Anthropic, OpenClaw and Software Valuations: A 20VC Investor Roundtable

Harry Stebbings joins SaaStr founder Jason Lemkin and Scale Venture Partners general partner Rory O’Driscoll to discuss AI investment, software valuations, autonomous agents and leadership. The investors offer opinions and time-specific estimates; private-company figures, transaction speculation and stock forecasts are not independently verified in the conversation.

Stebbings introduces Anthropic’s reported $30 billion raise at a $380 billion post-money valuation. Lemkin says the size and interest reflect a venture market concentrating on a few exceptional companies as public software shares fall. He says he does not have an Anthropic allocation; the valuation does not by itself establish investment returns.

O’Driscoll points to rapid appreciation across previous private rounds and argues that multi-stage funds want exposure to one of the strongest sources of momentum. Stebbings says a $30 million early-stage round can attract much more demand. The examples describe the participants’ experience, not a representative measurement of all venture financing.

Lemkin compares the pull of a small number of fast-growing AI companies to gravity drawing capital away from other technology businesses. He cites an Anthropic coding-product revenue run-rate estimate as an example of escape velocity. O’Driscoll thinks investors may extrapolate too far, even as the narrative can dominate prices in the short term.

O’Driscoll describes extremely rapid reported growth at Anthropic and says both run rate and recognized revenue should be considered. He calls the growth historically unusual but contrasts its heavy losses and capital requirements with profitable early technology companies. The historical comparison and private figures are his estimates, not independently audited results.

Stebbings asks whether investors price in the danger of a badly timed compute commitment. O’Driscoll recalls Anthropic CEO Dario Amodei describing risks on both sides: underinvest and rivals could pull ahead; overinvest and capacity might not earn its cost. The discussion treats this as a business risk, not a prediction that the company will fail.

O’Driscoll argues that model builders resemble capital-intensive semiconductor businesses more than traditional cash-generating software companies. Reaching free cash flow might require slower investment and growth, changing how investors value revenue. His estimates of Anthropic’s current and future GAAP revenue are tentative, and forecasts are not reported earnings.

Asked whether Anthropic is taking customers from OpenAI, O’Driscoll says he lacks data on token usage and notes that enterprises can use both providers. He characterizes OpenAI as more consumer-oriented and Anthropic as more enterprise and coding-oriented. Lemkin sees competitive pressure on budgets; their revenue comparisons do not prove direct customer losses.

Stebbings asks about software spending in 2030. O’Driscoll says the answer depends on whether AI delivers real productivity and labor savings: sustained faster growth requires customers to see a return on their spending. A future slowdown in enterprise spending would be one possible warning sign, not a forecast of a specific outcome.

Lemkin believes many enterprises are choosing to pursue smaller teams and AI adoption even before the return is fully demonstrated. He stresses that a large employer cannot replace thousands of workers instantly. O’Driscoll agrees that companies appear committed to trying AI but distinguishes making the bet from being satisfied with its results.

O’Driscoll expects a period of substantial AI spending followed, perhaps, by retrenchment if companies overinvest. He compares AI to earlier executive priorities such as efficiency, the internet and cloud. His proposed one-to-two-year spending window and later adjustment are judgments, not verified budgets.

Lemkin worries that efforts to shrink headcount could reduce demand for software sold per employee. He cites executives’ ambitions as examples and says this makes him reluctant to call a bottom in public software stocks. Those anecdotes do not establish that a particular company will cut staff or that all seat-based products will decline.

O’Driscoll agrees about sentiment but argues that price eventually matters: even a company facing difficulty could become attractive at some valuation if future cash flows support it. His hypothetical annual-return range is not a promise. Lemkin counters that already-slowing growth could leave little room if AI further weakens demand.

O’Driscoll describes a shift in market narrative from enthusiasm for software to enthusiasm for AI, while cautioning that the businesses inside the software basket are different. He considers Shopify’s commerce and payments infrastructure less obviously displaced in the near term than some workflow tools. This is his assessment at the time, not a guarantee about Shopify.

Lemkin is also positive on Shopify but outlines a bear case: agents or conversational shopping might sit between customers and merchants, weakening the value of the storefront interface. Shopify could still provide payment and merchant infrastructure. Stebbings questions whether consumers will prefer conversation over browsing and discovery.

