The AI Code Slop: Risk or Opportunity?

AI labs are growing revenue faster than any technology companies in history, token costs are collapsing at breathtaking speed, and yet the market narrative is fixated on which SaaS companies will die. Elad Gil and Sarah Guo think most people are looking at the wrong side of the equation.

The Episode

Sarah Guo and Elad Gil, co-hosts of No Priors and both prolific AI investors, use this episode to step back from the month of “SaaS is dead” panic and examine the structural forces actually reshaping the software industry. The conversation moves from raw revenue data to historical analogies to practical founder advice, with Elad drawing heavily on original analysis from his investment team. The tone is bullish but measured: they see a revolution happening, but find the popular framing of it shallow and often wrong.

The Numbers Nobody Is Talking About

Elad’s team pulled data from Capital IQ on how long it took companies to grow from $1 billion to $10 billion in revenue. ADP took over 20 years. Adobe took about 20 years. Salesforce and SAP took eight or nine years. Google, Meta, and AWS took three to five years. The AI labs did it in roughly one year.

The projections from $10 billion to $100 billion are equally striking. Microsoft took about 27 years. Google took over a decade. For the AI labs, public projections suggest three to five years.

Simultaneously, token pricing is collapsing. GPT-4 equivalent model costs dropped 150x in 21 months, from $37 per million tokens to $0.25. For o1-equivalent models, costs dropped 88x in 11 months, from $26 to $0.30 per million tokens.

“We’re having pricing collapse on the token side while we’re having revenue ramp insanely on the usage side.”

The combination is unprecedented: the fastest revenue growth in software history happening alongside the fastest cost decline. Elad thinks the market is ignoring the revenue and usage side of the equation, fixating instead on which incumbents will lose.

The SaaS Apocalypse: What’s Right and What’s Wrong

The meta trend is real: the bottleneck on producing expensive software is loosening, and that’s genuinely revolutionary. But Elad pushes back on the specific company-level predictions.

The “SaaS is dead” narrative confuses a few things. First, software markets have always been a fight about distribution and customer relationships, not just code production. Reducing the cost to express a point of view in software means more ideas get built, not that existing companies with distribution advantages automatically lose.

Second, much of the recent panic was driven by a month of hype rather than structural analysis. Sarah frames it well: people are “projecting five person startup behavior onto the Fortune 100.” Large enterprises are deeply committed to their vendor stacks, and change management at scale is genuinely hard.

Where the narrative gets it right: the velocity of technological change in AI means displacement cycles that normally took a decade are now happening in one to two years. That’s the real structural shift, not some binary “SaaS is dead” declaration.

Tech’s Growing Share of GDP

Elad’s team analyzed tech as a proportion of US GDP over time:

• 2005: tech was about 4% of GDP. Google was worth $100 billion. Exxon was the world’s most valuable company.
• 2018: Apple became the first trillion-dollar company. Tech represented about 30% of the S&P 500.
• Today: the top eight tech companies are worth about $23 trillion, making up over 50% of the S&P 500. Tech is about 12% of GDP.
• 2035 projections: anywhere from 15-20% to 30% of GDP, depending on assumptions.

AI accelerates this trend by converting services and certain jobs into software spend. The implication: tech company market caps will grow even larger, and the question becomes how many more trillion-dollar companies the ecosystem can support. Is it two? Three? A dozen? Fifty?

The Power Law Hasn’t Changed

Sarah raises an optimistic view that the “tail” of smaller companies will dominate because the surface area addressable by technology keeps expanding. Elad pushes back with the power law: head and torso aggregate almost all value.

He cites Google’s ad revenue as an example. The “long tail” thesis was popular, but in reality, most of Google’s revenue came from head and torso advertisers. YC’s value is probably concentrated in five companies accounting for 80% of it. Why would AI change this distribution?

Their resolution: there will be many more $100 billion businesses because the total surface area is growing. But the distribution shape remains power law, and the biggest companies will be even bigger than before. More companies at the top doesn’t mean the tail matters more; it means the head got bigger.

When to Sell: A Framework for Founders

Elad offers perhaps the most practical advice in the episode. Most companies have about a 12-month window when they’re at peak value, and then they crash. For a very small handful of companies, the answer is never sell. For almost everyone else, the timing question matters enormously.

His tactical recommendation, borrowed from Ben Horowitz’s experience running Opsware: pre-schedule a board meeting once or twice a year specifically to discuss exits. By making it routine, it becomes non-emotional. You’re not signaling panic; you’re maintaining hygiene.

“You just set up a non-emotional meeting once or twice a year. You’re like, ‘Nope, still not time to do it.’ Or you say, ‘Oh, you know what? Actually, the competitive dynamic has shifted dramatically.’”

Historical precedents make this urgent. During the internet wave, roughly 1,000 to 2,000 companies went public, and maybe a dozen to two dozen remain relevant. Mark Cuban sold Broadcast.com to Yahoo at its peak and collared the Yahoo stock to protect against decline. That financial engineering is what made him a billionaire.

The Lotus example is equally instructive: it built one of the first killer apps in spreadsheets, grew explosively, and then was crushed by Microsoft Excel. The business looked durable until it wasn’t.

The AI Code Slop Problem

The episode opens with what Sarah frames as the anxiety of the moment: AI can generate enormous amounts of code that nobody reads, nobody deeply understands, and nobody can assess for quality. It’s “the slop problem, vibe coding slop in my actual production codebase.”

But she sees this as an open opportunity rather than a terminal risk: “Nobody knows how to manage that issue of human attention to engineering. I think it’s like open season around this really really big problem.” The implication is that managing AI-generated code quality at scale is itself a massive unsolved problem ripe for a new company to tackle.

The Bundle Defense

When the ground is shifting this fast, what actually protects a company? Both Sarah and Elad converge on the same answer: multi-product bundles.

Building a bundle means cross-selling multiple products into the same organization so you become a default part of the workflow. You’re used for five or ten different functions rather than one singular thing that’s easy to clone. Sarah notes that bundles are “often seen as offensive, but I actually think they’re amazing for defense.”

This cuts against the conventional SaaS wisdom of “do one thing well.” Elad argues that was always bad advice; pre-SaaS companies were highly acquisitive and multi-product. The single-product focus was a SaaS-era artifact enabled by slow technological change. In an era where every two years equals a decade of change, singular products are sitting ducks.

A Few Observations

This episode is more valuable for its data-driven framing than for any single insight. Elad’s team has clearly done substantial original analysis, and the revenue growth and token pricing charts tell a story that cuts through the noise.

• The revenue growth comparison is the episode’s strongest contribution. Seeing ADP’s 20 years next to AI labs’ one year on the same axis makes the scale of change visceral rather than abstract.
• The “every two years is ten years” mental model is useful for any founder or investor trying to calibrate decision speed. If displacement cycles have compressed 5x, your strategic planning cycle should compress too.
• The exit timing advice is refreshingly honest for a venture investor to give publicly. Most VCs have incentives to tell founders to keep building; Elad is saying the opposite for most companies.
• The code slop framing is underdeveloped but points to a real opportunity: the gap between code generation speed and code comprehension speed is widening, and whoever solves that gap captures a large market.