Anthropic's $149B Bet, SaaS Seat Collapse, and Atlassian's Survival Playbook

Anthropic quietly released a projection: $149 billion in ARR by 2029. OpenAI’s estimate for the same year sits at $180 billion. Together, that’s $350 billion out of a $700 billion global software market. The math demands a question: where does all that money come from?

The Episode

Harry Stebbings brings together three people with very different vantage points on the AI-SaaS collision. Mike Cannon-Brookes, co-founder and co-CEO of Atlassian, represents the incumbent: a company with 250,000 customers, $4+ billion in revenue, and a stock price that’s been beaten down by AI anxiety. Jason Lemkin, one of the most prominent SaaS investors of the past decade, brings the investor’s lens on what’s actually happening to software companies in the field. Rory O’Driscoll, a General Partner at Scale Venture Partners with a portfolio of public SaaS winners (Bill.com, Box, DocuSign), adds the growth-stage perspective on what’s repricing and why.

The conversation covers Anthropic’s revenue ambitions, Harvey’s $11B valuation, customer support as an AI battleground, Super Bowl ad economics, and what it takes to lead a public software company through a paradigm shift. The tone shifts between analytical debate and something closer to therapy for SaaS operators.

$149 Billion: Where Does Anthropic’s Revenue Come From?

Anthropic’s most optimistic internal projection puts them at $149B ARR by 2029. Combined with OpenAI’s $180B target, that’s roughly half the entire global software market.

Rory frames the core question: is this net-new demand, or does it cannibalize existing software budgets? His analysis: total worldwide software revenue is $700 billion. If Anthropic and OpenAI alone capture $350B, either the overall market has to expand dramatically or existing software companies face massive compression.

Jason argues the TAM expansion is real but unevenly distributed. Every category outside engineering and product faces “existential risk of shrinking seats.” His reasoning: when AI can handle the work of three SDRs or two customer support agents, companies don’t buy more seats, they buy fewer. The per-seat SaaS model breaks down precisely because AI makes individual workers more productive.

Mike pushes back on the existential framing. “The idea that software as a category is dead is ludicrous to me.” His argument: software has always gone through cycles of disruption and reinvention. The internet didn’t kill software; mobile didn’t kill software; cloud didn’t kill software. AI will transform how software is built and delivered, but the category itself will grow.

The nuance emerges in their disagreement about timing. Jason sees immediate pressure: SaaS companies that can’t demonstrate AI value in 2026 will face brutal repricing. Rory points to the capital structure problem: private AI companies operate with no marginal cost discipline, funded by investors willing to burn cash for market share, while public companies must deliver EPS growth every quarter. Mike acknowledges the unfairness but argues it’s ultimately a capital allocation challenge, not a death sentence.

Harvey at $11B: The Vertical AI Valuation Question

Harvey, the AI legal company, raised $200 million at an $11 billion valuation. The panel dissects what this signals.

Jason’s take is blunt: this is “classic top of the bubble stuff.” His framework: investors are paying for consensus, not fundamentals. Harvey is the perceived winner in legal AI, so capital flows in at any price. It’s the venture capital equivalent of “you can’t get fired for buying IBM.” He specifically flags the 1x liquidation preference as a tell: investors are paying premium valuations while requiring downside protection, which means even they don’t fully believe the price.

Rory agrees on the valuation concern but sees a structural dynamic worth noting. Many vertical AI companies are growing at extraordinary rates (10-20x year-over-year), which makes traditional valuation frameworks break down. The question isn’t whether Harvey is overvalued today, it’s whether AI-native legal tools can capture enough of the legal services market to justify the trajectory.

Mike offers the operator’s perspective: the valuations only make sense if these companies can capture workflow, not just intelligence. A copilot that helps lawyers draft documents faster is a feature; a platform that restructures how legal work gets done is a company. He’s skeptical that most vertical AI startups are building the latter.

Every Non-Engineering Seat Is at Risk

Jason drops the most provocative claim: “Every category that I know of outside of engineering and product is at existential risk of shrinking seats.”

His evidence comes from the field. Customer success teams are being cut by 30-50% as AI handles routine interactions. SDR teams are shrinking because AI can do outbound at scale. Marketing teams are smaller because content generation is automated. The common thread: any role that involves processing information, generating text, or following playbooks is compressible.

The counterargument, which Mike makes forcefully, is that compression of roles doesn’t mean compression of software spend. If a customer success team shrinks from 10 to 5 people but each person uses more sophisticated tools, the per-seat value could increase even as the seat count drops. The total spend might stay flat or even grow.

Rory introduces a timing nuance: the seat compression is already happening, but the per-seat value expansion hasn’t materialized yet. That creates a dangerous gap where SaaS companies see declining metrics before they can demonstrate new value.

“I just think we have to give up on TAM. We just have to let the revenue show us the path to TAM.”

Customer Support: Terrible or Terrific Investment?

Sierra has hit $150M ARR in AI customer support. The debate: is this a great category or a terrible one?

Jason argues it’s terrific but dangerously crowded. Customer support is the most obvious AI use case because the workflows are structured, the data is available, and the ROI is immediately measurable. Sierra, Intercom, Ada, and a dozen others are all growing rapidly. The problem: when every AI company targets the same use case, margins compress and differentiation becomes nearly impossible.

Rory takes the other side: customer support might be the “first battle” but not the war. The real value is in companies that use support as a wedge to own the entire customer relationship layer. Sierra’s advantage isn’t that they do support well; it’s that they’re building a platform for customer-facing AI interactions across the entire journey.

