Miles Clements on Cursor, AI Investing and Venture Fund Decisions

Harry Stebbings asks Accel partner Miles Clements how investors can find durable value when AI technology and revenues change quickly. Clements proposes two dimensions: how soon a customer gets value and how long that value lasts. His examples and assessments reflect his role as an investor in several companies discussed.

Clements says legal and accounting AI can take time to deploy because professionals must be persuaded to change workflows, but he believes the value may be durable once embedded. He mentions Basis, which Accel recently invested in. That is his investment thesis, not independent evidence of customer outcomes.

By contrast, he says some early vibe-coding products offered immediate value but did not retain users. He regards coding tools as attractive because they can deliver quick value and become embedded in team workflows. His claim that a user might become ten times more productive is illustrative and is not supported by a measured productivity study.

Stebbings cites criticism of Cursor, rising costs and reported revenue, asking whether Claude Code’s popularity threatens it. Clements says the market is expanding through new users and greater consumption, so one tool’s growth need not come entirely at another’s expense. Accel has invested in Cursor and Anthropic, making his assessment financially interested.

Clements argues that Cursor is not confined to its original editor interface and says its team has emphasized agents. He cites Cursor co-founder Michael Truell’s public figures, including an agent product growing 15 times in a year and cloud agents accounting for 35% of merged pull requests. Those are company-reported metrics, not independently audited measures of market share or quality.

On reliance on outside models, Clements argues that supporting multiple providers lets users choose and benefits Cursor when underlying models improve. He cites preliminary results from an Accel survey about developers changing models daily, but gives no sample, methodology or final report. The percentages should not be treated as representative evidence.

Clements says Cursor’s own model work is aimed at specialized coding tasks rather than every task a general-purpose model handles. He expects this to improve the product for professional users. This is a prediction by an investor, not independent proof that the company’s models will outperform alternatives.

Asked about investments at reported valuations of $9.5 billion and $27 billion, Clements describes a possible engineering platform spanning more of the development stack. He compares the ambition with companies that built valuable businesses in individual categories, including Atlassian, an Accel investment. His upside scenario depends on future execution and is not a valuation guarantee.

Clements estimates Accel’s first entry looked like roughly four to five times a year-end revenue run rate using later reported figures. The speakers cite a report that Cursor exceeded $2 billion in ARR; they do not independently verify the number or its accounting basis. Comparing a valuation with a subsequently achieved run rate does not establish the multiple at the time of investment.

He recalls a conversation in which Cursor’s team aspired to reach around $500 million in ARR by year-end and Accel modeled a more conservative $300 million from an approximate $100 million starting point. He says reported results exceeded those estimates. His recollection is approximate and the interview does not provide audited forecasts or financial statements.

Clements says forecasts are useful because they encode assumptions about products, pricing and customer penetration, even when the precise outcome is hard to predict. He is less concerned about a modest variance from plan than whether the underlying assumptions remain sound. His approach reflects a private investor’s priorities, not a universal financial rule.

Stebbings asks whether extraordinary AI growth makes a triple-triple-double-double trajectory uninteresting. Clements says no: founder quality, market, price and ownership matter as well as growth. He argues a portfolio can include both widely recognized breakout companies and less obvious opportunities, depending on the investment terms.

The pair disagree over opportunity cost. Stebbings asks why a fund would allocate capital to slower growth when 15-fold growth is available; Clements says a single growth number omits risk and ownership. He sees value in combining different strategies rather than only buying momentum or refusing all expensive rounds.

Stebbings challenges whether Accel’s fund size requires very large outcomes. Clements points to the growth in the number of very large public and private companies, based on his approximate figures, and believes multistage firms can make substantial returns. Both acknowledge such outcomes are rare or difficult; a larger market capitalization does not ensure a fund return.

They discuss vertical software growing more slowly than the hottest AI companies. Clements would not reject it solely for its growth profile, although he agrees the potential outcome must be large enough for the fund. Stebbings says his own team increasingly looks for products that replace labor and have a path to much larger revenue.

Clements recalls missing ServiceTitan after applying strict limits to the forward revenue multiple for vertical software. He says deeper understanding of the market and what the company could disrupt might have justified a higher price. His subsequent valuation comparison is an estimate; hindsight does not establish what a different investment would have returned.

Stebbings asks whether calling Cursor an engineering platform conflicts with the idea that there can be more than one winner. Clements says winning matters but argues a growing market can support multiple substantial firms. His market-share and company examples are broad comparisons rather than a legal or economic determination of monopoly power.

