20VC on AI Competition, Founder Control and Software Valuations

Harry Stebbings opens with reports that Canva has lowered its growth expectations, asking whether generative AI threatens the no-code design market. Jason Lemkin and Rory O’Driscoll consider both AI’s impact on costs and whether customers will still need a dedicated design application when a chatbot can produce a usable image or presentation.

The investors distinguish a change in growth expectations from the deeper question of product demand. In their telling, AI can add costs while also changing what customers expect to make for themselves. A business can respond to a margin issue, but the harder challenge is defending an interface customers might bypass.

O’Driscoll resists treating a slowdown as a complete verdict on Canva. The panel considers the difference between customers making simple assets and enterprises using a broader workflow. They describe a range of possible outcomes rather than claiming to know how quickly either group will switch.

Lemkin is more concerned about products serving individuals and small teams. He says people around him have begun using AI tools for work previously done in separate applications. His observation is an anecdote, not evidence of the overall size of customer losses.

The debate turns to what a customer actually buys: an isolated image, a familiar interface, or the surrounding collaboration and production process. A chatbot may be sufficient for one request while a company with a repeatable design process may value consistency, permissions and shared work.

The speakers ask whether adding AI features to an existing application is enough when users already pay for a general AI assistant. An incumbent may have access to the same models, but distribution and the location of the user’s attention could matter as much as any one feature.

Lemkin describes AI bundling as a serious threat to tools with a narrow consumer use case. The others question whether the same logic applies to professional design and collaborative work. Their disagreement is over the durability of the workflow, not a demonstrated outcome for any company.

Figma offers a contrast because it is embedded in professional design teams. The panel asks whether enterprise collaboration creates protection against general-purpose assistants or whether generated interfaces will eventually reduce the need for seats there too. Neither side treats the answer as settled.

Stebbings asks how investors should value these differences. The group considers whether customer adoption, switching costs and an enterprise sales motion can offset pressure from products that generate results directly in chat. They emphasize that price and product quality can change rapidly.

The conversation moves from product risk to shareholder liquidity. O’Driscoll notes that a founder and a venture investor may face different incentives around a public listing or a sale: one may prefer more time to build, while the other must consider the life of a fund and returns to its investors.

The panel discusses the temptation to wait for a higher valuation and the cost of missing a window to sell. A company can be impressive and still offer a difficult investment decision if its valuation already assumes many years of success. Their discussion is opinion, not a recommendation to trade.

The group asks what founders fear when a general-purpose assistant bundles their product’s core task. A response might be to own more of a workflow, create distinctive output, or make the product valuable to groups rather than individuals. The question is whether those advantages remain meaningful as models improve.

The speakers consider creative aggregation and specialist tools alongside chatbot interfaces. Stebbings discloses an investment in Higgsfield, and Lemkin also describes a portfolio interest. Those holdings are relevant to their discussion of whether specialist creative products can retain users.

The panel compares standalone applications with an assistant that can route a request among tools. An attractive output alone may not establish a lasting relationship with a customer; distribution, the ability to iterate, and the workflow surrounding a generated asset all enter the competitive calculation.

They return to the pace of change. A product can have a strong brand and still face new competition for habitual use, while enterprise adoption may be slower and more complex than consumer experimentation. Both the speed and eventual extent of substitution remain uncertain.

Stebbings asks which pre-AI companies made the right strategic bets in 2023 and 2024. The investors discuss companies that committed resources early, contrasting a serious product investment with merely adding a marketing layer to an existing application.

Intercom comes up as an example of a company that pursued AI in customer support. The panel notes its investment connections to the company. The lesson they draw is that a difficult strategic choice can take time to show up in results, and early commitment is not a guarantee of success.

O’Driscoll asks how much can really be learned from individual examples. In his view, the difficult question is whether a business had to change its core model to benefit from AI or whether the same underlying demand would grow as customers built more software.