Lemkin says agents can use a platform without a person regularly opening its interface, using his own Salesforce usage as an example. O’Driscoll replies that it is easy to imagine a threat and difficult to disprove it; investors must assess probability as well as the narrative. Their examples do not demonstrate which interface consumers will ultimately use.

Lemkin describes uneven outcomes in his portfolio: some businesses benefit from AI, one has mixed effects, and others face pressure. These are his observations about a particular fund, not a representative sample. O’Driscoll distinguishes older software from newer AI applications that can also be sold as software subscriptions.

O’Driscoll says established public software companies must maintain financial discipline while investing in AI; smaller private companies may lack the same scale and resources. He sees differences among workflow software, payments and other categories. Lemkin says some growing portfolio companies may decide to build sustainable businesses without another venture round.

The panel turns to Figma. O’Driscoll offers approximate contemporaneous valuation and growth figures and argues its price reflects both current results and uncertainty about future competition. He says markets may pay less for the same business when narrative enthusiasm fades; the numbers are not current quotes.

Lemkin praises Figma’s existing growth but argues it missed an opportunity in product prototyping now served by Replit and Lovable. His estimate of revenue Figma might have captured is explicitly a guess, not measured lost revenue. He says the speed of the new products illustrates how quickly a strong incumbent can face a competitive threat.

O’Driscoll agrees that the design and prototyping market is close to AI capabilities, but says exposure differs by category. He expects creative work and coding to change sooner than some structured applications such as accounting. His projected valuations and timing are scenarios, not established outcomes.

Lemkin discloses an investment in Monaco, an AI sales-development company, and describes using its product. He initially calls an opportunity a six-figure deal, then clarifies that the agent booked a meeting concerning a possible six-figure sponsorship; it did not close the sale. His demand and performance claims are based on his own experience and were not independently verified.

Lemkin says fast improvement also makes the vendors themselves vulnerable if customers switch to a better agent. O’Driscoll, whose firm has invested in Reggie, agrees sales agents have improved through iteration but says adoption has not moved at the same pace across every software market.

The guests disagree about how quickly competitive pressure will reach established software. Lemkin uses Figma and new prototyping products to illustrate rapid change; his estimates of market opportunity remain speculative. O’Driscoll argues that time to adapt matters and does not expect the transition to be uniform or complete within two years.

O’Driscoll says legal applications and healthcare documentation are examples where AI found demand in markets without an equally compelling older software alternative. Other structured applications may move more slowly. The comparison describes his investment framework, not a measured timetable for disruption.

Lemkin suggests investing in strong founders in categories changing more slowly, giving them more time to adapt. Stebbings says that approach may require patient funding. O’Driscoll prefers identifying the next market before it becomes obvious but not so early that capital must wait indefinitely; none of the strategies guarantees returns.

Stebbings says he sees law firms actively adopting AI in response to competitive pricing. O’Driscoll recalls earlier sales cycles in which emphasizing AI could deter customers, while now corporate leaders face pressure to have an AI strategy. Their experiences illustrate incentives to spend, not proof that every implementation produces savings.

The discussion shifts to Stripe’s reported private valuation of roughly $140 billion and Adyen’s lower public valuation. O’Driscoll says a simple comparison of market capitalization misses differences in business size, growth, profitability and available information. His approximate financial figures should not be treated as equivalent audited measures.

Stebbings suggests that communication and narrative may also affect how the companies are valued. O’Driscoll prefers Adyen in the hypothetical comparison because public financial data permits value analysis; Lemkin prefers Stripe’s flexibility as a private company. These are personal views at the recording date, not investment recommendations.

O’Driscoll predicts that major capital-intensive private companies may eventually seek public funding, while Lemkin values the freedom private companies have to invest without the same immediate market pressure. They describe an uneven public market that rewards some high-momentum stories but can be difficult for companies managing transitions. IPO predictions are uncertain.

Stebbings raises the news that OpenClaw creator Peter Steinberger is joining OpenAI and that the open-source project is to move to a foundation. Although the video title and conversation use acquisition language, the discussion does not establish the transaction structure. The guests speculate about a price and competing interest without confirming terms.

Lemkin says an earlier agent deleted his database and describes his concerns about software with broad access and continuing autonomy. At the same time, he sees value in the developer enthusiasm around OpenClaw. His claims about why other companies did or did not build comparable tools are interpretations, not verified accounts of their internal decisions.

The panel considers the costs of running agents frequently. Lemkin speculates that smaller models, intermittent operation and local hardware could help manage inference expense, while acknowledging the limits of his expertise. Costs depend on workload and configuration; the comparison is not a pricing guarantee.

