From SaaS to AI-First: How Companies Are Reshaping Innovation
Summary
The indiscriminate “SaaSpocalypse” trade mistakes a real long-run shift for an immediate extinction event. Gil argues that a fleet-management product like Samsara—with in-cab hardware, distribution, enterprise sales, and support—will not simply be replaced by a weekend vibe-coded app, though support can use vibe agents. Guo says Fortune 100 change management and security make a weekend-built CRM likely implausible. Decagon and Sierra do demonstrate a genuine move from per-seat software toward usage-priced support agents, but “this isn’t gonna be every single SaaS company.”
AI-native companies show that code can become abundant without making the rest of company-building abundant. Guo cites portfolio companies with hundreds of millions of dollars of revenue and fewer than 50 engineers, yet rapidly expanding from zero to nearly 100 salespeople: “Vibe sales is not happening.” Gil expects engineering productivity to be absorbed by enormous unmet software demand rather than simply eliminating engineering teams.
The investable engineering bottleneck may shift from writing code to allocating trustworthy human attention. If agents generate enormous volumes that nobody reads, “nobody deeply understands the code base,” leaving production systems fragile and full of “vibe coding slop.” Guo calls agent-first engineering management, testing, smart review, and formal verification “open season around this really, really big problem.”
A month of AI hype blurred partnerships, demos, and planted behavior into evidence of autonomous markets. Gil disputes claims that agents already choose vendors: partnership defaults that provision particular tools resemble Airtable silently running on AWS, not independent purchasing judgment. The “Malt book” Reddit-like forum example also seemed partly human-generated or planted for marketing. Guo’s sharper equity-research critique is that “the theory of competitive advantage didn’t just like poof, disappear”; production, distribution, change management, and product completeness still matter.
The underlying economics remain extraordinary even after stripping away the hype. Using Capital IQ data and projections, Gil’s team puts the AI labs’ journey from $1 billion to $10 billion of revenue at roughly one year, versus about 20-plus years for ADP and Adobe, while public projections imply $10 billion to $100 billion could take the labs three to five years. Meanwhile, GPT-4-equivalent pricing fell from roughly $37 to $0.25 per million tokens in 21 months—150x—and o1-equivalent pricing from $26 in December 2024 to $0.30 in November 2025—88x in 11 months.
AI could pull far more GDP into technology while concentrating much of the resulting value in a power-law head. Gil says the leading tech companies’ share rose from roughly 4% of US GDP in 2005 to about 12% today, with plausible 2035 scenarios ranging from 15%-20% to 30%; the top eight tech companies already represent about $23 trillion and well over half of the S&P’s value. Guo argues that expanding technological surface area could let the tail dominate, while Gil says there may be more $100 billion businesses but “the head and torso aggregate almost all the value.”
Faster growth does not confer SaaS-era durability: AI may compress a decade-long displacement cycle into one or two years. Gil warns that many startups get only about a 12-month window at peak value, so boards should schedule unemotional exit reviews once or twice annually; only a very small group should “never, ever, ever sell.” The strongest defense is a bundle spanning five or 10 aspects of the same vertical or application, reinforced by platforms, ecosystems, networks, or hardware—because “if every two years is 10 years,” a cloneable point product is a precarious control point.
Deep dive
1. The SaaS selloff extrapolates startup behavior onto enterprises
Gil separates the durable trend from the timing: AI may transform software over decades, but the near-term belief that companies will replace every paid application with vibe-coded internal tools is “incredibly shortsighted.” The market correction contains truth, yet its company-level conclusions are often wrong.
Samsara is his clean counterexample. Fleet operators are not going to replace its broad application surface with a weekend vibe-coded app; the product also includes in-cab camera hardware, distribution, enterprise sales, and ongoing support, although Gil imagines support using vibe agents.
The five-person startup with a custom CRM proves less than enthusiasts think: previously, it might simply have used a spreadsheet. Guo asks whether the same engineer wants to manage Bank of America’s security reviews, workflow debates, maintenance, and organization-wide change management. “The answer is, like, probably not.”
The genuine disruption is narrower but consequential. Gil points to Decagon and Sierra shifting customer support from per-seat software toward utilization-based agents; those businesses may pressure an earlier software cohort without implying that “every single SaaS company” disappears.
2. Code abundance moves the constraint to people, judgment, and trust
Gil’s demand-side case is that software remains radically undersupplied relative to what organizations want. Large productivity gains therefore get “soaked up” by more products and features, although engineers motivated by artisanal code craftsmanship may experience the transition very differently from those who treat code as a utility for building products.
Guo adds an identity risk: work once considered technically difficult or high-status may be comparatively easy for agents. Her borrowed advice from an Applied Intuition founder—“Keep your identity small”—is a prescription for adaptability when both the enjoyable work and its professional status are changing.
