The Fortnite Circle Is Shrinking: How AI Agents Are Maiming SaaS, and Why Ghost GDP Should Worry Everyone

Only one publicly traded B2B company has built a competitive AI agent. One. That claim from Jason Lemkin sets the tone for this episode of 20VC, where SaaS investor Lemkin and Scale Venture Partners GP Rory O’Driscoll clash over the speed and depth of AI-driven disruption, with host Harry Stebbings escalating the stakes at every turn.

The week’s landscape

The conversation opens against a backdrop of cascading market events: Anthropic’s new security product wiped $20 billion off cybersecurity stocks in a single day, the Citrini Research “2028 Global Intelligence Crisis” report shook public markets, and OpenAI leaked slides showing plans to double spending to $665 billion by 2030. Lemkin, founder of SaaStr and veteran SaaS investor (portfolio includes Algolia, Talkdesk, RevenueCat), brings the perspective of someone who runs security audits from an airplane using Claude Code. O’Driscoll, who led investments in Bill.com, Box, and DocuSign, brings the public-markets discipline of valuation and historical cycles. The two agree on more than they initially appear to, but their disagreements about timing and magnitude are where the real insight lives.

Anthropic’s security shockwave: old news, new panic

Anthropic’s security product release hammered CrowdStrike, Cloudflare, and other cybersecurity stocks. Lemkin’s reaction was not surprise at the capability, but surprise at the market’s reaction. He had literally run a full security audit on an airplane using Replit (powered by Claude Code) the week before, including static code review and penetration testing. He sent the results to Replit’s own engineering team, and they didn’t even know their product had become that good.

“When you are priced for perfection, anything less than perfection will be a kick in the nuts.”

O’Driscoll’s framing was different: technically unsurprising, commercially overhyped. He cited an HSBC report arguing that “software is the means by which AI will diffuse into the enterprise,” meaning companies like CrowdStrike will be the ones bringing AI security capabilities to customers, not Anthropic directly. But he conceded the valuation problem: CrowdStrike was trading at 16x revenues with 22% growth and 31% cash margins even after the correction. At that price, any increase in tail risk hits hard.

O’Driscoll added a dry observation: Anthropic markets itself as “the nice AI company,” yet creates stock market casualties comparable to weaponry.

“Maybe the Pentagon is wrong and they need to buy more Anthropic and just point it at the enemy. It’ll bring China to its knees, right?”

The Fortnite metaphor: SaaS territory keeps shrinking

Lemkin’s central metaphor for the episode was Fortnite’s shrinking circle. Claude, he argued, keeps consuming more of every company’s territory. You’re stuck on an ever-smaller island where you must capture more market share just to survive.

His headline claim: among all publicly traded B2B companies, only Palantir has a competitive AI agent. No other public company has seen a single ounce of revenue acceleration from their AI agents. Everything else is lip service.

He explained why public companies can’t build great agents with three structural reasons. First, every agent today is essentially custom, requiring training, data cleansing, and deployment; nobody at these companies wants to do this work. Second, the talent gap: the average customer success person with a green/yellow/red dashboard cannot train and tune an agent. Third, platform breadth is the enemy of agent quality. Companies like Shopify or Monday covering 100+ verticals can’t build specific agents that automatically serve churches, basketball courts, and refrigerator businesses.

“If you look at all the publicly traded B2B companies, there’s only one that has a competitive agent. It’s Palantir.”

O’Driscoll offered a four-path framework for how AI intelligence reaches enterprises: (1) buying directly from Claude, (2) building internally, (3) existing incumbents integrating AI, (4) new startups building on foundation models. He believes paths 3 and 4 are most likely. Foundation model companies won’t directly serve every church and basketball court. Lemkin agreed that not all revenue flows to foundation models, but insisted: if you don’t own the agents in your category, you’ll be maimed.

He also revealed that two investments he made last year, which he loves, no longer have a reason to exist in their prior form because of Claude. The products haven’t gone bankrupt because customers are renewing, but their standalone value proposition has evaporated.

“I have two investments I made last year I love them, those products no longer have a reason to exist today because of Claude.”

DoorDash and the agent intermediary

The Citrini report suggested agents would replace DoorDash. O’Driscoll called this absurd. DoorDash is a massive combination of logistics, restaurant onboarding, and customer service. No software replaces that. He compared it to Netflix’s recommendation engine, which has been running for 10-15 years but only drives 5-10% of blind viewing choices. Consumers don’t delegate personal decisions like food to algorithms.

Lemkin countered with a direct quote from DoorDash CTO Andy Fang, who publicly stated that “agent commerce will be transformative to our industry” and that DoorDash needs to “earn the right to service customers’ agents end to end.” He also cited data from over 10,000 restaurants through an investment that confirms consumers are willing to let agents make recommendations.

Stebbings cut to the synthesis: agents don’t need to replace DoorDash. They just need to become the intermediary layer between consumers and platforms, choosing optimally between DoorDash, Uber Eats, and ordering direct, weakening platform pricing power.

