Founder @SourcetableApp, the AI spreadsheet. Big fan of the Internet.
San Francisco
Joined January 2009
- Tweets8.7K
- Following293
- Followers1.6K
- Likes1.8K
The real governor on model progress seems to be cyber security. You can't release a model if it breaks the Internet.
No Internet, no revenue. The frontier paces itself.
How this was taught to me:
- investors live on email and are inundated by it. Everything is a filter to a call or meeting.
- they will spend 10 seconds on the email (usually, on their phone). If they like the email, they ask for intro immediately or will skim the deck.
- they will spend 1 minute flicking through the deck. If they like the deck, they will either ask for intro immediately, or spend a few more mins re-reading the deck and either ask for intro or pass.
Obviously, there are variations to this. Some emails convey more info, some founders won’t share decks pre-meeting for fair reasons, some founders counter-filter with a more-effort-up-front approach, and some founders are already known and can run whichever untraditional playbook they like. Generally though, you have 10 seconds then 1 minute, so clear communication matters.
Eoin McMillan retweeted
I’m Samuel, 19y/o Aussie building in drones/defence/aerospace, backed by some of the best investors in the country.
I’m spending 3 weeks in the US, embedding myself into critical industries like aerospace, manufacturing, materials, and defence.
If you’re building the future of your industry, dm me. I want to hear your vision for the future, and what’s holding us back. I will in turn help you out wherever I can + shout coffee.
I’ll be at AUSA, El Segundo + more. I want to meet the builders tackling the hard and important problems. Dm me!
Counterpoint: LEGO is 100% owned by the Kristiansen family and worth $30BN.
Markets reward value creation.
One of our use cases for LiteLLM
Just published this post about how we’re going to need default hard budget caps on pretty much everything simonwillison.net/2026/Oct/3…
What happens when warehousing and inventory management turns into local just-in-time manufacturing?
Yup, vibe manufacturing is coming in hot. In an Industrial Revolution, what happens when the revolution is a collection of all of our earned intelligence over time? What happens when the cost of that becomes a utility? What IS the next bottleneck?
Software is just about to be over.
Entrepreneurs - think accordingly. In order to rush in this new age, what do we need to build? I am looking for you. I know what's on that path. We will meet on it.
Love this. Give users what they want, and what they want is shared context across apps.
(The counterpoint is that this consolidates data network effects and will diminish personal sovereign AI.)
“Make something agents want!”
Every backend service should have a /feedback endpoint.
Agents are quickly becoming the heaviest users of most APIs. When one hits a missing feature or a bug, it should be able to say so right there, in a structured way.
If the request makes sense, another agent drafts the PR and a human approves it. Software that improves itself based on what its users actually tried to do speeds up recursive self improvement.
The ultra-premium version of this is to select a specific cow from a catalogue, clone it, wait for that special occasion and harvest...
To make the customer support agents seem more realistic, it's important to keep people waiting on hold for a while before patching them through...
The problem with this academic research is that everyone who has figured out how to hack LLM signal stays quiet. It's the dark forest of grey hats out there...
From word-of-mouth and personal experience, agentic recommendation is an entirely tractable problem.
Read the BERT paper and Google Antitrust docs to understand evolving cat & mouse efforts required to win prior algorithmic inbound via SEO.
Similarly, LLMs are exploitable algorithms with exploitable information supply chains. Personalization is just one of many variables that make this game fun.
🚨Our new research examines agentic shopping: can you consistently predict (or, using marketing, influence) what an agent chooses?
Nope.
We found that even small differences (viewing order of pages, memories) changed AI preferences in unpredictable ways. papers.ssrn.com/sol3/papers.…
This was excellent. So nice to hear articulate thinkers talk through every level of every decision at any level of depth with appropriate concision.
Neil Movva (@neilmovva) started his career at Nvidia, working on GPUs and kernels, and has an unusually deep understanding of inference, from software to chips to power.
We spend a lot of time on each of those layers, how they connect, and where the important tradeoffs are.
What makes this conversation special is how detailed it is (like a 401-level class), yet Neil makes it remarkably clear and easy to follow.
Today he runs Sail Research, a company building infrastructure for agents to make tokens as cheap as possible.
We discuss:
- Latency versus throughput
- Why there are no bad chips, only bad pricing
- The end of kernel engineering
- Buying chips and power no one else wants
- New chip architectures
- Nvidia lore + his contrarian view of the company
- Open source and the frontier labs
I learned a ton. Enjoy!
TIMESTAMPS
0:00 Intro
0:38 Building a “Token Factory”
4:21 The Future of Background Agents
13:09 Nvidia and the GPU Stack
23:27 Chips, Memory, and Transformers
36:14 The Future of AI Training Data
44:32 Chip Scarcity and Compute Arbitrage
52:44 Reinventing the AI Data Center
59:01 Power and the “Scavenger Strategy”
1:10:10 Open vs. Closed AI