Forget AGI. Intelligence is Cheap.
Forget AGI. Intelligence is Cheap.
Base intelligence is becoming a commodity. The real value is moving up the stack
Feb 22, 2026
6 min read
Welcome to Slow Ventures’ Snailmail, where we slow down and share what’s been on our partners’ and founders’ minds this week.
TL;DR:
- Forget AGI. Intelligence is Cheap.
- SaaS is Dead. Long Live Software
- Others: Lightning Labs released an open source tool enabling AI agents to make autonomous payments, Lightspark partnered with Cross River Bank for fiat payments via Bitcoin Lightning network, Breakfast with Slow and AIC (NYC)
Forget AGI. Intelligence is Cheap.
Sam Lessin
Context:
I’ve been thinking a lot lately about where all this AI “intelligence” is really heading, especially as the cost per token keeps dropping and the number of players in the space explodes. Picture me in the backcountry, slaying pow, turning over this idea in my head: if intelligence becomes super cheap and abundant, does it actually retain any value? Or are we just racing to the bottom, commoditizing the very thing we’re all supposedly building toward?
Market Signal:
We’re already seeing this play out. The supply of “intelligence”—whether human or machine—is skyrocketing. Unless demand for intelligence somehow outpaces this supply (which I’m skeptical about), the market value of intelligence as a raw ingredient drops, fast. I don’t buy the old narrative that whoever gets to AGI first “owns” intelligence and wins the game. That story is, frankly, dead. Instead, we’ll see a world with tons of intelligence providers, and the price to access “base intelligence” will keep falling.
In this context, it’s hard to see how anyone (Anthropic, OpenAI, whoever) can hold anyone “ransom” over terms of use. If intelligence is a commodity, you just buy it from someone else. The marginal cost per token is the key metric here, and as it approaches zero, the business model for selling generic intelligence gets shaky.
Takeaways:
Base intelligence is becoming a commodity. The days of selling “raw tokens” as a differentiated, high-margin product are numbered.
. Specialized use cases, proprietary IP, and orchestration layers (think: routing, context, vertical integration) are where the business models will shake out.
The “power” analogy is instructive, but incomplete. Sure, electricity is regulated, but intelligence on the internet isn’t governed by the same borders or regulatory capture. Unless governments force everyone to use a single provider (unlikely), competition and abundance will drive prices and margins down.
Scarcity is where value accrues. What’s abundant is cheap; what’s scarce is valuable. The question for founders and investors is: what can you build on top of commoditized intelligence that isn’t abundant?
SaaS is Dead. Long Live Software.
Yoni Rechtman
Context:
We’re living through the SaaS comedown. The public markets have repriced pure software businesses, and the assumptions that underpinned SaaS as the “forever” business model (zero marginal cost, non-ephemeral value, and high switching costs) are all getting eroded by AI.
Market Signal:
Shrinking TAMs and Depressed Multiples: SaaS isn’t going away, but the market for it as we knew it isn’t growing. Multiples are down and likely staying there.
Margin Compression: What used to be 25% income margins might be 10-11% going forward. The economic engine of SaaS is fundamentally dented.
AI Eating the Margins: AI increases marginal costs (inference isn’t free), accelerates value depreciation (new tools obsolete old ones faster), and destroys switching costs (software is easier to replicate and agents don’t care about brand loyalty).
Takeaways:
Software is Still a Business, Just Not the Same Business: Pure application software is starting to look more like traditional services: easy to start, hard to scale, low need for outside capital, and little defensibility.
Switching Costs Are King: The only way to preserve margins and pricing power is to build real switching costs—through network effects, brand, monopoly, or critical workflows.
Hybrid Models Win: The future is software bundled with services, hardware, or unique data. “Sin-eater” businesses, those that absorb operational risk or complexity, will persist (infra, security, fintech).
Don’t Bet Against the Models: Align with the foundation models. Don’t root for their limits instead, find ways to leverage their strengths.
The Next Big Opportunity: Connecting foundation models to high-value, complex real-world use cases is the biggest wealth creation opportunity in recent memory. It won’t look like the last SaaS/cloud wave.
Asks:
Founders: Stop pitching pure SaaS with “defensible” high margins unless you have a real moat. Show me your switching costs, your network effects, your hybrid model.
Operators: Rethink your margins and product strategy. How are you decommodifying your code? Where are you absorbing risk or creating lock-in?
Investors: Update your frameworks. The old SaaS multiples don’t apply. Look for new forms of defensibility, not just recurring revenue.
More Musing From The Team
OpenClaw vs Meta vs OpenAI: The Personal Agent Wars Heat Up - YouTube
Tap to unmute
OpenClaw vs Meta vs OpenAI: The Personal Agent Wars Heat Up More or Less Podcast
More or Less Podcast59K subscribers
Disagree with anything? Hit reply, we read every note.