The harder operational problem begins when agents produce code faster than humans can understand it. Guo fears “vibe coding slop in my actual production code base,” with quality uncertain and fragility accumulating; the opportunity is software that manages scarce human attention through testing, intelligent review, agents, or formal verification.
3. AI’s capabilities are revolutionary; the latest narratives were not
Gil rejects claims that agents already make meaningful vendor-purchasing decisions. A commercial partnership that causes software to provision a Superbase instance is no more evidence of autonomous purchasing than Airtable invisibly creating AWS infrastructure: “That’s always happened.” True agentic commerce might come later, but it requires understanding the buyer’s persona and needs.
His assessment of the recent news cycle is deliberately blunt: “I think we had a month of kind of bullshit hype.” A lot of the “Malt book” material seemed human-generated, and he thought some of the forum behavior may have been planted for marketing, while media treated emotionally resonant behavior as evidence that autonomous agents were already excluding humans.
Guo’s pushback on bearish equity research is that a compelling demo is not the complete software customers require. Cheap code does not erase distribution or structural advantage; it lowers the cost of expressing a point of view in software, allowing more competing ideas about engineering, productivity, and other workflows to reach the market.
4. Revenue is accelerating while equivalent intelligence gets radically cheaper
Gil’s team uses Capital IQ data and projections to compress software history: ADP and Adobe needed about 20-plus years to grow from $1 billion to $10 billion of revenue; Salesforce and SAP took roughly eight or nine, Microsoft seven or eight, and Google, Meta, and AWS approximately three to five. The AI labs were put at “roughly a year.”
The public—not necessarily company-supplied—projections are equally aggressive beyond $10 billion. Microsoft took about 27 years to reach $100 billion; Google, AWS, and Meta required over a decade or roughly that long. The AI labs are projected to cross the same interval in approximately three, four, or five years.
Cost curves are moving in the opposite direction. GPT-4-equivalent model pricing fell from around $37 per million tokens to $0.25 in 21 months, a 150x reduction; o1-equivalent pricing dropped from roughly $26 in December 2024 to $0.30 in November 2025, “88 times cheaper” in 11 months.
Guo says consumption at inference clouds such as Baseten, Model, or Fireworks is growing by 1,000x, even as efficiency makes revenue grow more slowly. The humorous benchmark still carries a technical point: a human brain uses about 12-20 watts, leaving substantial room for model-compute efficiency.
5. AI’s terminal value depends on how much GDP technology absorbs
Guo identifies a reflexive competitive risk for incumbents: they possess acquisition currency until falling valuations deprive them of it. If labs and leading applications rapidly reach billion-dollar run rates and their valuations follow, challengers gain the financial capacity to compete while legacy vendors may lose theirs.
Gil’s historical scale markers show the regime change. In 2005, Google was worth about $100 billion and Exxon roughly $400 billion; Apple became the first $1 trillion company in 2018. Today, the top eight technology companies total about $23 trillion and represent well over 50% of S&P value.
Gil says the leading tech companies’ share of US GDP rose from around 4% in 2005 to about 12% today. Depending on growth assumptions, he sees technology reaching 15%-20% or as much as 30% of GDP by 2035 as services and jobs become AI-augmented software spend—supporting more, and potentially much larger, trillion-dollar companies.
Guo argues that expanding technological surface area could let the tail dominate. Gil concedes that there may be more $100 billion businesses as the surface area grows, but preserves the power-law point: market capitalization and customer value remain concentrated, so broader opportunity does not imply the tail captures most economics. “It’s all the head and torso.”
6. AI-era leaders need exit discipline and defensible control points
Guo observes a large handful of companies reaching $100 million-plus run rates faster than SaaS predecessors, yet even a billion dollars of revenue may not settle the question she frames as: is it “you,” Ant, or OpenAI over time? Category leadership once felt protected at scale; now another capability jump can reset the leaderboard.
Gil’s internet precedent is unforgiving: roughly 450 companies went public in each of 1999 and 2000; he says perhaps “one to 2,000” companies went public during the internet age, but only about one or two dozen remain relevant. Lotus reached hundreds of millions in revenue rapidly, then Excel took the market and Lotus collapsed into IBM’s arms—an exit, but no longer a standalone business.
Most companies, Gil argues, have about a 12-month period when they are worth more than they will ever be, often followed by a crash-out even after substantial traction. Boards should pre-schedule exit discussions once or twice yearly so a changed competitive structure or an offer higher than anything the company could achieve over the next five years can be assessed without panic; only a “very small handful” should never sell.
The defensive answer is breadth: build a bundle used for five or 10 different aspects of the same vertical or application, then reinforce it with ecosystems, networks, platforms, or hardware. SaaS-era “do one thing well” advice is dangerous when displacement cycles shrink from a decade to one or two years: “If every two years is 10 years,” founders must react accordingly.