Navan, Canva and the AI Investment Boom

Harry Stebbings, Jason Lemkin and Rory O’Driscoll discuss AI talent acquisitions, legal software, potential IPOs and the future of enterprise tools. Their company figures are those cited during the conversation, while proposed transactions and future outcomes remain subject to confirmation.

The panel starts with reports that Nat Friedman and Daniel Gross could join Meta after earlier AI-related deals. O’Driscoll frames the question around why Meta might pay for talent, why individuals might sell and how labor-market rules shape competition. He is interpreting reported developments rather than confirming their terms.

O’Driscoll speculates that Meta’s greatest concern is an AI assistant capturing time and attention that users now spend on its social products. He treats large AI spending as a form of insurance against a potential platform shift, not proof that Meta will lose users or that the investment will work.

He compares the possible AI effort with Meta’s past virtual-reality spending and guesses at a $100 billion budget. The amount is his rough estimate, not a disclosed Meta budget; success and the appropriate level of spending are uncertain.

The speakers cite a third-party comparison of recent mobile downloads for ChatGPT and major social apps as a possible signal of changing attention. App downloads are not the same as active usage, revenue or time spent, and the numbers are attributed to the discussion.

O’Driscoll argues that experience building leading models has become valuable in the talent market. He discusses California’s restrictions on noncompete agreements as one factor allowing people to move and create new companies; his historical examples and interpretation are not a legal opinion about any individual contract.

The panel discusses reports of very large offers to AI researchers. O’Driscoll says public claims about offers could also affect expectations among employees, but admits he does not know whether any particular communication strategy was intended. The compensation figures are not independently verified.

Stebbings says he was offered $75 million for his wholly owned media company but declined, recalling his mother’s question about what he would do after a sale. He says he would have started a similar show. The account and offer amount are his own, not verified transaction documents.

Lemkin worries that people leaving founders, employees or limited partners for lucrative new opportunities may weaken loyalty. O’Driscoll emphasizes that a transaction can generate returns for investors and that people respond to market incentives. They disagree about the responsibilities that accompany a move.

The group turns to a reported Harvey round of $300 million at a $5 billion valuation. They cite conflicting revenue estimates and explicitly say they do not know the actual figure; the $100 million end-of-year number is discussed as an expectation rather than achieved revenue.

O’Driscoll says many legal AI demonstrations looked compelling and credits Harvey with establishing early recognition among law firms. He relates an anecdote from a customer reference and offers his own assessment of product evolution; it is not an independent study of legal quality or adoption.

The investors debate what market size could support the valuation. O’Driscoll distinguishes software spending, spending on legal information and potentially automating some legal work. The idea that AI can capture part of labor spending is a hypothesis, not proof of legal accuracy, savings or future market share.

Lemkin describes another legal startup with rapid reported ARR and says his own experience suggests some customers value usefulness more than the sophistication of the underlying model. O’Driscoll replies that integration into a law firm’s existing systems and the cost of switching can make distribution valuable.

Stebbings notes competition from Legora and discloses that his firm invested in Solve, which addresses patent work. He argues the legal market can be fragmented into specific workflows, so a broad market estimate may not belong to a single vendor. His investment is relevant to that view.

O’Driscoll separates two questions: can software automate work people currently perform, and can its provider retain enough of the value? Competition could pass the savings to customers even if automation works. The calculations are an investment framework, not a measured outcome.

Lemkin says he uses Claude to review documents and then asks counsel to confirm the result, including in a particular limited-partner transfer example. He reports that counsel agreed in that instance; the anecdote does not establish that AI legal answers are always correct or replace professional review.

O’Driscoll imagines routine document review becoming faster and cheaper, while incumbent firms and new AI-enabled firms compete. Whether the savings, legal reliability and customer adoption materialize as described remains uncertain.

Stebbings argues that patent, immigration and other specialist tools could divide the market. O’Driscoll agrees that focused products may serve distinct processes while a general corporate-law tool addresses another segment. Their descriptions are market hypotheses, not verified addressable-market counts.

Lemkin predicts AI will expose slow or low-value knowledge work, while O’Driscoll expects pressure on inefficient processes. They are debating possible changes to services and employment, not reporting that specific professionals have been displaced.

