Building | prev: Founding team @ Long Lake
San Francisco
Joined October 2025
- Tweets624
- Following320
- Followers2.2K
- Likes1.8K
Pinned Tweet
Increasingly, I believe companies may need to be rebuilt from the ground up, where you have a single timeline of all observability + product metrics + file changes laid out in a retrievable system, like Datadog + Posthog + Google Drive + Slack (really unified filesystem of Claude Code chats + Codex chats). This might be the new data foundation for any and all companies to maximize AI. Needs to be rebuilt because keeping track of diffs on existing system basically impossible to produce longitudinal information on decisions and rollbacks, something coding agent storage companies are actively trying to figure out, but this should extend to businesses as a whole.
Highly skeptical existing businesses will adopt this though because it means overhauling everything about their instrumentation and business data, but I think businesses built on this foundation probably can execute 100x better and faster
See, the problem with these posts are that they are not actually singularity pilled. If you are, then you must believe that we would get to tens, hundreds of breakthroughs in fundamental physics of classical physics to quantum that what gets discovered and “produced” likely doesn’t even belong as carbon based life form, let alone it being space / temporally bound the way humans are. We’d harness the energy of the universe trillions times more than what we have. None of this would be understandable by any human intelligence. There may not be because such lifeform wouldn’t even need to be confined to our timeline. And once it does, then it must actually already be omnipresent.
And you basically end up with deriving christianity from first principles.
If you are truly Singularity pilled, you'd realise we are probably gonna skip multiple branches of the tech tree.
Like everyone's picturing mechanical humanoids, but if ASI exists you're gonna skip right to the most optimal form, which for many use cases is probably biological.
So yeah you're gonna have lots of mechanical robots for sure (different use cases), but the Singularity is going to be far weirder and far faster than almost anyone is capable of comprehending.
Jev may nuke 99% of actually defensible fine tuning LLM use case of solving for low latency high volume analysis and classification particularly useful in fintech (e.g. credit card usage data, real time fraud detection)
My bet with white collar work always was to bet on the bitter lesson, which is:
Process mining and optimization in a software consulting traditional way doesn’t really make sense if we get vllms that can click through tens of pages at superhuman speed per computer at $10/hr. You don’t need a Fivetran type of connector for writes (you still need it for search and historical analysis that’s really superhuman). As long as it’s observable, interruptable, more accurate and faster than humans, you can bypass significant amount of FDE plumbing.
Most FDE companies and AI consulting companies haven’t really internalized this
Slack API is painfully slow and rate-limited. It clearly wasn’t built for agents.
Last night I asked Codex/Astra to search a huge Slack channel. My auth broke, so instead of using the API, it just used computer use.
I watched it scroll through Slack quickly, faster than API.
AI won’t wait for APIs, maybe.
Labs will believe that they will command premium prices for the non public models that they can share to enterprises. This will be true for select industries where they can get alpha from it . My take always has been that it's pareto principle useful for unlocking the next quadrillion economy, but the existing 60T will want to save costs and prefer open source models as intelligence saturates decision making and all that gets really left is better context following. Model "judgment" is largely a bad thing for enterprises unless true continual learning is solved, which will require fundamental rethinking of architecture.
Severely severely underpriced future. Always bet on ingenuity and irrational grit of a few changing the world.
He does need bodyguards though given how much the economy relies on fake continual learning ("look, the models are released faster - that is continual learning")
Rich Sutton - Continual Learning will be solved in two years, I am willing to bet my reputation on it.
Doina Precup - The learning loop currently includes of talented research engineers doing clever data mixture engineering.
#RLC2026
Watch our AI tutor teach by drawing on a whiteboard in real time.
Two AI agents work in parallel: one explains the concept while the other visualizes it, creating a learning experience you can interact with as it unfolds.
This is the future of learning anything.
Opus 5 enables this crazy new live AI tutor concept I'm prototyping for @AlphaSchool.
It powers chat + direction under the hood, while a custom trained tiny world model(ish) acts as a real-time renderer and interactivity layer.
Still rough, but it's already kind of insane.
John Suh retweeted
A billion people produce the most valuable dataset in the world every day – and delete it every night.
We’re recording it.
Introducing @AttentionInc
Watch me learn logarithms in real time.
This is the future of education. No useless multiple-choice drills. Just pen, voice, and real active problem-solving, without having to wait for a $60/hr tutor scheduling for that one hour you don't have swimming practice.
Demo:
(P.S. @tldraw is an incredible piece of software)
I actually think YC doing these big events dismantles higher education at scale. It is largely a good thing for the world even if some immediate LP money may go down the drain. It's basically 10x better philanthropy than what any non-profit may do
I have great respect for YC, but need to say what many are surely thinking:
Startup School in a stadium feels antithetical to the YC brand and to what makes startups work.
The YC brand felt like it was designed to filter out status chasers.
The bare Mountain View office. The utilitarian design of Hacker News. Everything about YC said: if you’re a curious hacker, we’re your people. If you’re looking for other kinds of prestige or excitement, we’re not for you.
Great early stage startups are designed the same way: they’re for the people that get a thrill from the work itself, not surrounding fanfare.
If it were just the arena I wouldn’t comment, but you can feel the gravity of a status vortex slowly tearing down what made YC special.
You can feel the herd rushing into an overflowing barn. A stadium sized barn.
You see how YC is now more prestigious than most college degrees. How prestige seekers name drop YC partners. How many YC founders now quit in the early innings—because they didn’t really love the work.
Maybe it’s inevitable. It happens to startups too. Eventually the underestimated, curious, hard working teams become the winning teams. And then the status chasers flock to them.
If YC is not strong enough to fend it off, maybe no one is.
John Suh retweeted
We’re launching BenchmarkList: one place to track AI benchmarks, models, and capabilities
We aggregated 2,400+ benchmarks scattered across papers, websites, and posts.
Now you can:
-Track new benchmarks, models
-Compare models side by side
-Map AI progress on human jobs
Basically never believe things on social media. Can't name the co but growth people from top lab said Twitter accounts with more than 10k get paid 6+ figures to shill / defend launches. They get offers from basically all companies. Apparently an absolute arms race to cover the feed