Mike, as someone who actually runs customer support at scale, adds practical nuance. Atlassian has been deploying AI aggressively in their support operations. The results are genuinely transformative: faster resolution, higher customer satisfaction, lower cost. But he warns that the transformation is harder than it looks. Bad AI support creates worse outcomes than bad human support, because customers get stuck in loops with no escalation path.

The Super Bowl Ad Economics

Anthropic ran a Super Bowl ad. So did Google, Meta, OpenAI, and a slate of AI startups. The discussion reveals how differently each panelist sees the economics.

Rory’s take: Super Bowl ads for consumer brands (Doritos, beer) are rational because they reach massive audiences with measurable sales impact. For enterprise software companies, it’s mostly ego. The CEO wants the prestige of a Super Bowl ad, so the marketing team finds a justification.

Mike disagrees partially. For some companies like Wix, which sell to small businesses and consumers, a Super Bowl ad is a totally sensible investment. For enterprise-focused AI companies, the signal matters more than the direct sales impact. Anthropic’s ad wasn’t about generating leads; it was about establishing itself as a household name alongside Google and OpenAI.

Jason’s contribution: the Super Bowl ad phenomenon reveals something about capital discipline. When companies are spending $7-8 million on a single ad spot (plus production costs), it tells you they’ve already spent on everything else. It’s the marginal dollar at work. Companies that are still fighting for product-market fit or scaling their go-to-market wouldn’t spend that money on brand. The companies running Super Bowl ads are the ones that have surplus capital and are looking for incremental awareness.

Mike makes the sharpest observation: “That ego stuff happens when capital is not being efficiently spent in a business.” He’s seen it before across tech cycles. When capital is abundant and accountability is loose, vanity projects proliferate. When capital tightens, those same companies suddenly discover that brand awareness doesn’t solve retention problems.

The Public Company Handicap

Mike is remarkably candid about Atlassian’s position. The company delivered strong quarterly numbers, but the stock price has dropped significantly. The market has repriced all SaaS companies on fear of AI disruption, regardless of individual performance.

The tension he describes: public companies must simultaneously invest heavily in AI transformation and deliver short-term earnings growth. Private competitors face no such constraint. They can burn cash, offer below-cost pricing, and hire aggressively without answering to quarterly expectations.

But Mike flips the narrative. “We’re a better company because we’re a public company.” The discipline of public markets forced Atlassian to become better at forecasting, planning, and executing. The danger is when that discipline replaces strategy, when companies optimize for EPS at the expense of long-term investment.

His framing of Atlassian’s position is striking: “If we refounded Atlassian today with 350,000 customers, 50,000 enterprise customers, a couple of billion bucks in the bank, and 10,000 people in R&D and a great distribution engine… you’d be like, it’s a pretty good starting point to start a business.”

Rory highlights the SBC (stock-based compensation) asymmetry. Public companies get scrutinized for every dollar of SBC, while private companies hand out equity “like candy” with no comparable accountability. Both are diluting shareholders, but only one side faces public pressure for it.

CEO Mental Health in the AI Era

The conversation takes an unexpectedly personal turn. Mike reveals he starts working at 5 a.m. every day now. His co-founder Scott Farquhar retired a year and a half ago after 23 years, adding to the workload. The AI disruption has intensified the pace for every CEO he talks to.

Jason observes that not every SaaS CEO has the stomach for this transition. He mentions one CEO who quit at the end of the year, at hundreds of millions in revenue, with the goodbye message: “I was told AI wasn’t important in our space.”

Mike’s advice to these CEOs is unusually thoughtful for a tech podcast: “Dude, you just got to accept reality and then go build something.” He distinguishes between the founders who should push through (they’re energized by the challenge) and those who should step aside (they’re exhausted and pretending). He calls the latter “a hotter call to make” but ultimately more admirable than “hanging on waiting to get the headshot.”

“Part of creation is destruction. You got to go build value for customers. You got to go build new products and technologies and services. We’re going to create our way out of this problem.”

His philosophy on balance: spend time with your kids, walk in the trees, maintain relationships outside work. “If you spend 100% of your time at work, you make worse decisions.” But he’s not naive about it: he doesn’t spend 60 hours a week with his kids, and he doesn’t spend 100 hours a week at work. The intentional trade-offs are what matter.

Rory shares his partner’s framework: “Our business is a learning business, and the day you stop learning is the day you should say to your partners, in about a year and a half you should be replacing me.”

Some Thoughts

This episode works because Mike Cannon-Brookes isn’t playing the typical CEO game. He doesn’t pretend Atlassian is immune to AI disruption, doesn’t minimize the stock price pain, and doesn’t hide behind buzzword strategies. He simply says: we have customers, we have engineers, we have capital, let’s go build.

A few things worth sitting with:

• The $149B + $180B = $350B math against a $700B global software market is the single most clarifying data point for understanding the SaaS repricing. If even half of that materializes, the displacement is enormous.
• Jason’s “every non-engineering seat is at existential risk” framework is more useful than generic “AI will change everything” takes because it identifies specifically where the compression happens: support, sales, marketing. The roles that process information and follow playbooks.
• Mike’s distinction between capital discipline and strategy is the most actionable insight for SaaS leaders. Public companies that mistake cost-cutting for transformation will be caught in the worst of both worlds: shrinking revenue without new product capabilities.
• The most telling moment might be the CEO who quit saying “I was told AI wasn’t important in our space.” That’s not a personal failure; it’s an organizational information failure. The CEO was surrounded by people who told him what he wanted to hear.