Lemkin used YouTube as a decisive counterargument: YouTube is the best recommendation engine on Earth, channels and followers don’t matter anymore, and it gets better every single day at knowing what you want to watch. That’s an entirely recommendation-driven consumer experience, and it’s utterly disruptive.

“Almost all the B2B software we use today is terrible now because AI software is so good.”

The three ultimately agreed: creating an entirely new AI competitor to replace DoorDash is “clickbait stupid.” But agents as a decision intermediary will maim growth, and that’s all it takes when you’re priced for perfection.

Ghost GDP: productivity without consumers

Ghost GDP was the episode’s most provocative concept. Lemkin defined it simply: AI productivity gains don’t flow to human workers who then spend it. Productivity rises, profits concentrate among fewer people, but there aren’t enough consumers left to drive the economy.

His own company is the laboratory. SaaStr went from 12 people to 2-3 while maintaining eight-figure annual revenue. Their 10 AI agents (Repley, Art, Qualie, Monty, and others) generate millions in revenue but buy nothing.

“Our 10 agents at SaaStr generate millions of revenue, but they buy nothing. They buy nothing except tokens.”

O’Driscoll pushed back hard on the macro framing. Productivity gains have been the engine of growth for 200 years. 80% of people used to work on farms; now it’s 4%. The rest do other work, and the sum total of human achievement expanded. You can’t argue against productivity being good over the arc of history.

Stebbings cited Japan in the 1990s as a counter-example: massive mechanized productivity gains that didn’t disperse to the broader population, leading to long-term economic stagnation. Lemkin reinforced this with firsthand experience. At a VIP dinner in Japan last November with B2B founders (IPO $20M+ revenue only), everyone was discussing how their seat base naturally shrinks each year. This was before the current AI panic; it’s structurally driven.

Lemkin then let slip a remarkably candid admission:

“I think in the short term this is all great for us as investors… we should put the money in the bank and flee to Miami or Monaco because I’m not sure it’s good for everybody.”

O’Driscoll’s sharpest rebuttal was a framing exercise: separate micro from macro. At the micro level, B2B software is genuinely threatened. But the entire software industry is only about 2% of US GDP. Even if it were completely wiped out, the impact wouldn’t match the car industry’s collapse 10-15 years ago.

“We didn’t bleed in Silicon Valley when the car industry went down the toilet. Don’t hold your breath thinking they’re going to come for us.”

The employment math

Stebbings estimated 30-40 million jobs could be affected when combining customer support, legal, bookkeeping, and Waymo’s impact on driving within 4 years. Lemkin asked Claude to model a 50% tech headcount reduction scenario. The output: $600-900 billion in GDP impact, 4-5 million total jobs lost including multiplier effects, and local economic devastation in 5-6 tech-concentrated cities. It would be one of the largest economic shocks in US history outside of a world war or pandemic.

Shopify provides a real-world signal: three years without adding a single net employee while growing revenue 50% to $12 billion. Lemkin’s observation was subtle but important: not growing headcount for a decade is losing headcount too, because revenue-per-employee keeps climbing while the multiplier effect on local economies stagnates.

“One of the leaders in the next 12 months is going to do an Elon Musk and just cut half their team in one day.”

Leveraged PE-owned SaaS: the first dominos

O’Driscoll identified which companies will fall first: highly leveraged PE-backed SaaS companies at 6x EBITDA debt with single-digit growth. If you’re unleveraged and public, you can follow Shopify’s Tobi playbook (freeze headcount, rely on growth). But the math doesn’t solve for leveraged companies. The cascade: debt inverted, equity wiped, creditors extend to avoid crystallizing losses, dramatic headcount cuts, reduced R&D investment, less attractive as a vendor, slow customer churn due to contract inertia, 5-year painful grind to death.

Lemkin predicted the consolidation response: 5-8 startups at $50-200M revenue each, bundled together at 1.5-2x revenue, creating “Frankenstein B2B companies.” Twenty unicorns merged into one entity, possibly attempting an IPO in 2027. O’Driscoll agreed with the direction but was skeptical about the IPO timeline.

OpenAI’s spending bet

OpenAI plans to double spending to $665 billion by 2030 while raising revenue forecasts 27% to $280 billion, including hardware, ads, and products that largely don’t exist today. O’Driscoll dissected the revenue stack: the consumer subscription product that exists, the enterprise product trailing Anthropic, agentic products that don’t yet exist, and $30-77 billion in “non-subscription consumer monetization” (essentially ads) layered on top.

He noted an interesting narrative shift: Claude now gets the benefit of the doubt while OpenAI has gone from darling to questioned. “The truth is probably no one is ever as good or bad as they seem.” 2023-24 were good years for OpenAI; 2025 was a good year for Anthropic. In a 10-year race, two years up and one year down isn’t fatal.