O’Driscoll says businesses need to balance agent capability with controls over access to data and devices. Lemkin anticipates products that perform tasks while users are away, citing a proposed future Replit release as an example. His description of future functionality and timelines is a prediction, not confirmation of a shipped feature.

Lemkin makes claims about security incidents and Anthropic’s response to the project’s earlier name. Those claims and his account of the companies’ motives are not independently established in the interview. Both guests nevertheless agree that autonomous access can create serious safety and data-protection questions.

Stebbings asks whether responsibility for safeguards belongs to the agent vendor, the data holder or an independent provider. Lemkin says enterprise leaders already face responsibility for the actions of their products. O’Driscoll distinguishes accountability when something goes wrong from the separate question of which tools can prevent it.

O’Driscoll worries that an enterprise agent with broad access could expose customer data and create severe consequences for its provider. He expects demand for agent security products. Stebbings says his firm is about to invest in one, a financial interest relevant to the discussion; the guests do not evaluate a specific product.

Stebbings brings up Thrive’s roughly $10 billion fund and asks how large venture funds might become. O’Driscoll argues that financing very large private companies requires much larger pools of capital than early-stage venture. He points to Thrive’s investments in Stripe, Databricks and OpenAI as examples of its growth strategy.

O’Driscoll argues that investors might access such businesses more cheaply if they were public, while private ownership creates demand for large intermediaries. Lemkin describes a strategy of continuing to invest in later rounds of winners. The group notes that such a strategy still depends on the prices and economics working out.

Stebbings discusses Arif Janmohamed leaving Lightspeed to start a firm and asks whether departures reflect a desire to return to early-stage investing. Lemkin emphasizes economics; Stebbings and O’Driscoll also suggest autonomy, bureaucracy and preference for working with founders. They do not claim to know Janmohamed’s personal reasons or terms.

Lemkin says leaving a venture firm can carry an economic cost because fund interests may vest over years, although terms vary by firm. O’Driscoll adds that even without forfeiture, a departing partner may leave future opportunities behind. Their general discussion does not establish the contractual terms of any named individual.

The panel turns to Workday cofounder Aneel Bhusri’s return as CEO. Lemkin argues that founders can drive rapid change in a period of disruption. O’Driscoll considers whether product-roadmap questions may favor someone with deep knowledge of the original design, but says the reasons for the leadership change and its results are not established in the discussion.

O’Driscoll distinguishes generic operating skills from the specific knowledge a flexible founder might bring to redesigning a core product. Asked whether the return will lift Workday’s shares, he says he lacks a developed opinion and sees no guaranteed quick fix for a mature market. Lemkin expects faster decisions but acknowledges that the bets may fail.

Lemkin says organizational resistance can make change difficult and describes his conversations with executives. Stebbings cites a sales-compensation example from an ElevenLabs interview, and O’Driscoll argues that strong customer demand can support unusually large sales quotas. These examples do not establish how every employee or company behaves.

Asked which founder might return to a public-company CEO role next, the guests consider several possibilities and admit they lack a convincing candidate. O’Driscoll notes his prior board relationship with Box while praising persistent founders. Their comments about individuals’ future choices are speculation, not reports of planned appointments.

Stebbings asks about Monday.com after citing a market value of roughly $3.8 billion and its reported revenue, growth and guidance. O’Driscoll says assessing it requires cash flow, stock-based compensation and assumptions about durable growth. The conversation’s operating-income figure is not interchangeable with free cash flow, and these are historical figures.

The guests say a low valuation could be attractive if growth remains durable, but could be misleading if AI weakens the product’s position. Lemkin compares Monday.com and Salesforce and expects smaller customers to change or leave faster than large enterprises. These are conditional assessments rather than verified forecasts of churn or investment advice.

Lemkin says he considered buying a basket of public software stocks for the show but did not execute the purchases because he could not see a floor. Stebbings offers to match a proposed $200,000 exercise, with four stocks selected later. The exchange is a proposal, not evidence that a trade occurred, and Lemkin’s assumptions about losses and tax treatment do not limit actual risk.

The group agrees to revisit the proposed selections in roughly two weeks. O’Driscoll says the exercise forces investors to confront what they actually believe; Lemkin repeats that he had not bought the stocks. They close without naming a completed portfolio or establishing any future return.