The investors distinguish AI-native applications from infrastructure providers that can benefit from additional usage. They discuss Datadog and Cloudflare as examples of businesses whose existing position may capture some of that demand, while recognizing that valuations and customer spending also matter.

The group considers whether older software companies can successfully incorporate new technology without undermining their existing economics. Some can use their customer relationships as an advantage; others may face a costly transition if new products cannibalize high-margin recurring revenue.

Palantir enters the discussion as an example of a company that made a long-term bet on deploying software closely with customers. Lemkin praises its execution and the role of forward-deployed teams, while the speakers’ descriptions of growth and market reception remain their own assessments.

A forward-deployed approach, in the panel’s account, is not simply about handing over a model. It includes working with a customer’s data and operations to make an application useful. That depth can create a valuable relationship, but it also requires people and sustained execution.

The panel challenges the idea that all winners fit the same template. Some sell a tool directly, some sell infrastructure, and others embed with customers. The relevant test for an investor is how the company captures value from adoption, not whether its presentation uses the latest AI vocabulary.

Discussion returns to the valuation of a private design company and how quickly changing growth expectations can alter the story investors tell. The figures raised are estimates and hypotheticals, not verified current financial results. The panel disagrees about how severe the competitive shift will be.

One concern is that a company can wait too long to confront a changed narrative. The investors discuss how management communicates a slowdown, the tradeoff between near-term monetization and customer goodwill, and whether a public market would reward or punish a reset.

A venture fund must also decide when to realize gains. The speakers describe the pressure to hold the largest winners against the reality that limited partners may need distributions. The right decision depends on the remaining upside, the size of the position and the fund’s circumstances.

The discussion turns to the power law in venture investing: a small number of exceptional outcomes can account for much of a fund’s return. The panel notes that this logic can favor holding winners, but that mathematical possibility alone does not resolve the real risk of valuation declines.

Private marks can make a portfolio look strong without putting cash in investors’ hands. The group contrasts paper value with the ability to sell, and asks whether another financing round changes the underlying business or merely sets a new reference price.

Lemkin argues for being candid when the operating picture changes, even if the news is uncomfortable. The panel’s larger point is that optimistic market narratives and a company’s actual ability to grow profitably need to be examined separately.

Before moving to talent, the investors acknowledge that their judgments about company prospects are shaped by their portfolios and experiences. Their examples illustrate competing investment frameworks rather than a reliable forecast of any individual company’s outcome.

Stebbings raises reports of changes involving Google AI leaders Jeff Dean and Demis Hassabis. The group discusses what a large organization can offer senior researchers and what they might seek elsewhere. Its suggested explanations are speculation; the speakers do not establish either individual’s private motives.

O’Driscoll suggests that a researcher with substantial wealth may value autonomy, scientific ambition and access to compute differently from someone earlier in a career. Stebbings presses him on whether those assumptions accurately describe particular people. The participants do not have direct evidence of their decisions.

The conversation considers how a major company balances scientific research with commercial pressure. A researcher may want freedom to pursue a long-term agenda, while a company must allocate resources across products, revenue and infrastructure. The tradeoff varies by person and organization.

The investors also discuss the experience of senior executives inside large companies. Lemkin draws on his own time in a large organization to describe the constraints of managing a business unit. He does not claim that his experience explains the decisions of Google’s researchers.

Stebbings asks what a change among well-known researchers would mean for Google itself. The group distinguishes the contribution of exceptional individuals from the systems, teams and accumulated expertise that can remain. They disagree over how much strategic damage or opportunity such changes might create.

The panel debates whether commercial AI products and ambitious scientific research necessarily compete for the same leadership attention. One view is that a company needs leaders motivated by the demands of operating products at scale; another stresses that strong research and product execution can coexist.

The discussion broadens to other AI labs and recruiting. A startup may promise ownership, autonomy and a chance to build a new organization, while an established lab may offer resources, colleagues and access to large-scale computing. Different researchers can rationally make different choices.