Lemkin’s read was pragmatic: the leaked slides were for believers (SoftBank and others). If you want to believe, you’ll believe. If you’re a skeptic, you take a marker and delete the unrealized revenue bars. It felt like “a startup board meeting on steroids.”

Figma fights back

Figma’s Q4 2025 numbers were flawless: $1.22 billion ARR, growing 40% year-over-year (accelerating from 38% in Q3), $10K+ customer retention at 97% GRR and 136% NDR, stock up 15% post-earnings.

O’Driscoll held this up as what fighting back looks like. Dylan Field clearly knows what to do: add AI capabilities, extend from design to coding. Figma cites native integration with Claude Code as one of its biggest growth drivers. In a sector where AI-native disruption is among the most credible threats, this is a generational talent entrepreneur at his peak, punching back.

Lemkin was conflicted. The quarter was epic, but “we’ve given up on the present, we’re all panicked about the future.” He believes that within 8-18 months, Claude Code will automatically generate designs as elegant as a professional designer’s. Claude has ingested every website and mobile app on Earth. Right now Claude Code websites “all look the same, same purple color scheme, same icons,” but only because they haven’t focused on it yet.

This is the Fortnite metaphor playing out in real time: Figma’s biggest growth driver is integration with the very thing that might eventually subsume it.

Momentum vs. value: four stock picks

Lemkin revealed a shift from value hunting to momentum investing. He identified only five B2B software stocks up over the past year: Palantir, Figma, Cloudflare, Shopify, and one other. His decision: only buy winners.

O’Driscoll placed this in academic context: over 6-18 months, momentum plays work. Over 5 years, value plays work. The trick is knowing which regime you’re in. Jason is explicitly calling this a momentum market.

Two comparisons crystallized the tension. Klaviyo vs. Shopify: Klaviyo is essentially a Shopify derivative, yet Shopify is up 2.63% while Klaviyo is down 58% over the past year. O’Driscoll argued Klaviyo is a trap, because Shopify’s Tobi must build agents that absorb Klaviyo’s market to survive.

The more striking comparison: Palantir (up 41%) vs. Atlassian (down 74.85%). Atlassian is accelerating from 20% to 23% growth at $6.3 billion revenue, making it the only beaten-down stock that’s actually accelerating. O’Driscoll called it “the greatest dislocation in the market”: biggest price decline, most revenue acceleration, still above-the-fold quality.

Jack Altman joins Benchmark

Jack Altman (Sam Altman’s brother) gave up his own fund Alt Cap (raised $400M total in two years, essentially solo GP) to join Benchmark. O’Driscoll saw it as Benchmark’s consistent 15-20 year playbook: offer equal partnership with high autonomy, attract proven talent.

Lemkin found the signal more interesting. 95% of VCs dream of having what Jack had: $400M, complete autonomy. Even just managing the fees provides a comfortable lifestyle. That he gave it up says something about 2026: the difficulty of going it alone is great enough that people will sacrifice independence for institutional backing.

Lemkin then shared his own counter-story: a top mega-fund once offered to “make him whole” with guaranteed $10M per exit and $2-3M annual salary when he’d just raised his first $70M fund. He didn’t consider it for 10 minutes.

“I didn’t sell my last company to go work for somebody.”

Some Thoughts

This episode’s value isn’t in predicting which SaaS companies live or die. It’s in the collision between two investment epistemologies. Lemkin is a founder-turned-investor whose judgments come from hands-on experience (running a security audit on an airplane, watching his own headcount shrink from 12 to 2). O’Driscoll is a classic public-markets analyst who views the same facts through valuation, historical cycles, and macroeconomics. Both are right within their own frames, and the tension between those frames is more instructive than either frame alone.

• The most revealing moment came during the Ghost GDP discussion, when Lemkin said, almost in passing, “In the short term this is all great for us as investors… I’m not sure it’s good for everybody.” This level of candor is rare. The beneficiaries know they’re benefiting, and they know it may come at others’ expense.
• O’Driscoll’s “2% of GDP” rebuttal deserves serious consideration. If the entire B2B software industry were wiped out, the broader economy’s response might well be “I’m willing to lose those guys.” The existential stakes that dominate tech Twitter may not register outside the bubble.
• All three participants default to “agents” as the core form of next-generation software, yet each defines “agent” differently. Lemkin’s agent is a Claude Code-like system completing tasks end-to-end. O’Driscoll’s is closer to traditional software with embedded AI. Stebbings’ is an intermediary layer between consumers and platforms. This definitional gap directly shapes their divergent conclusions about SaaS survival.
• Lemkin’s Fortnite metaphor, while playful, captures something the standard “disruption” framing misses: it’s not about replacement, it’s about surface area. Companies don’t die; they just occupy progressively less of the value chain.
• The Atlassian case (accelerating growth, stock down 75%) may be the single most actionable insight in the episode. If O’Driscoll is right that value plays win on 5-year horizons, and the stock is the only beaten-down name that’s actually accelerating, it represents a rare intersection of analytical frameworks.