"Cursor is Dead" is Total BS: Here is Why | Miles Clements

Summary

• “Cursor is dead” is total BS, says Accel growth partner Miles Clements — responding to Jerry Murdoch’s “no one’s using Cursor anymore” and Chamath’s too-expensive tweet with actual numbers: 2x more Cursor users use agents than the tab feature, 90% of users are daily-active on the agent product, agents grew 15x last year, and cloud agents launched ~October 30 already drive 35% of merged PRs in Cursor — alongside a leaked $2B ARR. His verdict on the critics: “You’re not looking at any real metrics. Who are these people to make these judgments?”
• Claude Code’s surge — tethered to Opus 4.5/4.6 — is market-expansionary, not zero-sum: new cohorts who “would not have been software developers a year ago” plus off-the-charts consumption revenue mean both tools win. Accel’s near-final developer survey: 50% of developers switch model families daily, 95% switch models daily — the world wants to be multi-model, and that makes Cursor “an index of AI innovation.”
• The Cursor underwrite: Accel entered at $9.5B — roughly 4-5x year-end ARR — on the thesis that no platform company has ever owned engineering the way Salesforce owns go-to-market or CrowdStrike and maybe Palo Alto own cybersecurity. Michael (likely Truell) guided ~$100M ARR to a 500 aspiration; Accel haircut it to 300; the company “ended last year somewhere in the billions.” Lesson: forecasts encode assumptions, but “we’re not managing to earnings calls.”
• Portfolio dogma for this cycle: “You can be successful investing in consensus. You can do really well investing in non-consensus. You get hammered sitting in the middle.” The scar tissue is ServiceTitan — Accel had fallen in love with the opportunity, chased the round at ~$250-300M, and lost by clinging to rigid 6-10x-forward vertical-SaaS multiple rules; it became a $9B company. Jim Brier’s line: the science is valuing a company, “the art is understanding when to break the rules.”
• The Rippling miss “stings”: Parker Conrad is the world’s best at “marginal ease of ARR accumulation” — building future growth levers like laptop provisioning that beat put-in-a-dollar-get-$1.20 marketing math — and Accel passed over reputation concerns and ownership rules. Counter-dogma on winner-take-all: AWS has ~35% share, and Deel at $1B ARR is “120th the size of ADP” — “who will win” is a narrow-minded framing.
• Biggest change of mind: a year ago Miles believed “all of the generational investments in AI had been made” — “a really stupid thing to say and I no longer believe it.” Accel invested in Anthropic in the 180 round, implicitly underwriting trillion-dollar comps (Google, Microsoft, Amazon), and he frames the Pentagon conflict around principle: ~1M net-new consumer signups a day and passing GPT in the App Store — “if you behave the right way you will be rewarded.”
• End-state for the 2021 vintage: “a good time to be in the Thoma Bravo, Vista, Blackstone, KKR business” — LBO homes for Snyk-type assets ($300M revenue, 15% growth, last priced ~$7B), while IPOs in the $2-5B range “never really break out” so companies wait for clear line of sight to trading above $5B. Most oversold name he’ll volunteer: Figma — the “SaaS apocalypse” is an over-rotation.

Deep dive

1. Value AI on two axes: time to value × durability — coding wins both

• Miles’s framework for “true value in an AI world”: score companies on time to value and durability of that value. Legal and accounting AI (like Basis, a new Accel investment) is slow to deploy but “once it is hooked, the durability of that value is transformational.” Early vibe-coding apps were the opposite — a “weekend warrior pickleball app ready to go overnight,” but “the bottom just fell out” because there was no durability.
• Coding is “the vertical in AI” precisely because it shines on both: “you can start using Cursor in an afternoon and by that evening you’re 10 times more productive,” and the value compounds as teams adopt — which is why coding is the battleground of AI today.