Clements cites payroll and workforce software to question declarations that one young company has already won an entire market. He contrasts reported ARR at Deel with larger incumbents, using figures he does not verify in the interview. He says Accel invested in Remote, while Stebbings says he invested in both Deel and Remote.

Clements says missing Rippling still stings. He admires founder Parker Conrad’s ability to build adjacent revenue opportunities, such as equipment provision, into a broader product. This is his assessment of product strategy, not a verified comparison of company unit economics; Accel’s position in Remote is relevant to the discussion.

Asked why Accel did not invest, Clements says the team moved too slowly and adhered to valuation and ownership rules. He also refers to a past reputation but explicitly declines to judge whether it was deserved. He does not provide sufficient detail to establish any underlying allegation, and regards the missed opportunity as a possible case for flexibility.

Stebbings asks whether current Series A prices justify breaking ownership rules. Clements quotes a mentor’s distinction between the science of valuing a company and the art of knowing when to break rules. He says the meaning of an A round has changed and exceptions should be rare; investors do not have to participate in every deal.

Clements describes a trap he has experienced: extrapolating a spectacular quarter into a permanent growth trend. A company can appear to have found product-market fit, only for the quarter to prove anomalous. He says competitive pressure increases the temptation to make a decision from limited early data.

He argues that strong growth can hide underlying weaknesses, so investors should inspect how intensely customers use a product and how usage changes. This matters for both enterprise and consumer businesses in his view. The principle is a diligence recommendation, not proof of any particular company’s durability.

Stebbings describes the stress of a market that can feel binary. Clements says he enjoys the competition for founders and deals, while acknowledging no investor has a perfect record. He says Accel reviews companies it missed to understand errors and improve future decisions.

Asked why Accel missed early investments in foundation-model companies, Clements says the firm recognized the miss and course-corrected. He describes partnership meetings that review important private companies, where Accel has meaningful relationships, and which future companies it wants to pursue. The account describes the firm’s process, not verified results.

Clements says the aspiration is complete market coverage and a perfect win rate, but acknowledges nobody achieves both. Asked for Accel’s actual term-sheet win rate, he says he does not know and speculates that 80% would be healthy. The hypothetical benchmark is not a reported firm statistic.

He says losing a competitive deal can mean a firm is pursuing difficult opportunities, while finding overlooked founders can also be rewarding. He points to bootstrapped companies outside Silicon Valley as a continuing opportunity, though harder to identify. His claim that Accel excels at finding them is a self-assessment.

The speakers discuss Accel’s approximately $1.4 billion growth fund and a larger later-stage pool, using approximate sizes. Clements says multistage investors can build ownership through early rounds, tenders, growth rounds and an IPO rather than acquiring all of it at the start. Actual ownership, prices and fund returns depend on each transaction.

Stebbings questions whether larger eventual outcomes will compensate for smaller ownership and points to a reported Figma valuation. Clements invokes an Accel founder’s advice to focus on consistent execution and let exceptional outcomes emerge. Stebbings disagrees, saying venture funds must deliberately seek very large outcomes; neither approach guarantees success.

Asked whether Accel itself chases momentum, Clements says some momentum is justified when business fundamentals support it. He cites Anthropic as an important technology partner to portfolio companies and confirms that Accel invested in several rounds. The firm’s financial and commercial relationships are central context for his positive assessment.

Clements says some private companies could plausibly become trillion-dollar businesses, and acknowledges that investing at a reported $180 billion valuation implicitly requires a very large outcome. That is a speculative underwriting scenario, not a valuation forecast or promise of a tenfold return. He warns against applying the same assumptions to every early-stage startup.

Stebbings asks whether the Anthropic investment looks better or worse after its dispute with the US Department of Defense. Clements says he respects the founders for maintaining their stated convictions, but explicitly says he does not know how the dispute will end. His position is that of a shareholder; the discussion does not independently adjudicate the dispute.

The pair turn to companies whose previous private valuations were high relative to slower growth. Stebbings gives approximate figures for Snyk; Clements says a good product and customer base may still support an outcome even if it disappoints investors who entered at a high price. The figures and eventual outcome are not verified.

Clements thinks the market reset may create opportunities for leveraged buyout firms and that some companies may find different homes from those originally envisioned by founders. This is a market prediction, not a report of specific transactions. He emphasizes how management and investors behave when conditions become difficult.

Both speakers value founder leadership, citing companies they know. Clements also acknowledges that professional CEOs can be excellent and gives an example. His praise for particular leaders is subjective and tied in some cases to Accel’s portfolio, not evidence that all founder-led companies outperform.