The investors describe an unusually competitive market for a small number of highly sought-after researchers. Reported packages at the very top are not representative of typical employees, and the panel treats much of the compensation discussion as estimates rather than verified contracts.

The distinction between cash and equity matters. A headline package may include shares whose eventual value depends on the company’s performance, vesting and liquidity. The speakers warn that comparing a possible future stock outcome with a guaranteed salary can produce a misleading impression.

They discuss how extraordinary returns from early AI equity can distort expectations among subsequent hires. O’Driscoll argues that outliers attract attention precisely because they are unusual; a later employee should not assume the same multiple is available again.

Stebbings asks whether very high compensation is sustainable for more than a handful of people. The panel distinguishes ordinary employees, sought-after AI specialists and a tiny group of perceived superstars, while noting that market conditions and company fortunes could change.

A startup offering meaningful equity may recruit people who care about both financial upside and influence over the work. The investors also recognize the risk: a valuable-looking grant can be diluted, fail to vest or prove difficult to sell.

The group argues that attention to exceptional pay can obscure the wider labor market. They contrast a few large technology packages with the incomes of people in communities where infrastructure projects may be built, and transition to the politics of those projects.

The panel debates local opposition to data centers and proposals affecting their approval. Lemkin argues that constraints could impede infrastructure development and that construction and operation can bring jobs. Those are his policy judgments; benefits and impacts depend on the specific project and community.

The speakers acknowledge that residents may be uncertain about what they receive and what they bear. O’Driscoll argues that developers need to explain the deal clearly, including how electricity costs will be handled, rather than simply asserting that a project benefits the country.

O’Driscoll suggests that protections against higher electricity bills and a local dividend might improve support. He presents these as possible policy designs, not as evidence that any particular project has offered them. Actual power, water and environmental effects vary by location.

The group contrasts state and local approaches, arguing that different jurisdictions may make different choices. They also identify access to power and equipment as practical constraints. Their optimism about competition between jurisdictions is a political view, not a guarantee that infrastructure concerns will be resolved.

Stebbings turns to the proposed TerraFab semiconductor project associated with Elon Musk. He cites reported spending and potential jobs, while the panel discusses the ambition of building capacity in response to constrained chip supply. The project’s final cost, output and employment are not established by the conversation.

O’Driscoll describes the plan as a highly integrated bet, spanning infrastructure, manufacturing and downstream applications. He points to previous integration in Musk’s businesses but argues that the scale adds risk, especially if AI spending slows. That is a scenario, not a forecast of the project’s outcome.

Another argument is that firms may pursue their own capacity because they cannot secure enough chips on acceptable terms. The panel discusses supply-chain bottlenecks and mentions Intel’s involvement and financing based on reports. The speakers explicitly acknowledge that at least some of those details were not verified.

Stebbings introduces a reported, not yet implemented, Revolut founder compensation proposal, described as tying additional ownership to company valuations. The figures and terms discussed are not confirmed by the panel. He asks whether such packages could become more common.

O’Driscoll says unusually large equity packages are a specialized instrument, not a normal template for every public company. They may be designed to retain a founder whose existing equity has vested, but a board must weigh the cost to other shareholders against the benefits of retention.

He questions packages based only on stock price. A chief executive can execute well during a falling market or benefit from a rising market without creating the expected operational improvement. He favors performance conditions that specify what the company must actually achieve.

The panel also considers control. Lemkin suggests that a founder may care about both ownership and the ability to direct the company, while O’Driscoll notes that extra voting rights and additional economic ownership are different things. Their discussion of individual motivations is inference, not confirmed intent.

Even formal control has limits, the investors argue. Shareholders can choose not to buy or hold a company’s stock, and a founder who controls decisions also bears responsibility for the consequences. The panel discusses how governance arrangements may affect both decision-making and valuation.

O’Driscoll says he has become more willing to accept founder control if it encourages strong companies to list publicly. He still sees a need to protect other shareholders and suggests that control arrangements may deserve a valuation discount. Lemkin emphasizes the burden public-company founders can feel.