2. “Cursor is dead” collapses on contact with the numbers

• Harry stacks the case: Jerry Murdoch’s “no one’s using Cursor anymore, everyone’s using Claude Code,” Chamath tweeting he’ll move off Cursor as too expensive, the Twittersphere meme — versus a leaked $2B ARR. Miles gets “a little bit spun up”: “All due respect, I thought about playing in the NFL, but instead I walked onto a college football team and was the fifth-string inside linebacker. You’re not looking at any real metrics.”
• His actual metrics, from Michael Truell’s public post: 2x more people use agents in Cursor than the tab feature, 90% of users are daily actives on the agent product, the agent product grew 15x last year, and cloud agents — in market only since ~October 30 — already account for 35% of merged PRs in Cursor. The IDE framing is “Cursor being a victim of its own success.”
• The market is expansionary on two dimensions, so Claude Code’s rise — “very much tethered to the success of the underlying model,” Opus 4.5 and 4.6 — isn’t at Cursor’s expense: new user cohorts (“people who would not have been software developers a year ago”) plus consumption revenue “off the charts for both tools.”

3. Multi-model is the moat, and the underwrite was platform-for-engineering

• Accel’s near-complete developer survey (Miles owes Harry the full readout): 50% of developers switch model families daily; 95% switch models daily. The world wants to be multi-model, Cursor enables it, and that makes it “an index of AI innovation” — every model improvement compounds into the product. Building its own specialized coding models isn’t a mistake either: they “don’t need to be good at poetry or teach you how to make an apple pie.”
• The thesis at entry — $9.5B, roughly “four times, five times year-end ARR” — was that Salesforce owns go-to-market and CrowdStrike and maybe Palo Alto own cybersecurity, but “there has never been a platform company for engineering as a vertical”; Atlassian and Datadog built $50-100B companies on single slices of the stack.
• The forecast story is the tell on this market: Michael (likely Truell) guided ~$100M ARR to a 500 year-end aspiration, Miles and Andrew Braccia “haircut it and call it 300” — and the company ended the year “somewhere in the billions.” His conclusion: predictions matter because they encode assumptions, but “the idea of having a budget so you can hold the founder’s feet to the fire quarter after quarter is just not relevant… we’re not managing to earnings calls.”

4. Consensus works, non-consensus works — the middle gets hammered

• Is triple-triple-double-double dead next to a 100-to-a-billion Cursor? “Absolutely not — send me all of your triple triple double double companies.” His market map: “You can actually be successful in this market investing in consensus. You can do really well investing in non-consensus. You get hammered sitting in the middle.” Investors have flocked to the extremes — AI maximalists buying the basket “ownership, valuation be damned,” or valuation strikers sitting on hands — when “the best funds in the world embrace the nuance.”
• Harry’s opportunity-cost pushback lands a concession: he agrees outcomes must be huge — “a billion-dollar exit doesn’t do shit for us now” — and if the founder can’t articulate the big outcome, that is probably a sign not to be involved. The scar: ServiceTitan, where Accel had fallen in love with the opportunity, chased the round at ~$250-300M, and lost by sticking to rigid 6-8x/10x-forward vertical-SaaS multiple rules. “That went on to be a $9 billion company.”
• To Harry’s “your funds are too big to embrace nuance”: a decade ago zero trillion-dollar companies existed; five years later six; today a dozen in publics, plus the labs and SpaceX private. But he concedes pure late-stage momentum-buying is hard — the answer is multi-stage, multi-strategy.