The panel moves to Navan’s IPO filing and cites a previous valuation in the vicinity of $10 billion. The discussion concerns a filing and potential future listing, not proof that the IPO has priced or completed.

Lemkin observes strong recent IPO performance and expects more eligible companies to pursue listings. O’Driscoll thinks US markets can absorb supply while investor demand lasts, but acknowledges the cycle could change. His remarks are predictions based on conditions at the time.

The investors discuss whether prospective public-market prices and valuations can motivate private companies to list. A favorable price is one consideration alongside liquidity, reporting obligations and the burden of operating as a public company.

Circle’s sharp post-IPO price moves prompt debate over speculation and limited public float. The panel’s quoted share prices are historical points from the discussion; a fast price rise does not by itself establish a company’s fundamental value or predict subsequent trading.

O’Driscoll says neither bankers nor investors can know a stock’s future price with certainty. He warns that buying mainly because someone else may pay more can end badly, while distinguishing that market behavior from the underlying company’s business.

Stebbings asks why Canva might remain private when public markets appear receptive. O’Driscoll says relative cost of capital can influence the decision, but concedes that a fleeting high price alone would be a weak reason to accept the lasting obligations of a listing.

The speakers say Canva is generating substantial revenue and discuss a possible scenario in which cash flow allows it to repurchase shares or pay dividends. Lemkin’s suggested future free cash flow, buybacks and payouts are illustrative calculations, not confirmed Canva plans or reported results.

O’Driscoll suggests that focusing on AI development without the distraction of a listing may also be rational. The panel does not claim to know Canva’s private board deliberations or establish a date for an IPO.

The group turns to Oracle and Larry Ellison’s ownership, buybacks and infrastructure spending. They cite approximate shareholding and capital-expenditure figures from memory and interpret the change as a strategic bet; those figures and the reasons for decisions are not independently verified.

Lemkin discusses Cluely and his interest in real-time assistance for sales teams. He sees potential for a consumer-like product to support go-to-market work but says he does not know whether this particular company will win. Later in the discussion, Stebbings says Lemkin has invested; that relationship is relevant to his enthusiasm.

Stebbings questions whether provocative marketing has gone too far. The panel separates the question of attracting attention from whether the product works and whether particular marketing tactics help the business. Their descriptions are opinions about marketing, not a finding of misconduct.

O’Driscoll summarizes an outside essay arguing that an AI company might market ahead of its current capabilities while model quality improves. He treats it as a provocative framework. The essay’s characterizations of Harvey and Lovable are opinions, not established descriptions of their technology or evidence that either company deceived customers.

The investors discuss the hypothetical terms on which they might invest in a sales-assistance startup. Lemkin says he would need to meet the founders and understand their plans. These are illustrative check sizes and valuations, not an announced financing.

The panel discusses reported restrictions on Slack access and their possible impact on third-party AI tools. Lemkin predicts established enterprise vendors will increasingly restrict access; O’Driscoll questions whether customers would tolerate that and speculates that a paid access model could emerge. Neither presents this as a confirmed policy outcome.

O’Driscoll asks how customers should compare an incumbent’s own agents with third-party products. He suggests measuring task resolution and cost, using hypothetical rates to show why performance matters. The sample percentages are not reported results for Salesforce or any named competitor.

In a prediction-market round, the speakers take different positions on whether OpenAI might accuse Microsoft of antitrust violations. They distinguish an allegation, a lawsuit and a successful legal claim; their bets are opinions about a future event, not evidence that a violation occurred.

They also speculate on whether the US government could take control of an AI project during 2025. Their answers are probability judgments made at the time, not announcements or verified policy plans.

The final prediction concerns a proposed Trump-branded phone. Lemkin says he has not seen evidence of production and describes using a tool during the conversation; the panel’s timing and manufacturing comments are speculation, not a verified supply-chain investigation.

O’Driscoll suggests that a company with overseas manufacturing exposure might consider limited US production for political reasons. The speakers close with jokes about the show and the sales-assistance product; those comments do not change the uncertainties around the companies discussed.