Clements says he changed his mind about believing all generational AI investments had already been made. He now thinks existing companies could grow larger than he expected and new opportunities remain. He cites Accel’s early relationship with Scale AI and discusses a reported transaction; his forecast and figures are not independently established.

Asked about companies providing data and services to model developers, he says he struggles with very high recurring-revenue multiples for services businesses but recognizes their strategic value. The conversation does not establish comparable margins, revenue definitions or a fair valuation for any named company.

Stebbings asks why a founder would go public when employees can obtain liquidity and companies can conduct acquisitions while private. Clements agrees those alternatives exist for some leading private firms, but says many other businesses still need public markets. He distinguishes the options available to a small set of companies from those available to everyone.

Clements says some companies listing at a few billion dollars have struggled to break out and suggests founders may wait until they have clearer visibility to a larger valuation. This is his interpretation of market dynamics, not a reliable threshold for an IPO. He regards the software sell-off as partly justified by changed cash-flow expectations but believes it has gone too far.

He names Figma as a company he believes is oversold while saying Accel is not an investor. That is a personal view based on his assessment of its founder and financial profile, not independent valuation analysis or a recommendation. Stebbings then asks when investors should sell shares in private companies.

Clements says liquidity decisions should first consider what is appropriate for the company and founders. If a company offers a tender and permits investor participation, he thinks diversification may be sensible. Stebbings challenges the potential conflict between an investor’s need to return capital and a founder’s preference for the investor to hold.

Clements agrees the answer is situational: he says he might have sought liquidity in a hypothetical WeWork holding at a very high valuation. He also says Accel did not sell Miro at a cited private valuation, emphasizing the founder’s course. Neither counterfactual establishes what an investor should do in another case.

He contrasts that with CrowdStrike, where he says Accel invested early, followed on and bought in the IPO. He credits holding for allowing the firm to participate in later growth. The company value and transaction history are his account; the example does not guarantee that holding a different asset will produce similar returns.

Asked whether venture firms should manage evergreen public positions, Clements says multistage funds can support unusual founders for a long time, but not every company compounds indefinitely. He distinguishes technology investing from public-stock picking and says he does not expect Accel to be the best public-equity selector. He rejects a blanket holding rule.

Stebbings asks which Accel colleagues excel at sourcing, selecting and winning deals. Clements names colleagues and praises their persistence, judgment and ability to connect with founders. These are his assessments of teammates, not independently measured rankings of investment skill.

Do the best founders need investors? Clements says “need” is too strong. He sees a useful investor as a sounding board for a few consequential decisions each year, such as a partnership, acquisition or pivot, while avoiding micromanagement of ordinary product and hiring decisions.

On boards, Clements praises a colleague’s mix of operating and investing experience and the humility with which he offers feedback. He advises founders to be cautious of directors who dominate discussion or insist on teaching management something at every meeting. The guidance reflects his personal experience, not a validated measure of board effectiveness.

In the quick-fire round, Clements names venture firms and colleagues he respects across seed, growth and later stages. He also notes working with another investor on Cursor’s board. These preferences reflect professional relationships and competition, not independent recommendations to invest in the funds.

Asked whom he would add to Accel, he first names Atlassian’s Mike Cannon-Brookes in a hypothetical scenario, then praises an investor at Andreessen Horowitz. He emphasizes technical judgment and personal rapport. Accel’s investment in Atlassian and their professional connections are relevant to these favorable opinions.

For someone beginning a venture career, Clements repeats advice from Accel co-founder Arthur Patterson: maintain professionalism over long periods, respect the investment process and take partnership and portfolio-review rituals seriously. He presents the advice as a professional habit rather than a formula for returns.

Clements says Accel regrets not participating in ElevenLabs and wishes it had spent more time with the founder. He says both losing a deal and failing to see it are costly, though losing can feel worse. He also recalls wishing to work with the founder of Abridge; these are personal assessments, not verified performance forecasts.

Clements describes winning the chance to work with Linear as personally meaningful during a difficult period. He stayed near its founder in Southern California to be available for conversations while balancing family and friends. He says the founder’s eventual decision to work with Accel made that week memorable; both speakers have disclosed interests in Linear.

Stebbings says investors repeatedly sought introductions to Linear and Lovable before financing rounds, and discloses an investment in Lovable. The exchange illustrates how demand for access can burden founders and intermediaries. It does not establish that investment interest predicts either company’s future performance.

Asked what excites him about the future, Clements says he is most optimistic about younger members of Accel’s team and their development over the next decade. He says he cannot reliably predict the world that far ahead. The interview closes with thanks from both speakers.