5. “Who will win” is narrow-minded — and the miss that stings is Rippling

• Winner-take-all framing is oversimplified: “we don’t operate monopoly markets in this country… the best software company in the world is AWS. AWS has like 35% market share.” On Deel “winning” HR at ~$1B ARR: “ADP has $20 billion of ARR. You are 120th the size of ADP” — with likely Paychex, Paycom and Paylocity all still standing. Several AI verticals will end with “a couple of really big companies.”
• Deel isn’t a regret (Accel is in Remote, and he rates Job and Marcelo’s product vision); Rippling is. His coinage — the “brilliantly wanky” phrase Harry loves — is “marginal ease of ARR accumulation”: the downstream levers you install now that let you grow crazily in years four through seven, better than “I put in a marketing dollar and I get out $1.20 of revenue.” “Nobody in the world does that better than Parker Conrad” — laptop provisioning and physical IT leasing are bad standalone businesses but brilliant revenue line items.
• Why Accel passed: Parker’s prior reputation (“I like to think he’s now totally overcome” it) made them a step slow while competitors moved fast, and the deal would have required breaking ownership rules at a high valuation. “I don’t regret not breaking the rules in general, but this would have been a time when it could have been worthwhile.”

6. Break the rules rarely — and beware the anomaly quarter

• On Harry’s live dilemma — Series A prices moving from 20-on-100 to 20-40 on 200-400 — Miles reaches for Jim Brier: “Investing is an art and a science. The science is understanding how to properly value a company and the art is understanding when to break the rules.” In this market you face that constantly, but “breaking the rules is something you should do very, very rarely. It’s okay to say no.”
• The competitive pressure to pick fast forces extrapolation from early data, and he’s been burned: a company at $1M ARR prints a $4M quarter, “the product market fit snapped… it is the time to forward invest” — “and then it turns out sometimes they just had an anomaly quarter. I have fallen into that trap before.”
• The old benchmarks “are largely obsolete now”; what matters is usage intensity, because “growth can obscure and blind you to a lot of underlying ills in the business.”

7. Missing the labs, investing in Anthropic in the 180 round, and virtue rewarded

• On why Accel LPs aren’t in the foundation models: “a lot of firms missed the model companies early and we’re guilty of it — nobody has looked in the mirror harder than we did and course corrected.” The mechanism is a global offsite scoring the firm against the 50 best private companies: for how many are we “not just a passive shareholder but the investor of record” — “the most important conversation there is.”
• Accel has now invested in several Anthropic rounds, including the 180 round. There’s no squinting-to-3x model: “some of those businesses could be trillion-dollar companies,” with comps of Google, Microsoft, Amazon. The danger is ascribing Anthropic/Anduril/OpenAI/Stripe characteristics “to things that don’t really fit the paradigm at the Series A.”
• On Anthropic-versus-Pentagon, he frames it around principle rather than a calculated business move: “How can you not admire the founders for sticking to their conviction?… call me old-fashioned, but if you behave the right way you will be rewarded.” The reward is visible: a likely Mike Krieger’s ~1M/day net-new consumer signups and passing GPT in the App Store. Harry’s needle — “isn’t it ironic this is what was needed to surpass” — gets a flat no.
• His broader change of mind: a year ago he believed “all of the generational investments in AI had been made” after watching Dan Levine incubate Scale in 2016. “That was a really stupid thing to say and I no longer believe it” — outcomes will be bigger than he envisioned, there’s still time to participate, and “the innovation flywheel is just getting started.” On Scale’s $14.9B exit: appropriate congratulations, a full-throated thank-you to Alex Wang, “and then everybody gets the [expletive] back to work.”

8. Fund math inverted: ladder to 20% — and the singles-and-doubles fight

• Accel runs a $1.4B growth fund plus a larger, roughly $4B later-stage pool; to Harry’s “subscale versus David George’s six or seven and Josh’s nine,” Miles reframes: the market is 2000 inverted. Then you took 30% at the A and diluted to 20% at IPO; today “you do what the market will allow in the earliest possible investment, sponsor a tender, do a growth round, do an IPO round and ladder your way up to 20%” — only a multi-stage fund can.
• The sharpest disagreement of the episode: Arthur Patterson’s maxim — “focus on hitting singles and doubles and let the home runs take care of themselves,” because declaring $100B outcomes at the Series A makes you overswing. Harry rejects it outright: “That is not what venture is about… you want 30 swings — swing the [expletive] out of this.” Miles’s partial retreat: the real meaning is know what you’re good at and don’t “sling it into momentum-chasing opportunities where you’re not going to be any better than the next investor.” Bootstrapped outliers still exist — “we keep saying there’s no more bootstrap, and then you find a Laravel” — and Accel aspires to be best in the world at finding the 1Passwords in Little Rock.

