37% context left. I ⭐️ my own Github projects.
Sunnyvale, CA
Joined April 2009
- Tweets14.3K
- Following269
- Followers12.6K
- Likes16.6K
Pinned Tweet
Spent the past week on SlateDB's DST harness. It was a bit of a slog. The more state I explored, the more false positives I encountered.
rng.md/posts/deterministic-s…
I’m excited to have a smaller community of more involved humans that use AI constructively to make SlateDB better.
Chris retweeted
Building a cloud demo using TiDB database + Filesystem for AI agents, I found SlateDB a great fit for control-plane metadata.
Read-heavy, write-light, with object storage as the source of truth.
One less storage engine to reinvent. 😂
Thanks, @criccomini
Chris retweeted
AI may well be able to write massive software projects “from scratch”, but that almost certainly isn’t the fastest / highest performance / highest security / most token efficient way to do things. Good building blocks matter.
In addition to RL’ing the agents on final task loss, we should try to get them to regularize the building blocks as well. What would the standard library look like that makes the agents most effective?
It is likely at the limit that it turns out to be some alien organization of code, but given the existing human-bias in training, I bet that in the near term it would be lovely code that old programmers would appreciate.
I suspect the same could be said for mathematics, but that is out of my lane.
Chris retweeted
Help us get to 1 billion workflows/sec next.
We just scaled DBOS to 1M+ workflows/sec
To help our largest users scale, we made DBOS run on sharded Postgres (supporting any sharding engine: PgDog, Citus, Neki, soon Multigres, anything custom…), allowing you to horizontally scale your workflows across many databases.
We found that DBOS scales linearly to 1M workflows/sec sharded across 50 databases. We considered going for more, but the AWS bill from the benchmark servers had become too high
Just published a new blog post about how we built it (and the tricky architectural issues we had to solve to make sharding scale):
👇
Chris retweeted
In the last day I have requested changes on more pull requests in @ApacheDataFusio than I think I have in the entire 6 years with the project. We need to slow things down to reduce the chaos and spend our time actually thinking and planning the code.
this is a stunning amount of wealth
Who actually gets rich if Anthropic lists at $2 trillion?
The S-1 filed on Monday lets us answer from the filing itself.
The reported target is $1.5T to $2T, listing the week of November 16. The filing implies about 16.9B fully diluted shares. So $2T works out to roughly $118 a share.
The cap table at $2T:
- $199B for the 7 controlling co-founders, about 10%
- $28B for Ben Mann, if his stake matches theirs
- $100B for the Anthropic Public Benefit Fund, 5%
- $331B in employee options and RSUs, about 17% , That pool covers about 5,300 employees, an average of $62M each.
- But when you joined matters enormously.
The S-1 shows how much equity was granted each year, and at what price. [
Headcount went from about 300 to 500 in late 2023, to 940 at end 2024, to 5,300 now.
Match the two, and here is roughly how many people made what at $2T:
- $110M to $400M each: ~300 to 500 people hired by 2023
- $200M to $300M each: ~440 to 640 hired in 2024
- $70M to $95M each: ~1,360 hired in 2025
- $10M to $20M each: ~3,000 hired in 2026
So about 940 people hired by the end of 2024 average nine figures. Everyone hired since averages eight figures.
The 2024 class is the surprise. That year Anthropic granted 1.28B options at $1.30 a share. At $118, that one year of grants is worth about $133B.
Then there is the tender. In April, holders could sell at $25.91 a share. Investors lined up $6B, but only $4.7B sold.
Weeks later, Series H priced at $58.90 a share. At $2T, the people who held would be up 4.6x. And Dario and Daniela Amodei were among the sellers.
The 7 founders have pledged 80% of their equity to charity. That is about $159B at $2T. But the S-1 says the pledges are not legally binding.
Learning of the day: make skills self-learning. add evaluator-optimizer/agentic context engineering and tell the skill it should improve itself whenever it learns something skill-related that's generalizable.
Chris retweeted
So basically everyone is converging on - sandboxes, CI, apps, git, versioned file systems and managed inference?
Introducing Vercel CI
• Intelligently skips tasks with Turborepo
• 2-6× faster CI in our internal repositories
• Optimized for agents via Vercel MCP and CLI
After nearly 8 million tasks, we're excited for the future of agentic CI. In beta on all plans.
vercel.com/blog/introducing-…
Chris retweeted
Indeed job postings:
1/ Software development postings are rocketing upward: Y/Y growth of almost 25%. At that pace we're about a year away from pre-COVID levels (we're recovering from a deep depression in this category).
I was pretty bullish on local compute for LLMs. But the more I play with swarm stuff, the more I see I have an insatiable demand for compute. Renting from cloud seems preferable (even if it's open weight).
Chris retweeted
autoscaling is hard, but a good metric to leverage for S3-native systems is estimating the size of the working set
systems like neon/slate can use number of unique pages/blocks hit over the last 5/15/60m tracked in a hyperloglog.
then you scale when cache size < working set
