car salesman knows that he’s lying.”