9. The 2021 vintage’s endgame: Thoma Bravo time and the $2-5B IPO dead zone

• For the likely Snyk, Miro and 1Password class — great companies “done at such high prices” — the market “has gotten so humbling.” A $300M-revenue company growing 15%, last priced ~$7B, is “not a great setup” for peak-round investors, “but it’s a great business… there will be an outcome for that company.” His route: “it’s probably a good time to be in the Thoma Bravo, Vista, Blackstone, KKR business” — homes will differ from the founders’ original aspiration.
• On whether $2-10B IPOs are dead, he pegs the dead zone lower: companies going public in the $2-4/5B range “never really break out.” It’s not that investors are underwater — companies wait for “fairly clear line of sight to hitting the $5 billion threshold and trading beyond that, because it’s murky below that.” The SaaS apocalypse is repricing of future cash flows that isn’t wrong, “but this has been an over-rotation” — and the most oversold name he’ll give is Figma, even though Accel isn’t in it; Harry’s darker framing: Figma at $11B was “the swan song of software” and doesn’t return a growth fund.
• On taking chips off the table: first principle is what’s best for the company, but the CrowdStrike case argues for holding — Samir Gandhi and John Locke invested in 2011 at $160M post on $1M of software revenue plus a $9M consulting business, led the next two rounds, bought in the IPO, and it’s a $100B company. WeWork at $50B? “I probably would have been seeking liquidity.” Miro at $17B — Accel didn’t sell, because “what was the founder doing… that’s the only thing that matters.”
• Harry’s rant — “the casinoization of public markets, where a likely Citrini report can wipe billions off Anthropic and a security release impacts CrowdStrike — are you high?” — gets a shrug: “it’s not good or bad, it’s just a different asset class… I’m never going to be best in the world at understanding public stocks and I think that’s okay.”

10. Misses, the Linear stakeout, and who’s who at Accel

• Regrets, named: 11 Labs — “we haven’t spent enough time with the founder, which is our loss… that one stinks.” The regret he names: Shiv, likely at Abridge (“I think he’s generational” — he manipulated the seating chart at an AI dinner just to sit next to him). Losing versus not seeing? “They’re equally bad, but losing stings more.” A healthy term-sheet win rate is ~80%, and 100% claims mean “you’re not chasing competitive enough things.”
• The best win, as told: during “the worst couple of months I had experienced in a long time,” he stayed in a hotel near Linear founder Kari’s home in Del Mar — “a little bit creepy as I say it out loud” — flying back and forth for his best friend Craig’s birthday, waiting to see if Linear would raise. When Kari called, “it was pretty euphoric. I will always remember that week.”
• Accel’s internal depth chart: Christine Esserman best sourcer, Andrew Braccia best picker, Samir Gandhi best closer. Board wisdom via Ravi at Sequoia — “let me politely make an observation” — and the rule that “there’s generally an inverse correlation between how vocal somebody is and how helpful they actually are.” Founders need investors for the couple of big bumper decisions a year, not the little ones; and per a listener’s line Harry endorses, VCs are great at when to hire, awful at who.
• Quick fire: Liquid 2 at seed, Meritech at A (“great taste… trained at Summit Partners”), Thrive at growth (Miles Grimshaw, his “more articulate, intelligent VC alter ego,” sits on the Cursor board). Dream hire: talking Mike Cannon-Brookes into becoming an investor — “never bet against Mike.” Career advice from Arthur Patterson: professionalism sustained over decades — respect the rituals. What excites him most is the younger Accel bench; Harry’s closer: “Optimists make money. Pessimists are right.”

Verification Notes

• Raw captions do not clearly establish several resolved surnames or entities; references are qualified where needed.