The panel works through a hypothetical participation calculation based on the reported Revolut numbers. O’Driscoll considers the implied share of incremental value unusually high; Lemkin compares it with venture carry. The arithmetic depends on unconfirmed terms and assumed future valuations, not an actual payout.

Lemkin argues from his own portfolio that founder leadership can be especially valuable during a period of rapid change. He predicts poor outcomes for his non-founder-led investments, while acknowledging that counterexamples exist. This is his personal assessment, not a general rule about professional managers.

O’Driscoll challenges the open-ended cost of that argument. If additional ownership is granted every time a founder reaches a higher valuation, other shareholders face repeated dilution. The question for a board is how much of the value created should go to the executive and what protection remains for investors.

Lemkin says investors must model dilution more carefully. In his own seed-investing examples, he describes a higher effective entry price after later financings reduce ownership. That example is an illustration of how dilution changes returns, not a prediction for a particular investment.

The two find some common ground on incentives. O’Driscoll says he could support a very large award for exceptional performance if the operating milestones are carefully specified. Lemkin adds that timing matters: waiting until a company is struggling could make retaining its founder harder.

Stebbings asks about other financings and raises Whatnot as a chance to discuss a business outside the AI category. The group references a reported funding round and valuation, but its numbers are not independently verified in the conversation.

The panel describes Whatnot as a live-shopping marketplace and compares the format with television shopping and eBay. O’Driscoll says the idea is intuitive because people enjoy watching sellers demonstrate goods. That enthusiasm is his view of the business rather than proof of future performance.

The speakers cite estimates of gross merchandise volume and a possible marketplace take rate. GMV is the value of goods transacted, not the company’s own revenue. The growth and take-rate figures are presented as their estimates, and the conversation does not verify them.

Lemkin argues that investors should examine businesses serving durable needs outside AI, including commerce, restaurants and finance. AI may change how these businesses operate, but he does not believe every attractive opportunity is an AI application.

The group jokes about applying software or AI valuation multiples to shopping volume. Beneath the joke is a warning about definitions: treating transaction volume as revenue can inflate a comparison. They question whether investors are sufficiently attentive to the quality and composition of reported growth.

The discussion moves to reported software earnings, including Shopify and Atlassian. O’Driscoll says operating performance can change investor sentiment, while stressing that an infrastructure provider facing a customer spending change and an application company recovering from a weak valuation have different stories.

The panel’s comments about customer spending, revenue multiples and share movements are its own descriptions and estimates. They use those examples to argue that a single label such as ‘software’ can hide very different business models and sources of risk.

Lemkin highlights changes to free Loom seats and higher-priced features at Canva as signs he interprets as pressure to monetize existing customers. He cautions that a strong quarter can still require difficult decisions. The panel’s interpretation of management motives is not independently established.

O’Driscoll agrees that aggressive pricing or overage charges may be hard to sustain if they damage relationships. Lemkin adds that he admires Atlassian’s leadership but wonders whether AI agents will ultimately reduce demand for some developer seats. The impact remains a prediction.

Stebbings asks whether HubSpot might become an acquisition target. O’Driscoll discloses that his firm invested in an earlier round. Lemkin says a transaction is possible but does not claim an offer exists; any discussion of a buyer, financing or price is hypothetical.

Lemkin sees a more immediate competitive issue for small-business software: new entrants can build capable products quickly and sell into niches once dominated by established vendors. He argues that the challenge is not necessarily customers building everything themselves, but an expanding field of specialist competitors.

The panel discusses customer relationship management as an example. Incumbents that previously competed upward against larger rivals may now face younger companies attacking simpler customer segments. The speakers’ impressions of the new competitors’ growth and capabilities are anecdotal and not audited results.

O’Driscoll suggests that established software owners could seek out young teams and acquire products or talent to accelerate change. Lemkin responds that fast-growing companies are already pursuing similar acquisitions, so incumbents may have difficulty competing for the same teams. The conversation ends on that unresolved strategic challenge.