Everyone is quoting one line from Sam Altman's newest interview: "we are now in the singularity."
It's the wrong line to fixate on.
The interview — @ti_morse's first sit-down with Sam, about 11 minutes — is dense with founder wisdom. But the thread Sam keeps circling back to is far more useful if you actually build things:
The biggest temptation right now is pointing today's AI agents at the easy wins. The real opportunity is building for the models that don't exist yet.
Here's the interview, distilled for builders.
1. Picking today's fruit is the gentle trap
A 10-week-old startup today, Sam says, can ship what used to take a year. He met a two-week-old company that had rebuilt an entire office-productivity suite for a world where AI is a first-class consumer of documents, spreadsheets, and slides.
By rights, startups should be easier than ever. He loves moments like this — "startups have the biggest edge when the ground is shifting the most."
And yet:
Most people are building "an AI agent for vertical X."
That works. It'll be competitive. And it's "probably not the most successful, defining startup of that era." What surprises him is how few people take the crazy swing — building for what isn't economical this month but will be in two or four years.
That's the trap: today's agents are sitting right there, and picking a few easy wins is obvious to everyone. But the moment you pick, you lock yourself into today.
2. Trust the exponential — there's still "free money" in it
Asked what he'd most want founders to understand, Sam's answer was blunt:
Internalize that scaling laws will continue.
And it's hard — hard enough that, in his words, there's still "free money" left in betting on the exponential (in young high-growth founders, and in model progress). The market hasn't adapted.
Translated for builders: it is OK to start building things now that only work once models get smarter and cheaper. Don't design for today's capability. Design for the model shipping in six months — so when it lands, you're already there and the fruit-pickers are just starting to rewrite.
The proof that this is counterintuitive? Sam's own biggest mistake: he under-invested in compute. "Got psyched out by the financial markets." The person who preaches exponentials the loudest still flinched.
3. "The singularity is now" — but it's really a fork
The viral quote, in context, is subtler than the soundbite. Sam frames it as two things at once:
- It's one continuous, crazy exponential — no single moment is the tipping point.
- And we're somehow in another decisive period where the curve "can go one way or another."
Which two ways? His biggest current worry:
Are we heading to a world of AI authoritarianism, or liberty?
The bad path: because of real safety and economic concerns, we decide one single model should be "the machine God," and hand control to it or the company behind it. His stance: every time humanity has traded liberty for safety, it's been a long-term net loss. So the mission is to make intelligence abundant, cheap, and decentralized — in everyone's hands.
For builders, the subtext: the next decade's upside sits on the side of putting more capability into more hands — not helping one giant hoard it.
4. Four hard lessons for builders
Beyond the headline, the interview is full of operating wisdom. Four that hit hardest:
Chaos is learnable, not teachable. How do you get good at operating in constant chaos? Reps. Young founders' real weakness isn't intelligence — it's not having made emotional peace with "the company might die" moments. Survive nine and the tenth stops feeling fatal. Further: learn to enjoy the pain. As Naval put it — if life had a fast-forward button, your life would just be over.
Kill what's working. The failing project is easy to kill. The hard call is killing something that works well to make the best thing better. They killed robotics when GPT-3 hit; killed Sora and the browser when coding agents took off.
Plan forward, not backward. Sam doesn't reverse-engineer from a rigid 20-year vision. He holds a small number of strong convictions as directions, then plans forward from today. Too many beliefs about the future → the "space company that becomes an AI company" drift. Few convictions, extreme flexibility on the rest.
Always ask for what you want. Codex is his example. They were far behind Claude Code; building a competing coding app looked like a kamikaze mission, the kind of thing that "never works." But they thought it mattered, asked the team to do it — and it became the tool most of the best coders he knows now use. If they hadn't asked, it wouldn't have happened. People wildly overestimate the risk of action.
5. Real vs. fake trends — and the third wave
Sam's test for "is this real or a bubble":
- Fake: lots of hype, but people buy it and don't really love it, don't reorganize their lives around it, and it collapses (his example: VR).
- Real: persistent, near-daily use that becomes part of your life (his example: ChatGPT).
The only metric that matters is deep, durable usage. Ignore volume, watch retention.
By that lens: two waves so far — chatbots, then coding agents ("going totally nuts"). And the third is coming soon:
Persistent agents — chiefs of staff, coworkers, colleagues, whatever we end up calling them.
That's the most concrete answer to "build for the model that doesn't exist yet." The next ChatGPT moment likely lives in the persistent-agent / digital-coworker form factor.
The takeaway: stand in front of the exponential
Collapse the whole interview into one line and it's the thing Sam keeps repeating:
Don't let today's model capability be your product's ceiling.
Fruit-pickers build for today. The real opportunity is to trust the exponential and build — right now — for the stronger, cheaper model six months out, so you're standing there when it arrives.
Which implies a concrete requirement: you need to be able to switch to any newly-topping model the day it ships, instead of hard-wiring yourself to whichever one is best today. Keep one endpoint, one key, and swap a model name when a new one lands. (It's why I run everything through an OpenAI-compatible gateway whose pricing I can curl and verify — flatkey.ai — but any gateway that lets you switch on a dime works.) Because per Sam, the best model changes hands every few months.
Whether the singularity is "now" is a question for philosophers. For those of us actually building: stand in front of the exponential — don't pick fruit in place.
Based on a transcript of @ti_morse's interview with Sam Altman (native X video, 2026-07-25, ~11 min). All views are Sam's own; this is a structured secondary read.
Top comments (0)