
Ship first, optimize later - Founding Engineer | AI Security | @UMassAmherst @UCBerkeley 🐶 https://teapaw.org/t.co/invF2TOpTO
San Jose, CA
Joined July 2014
- Tweets1.3K
- Following973
- Followers541
- Likes5.3K
Eighteen months later, the OIDC-A questions mostly held up.
About five got answers. There are now more than 70 competing drafts. Cross App Access is a clean first hop, but it still doesn't identify the agent. I also wrote down what I got wrong.
10
15
1,530
five things on the IPO:
1. they are listing as a public benefit corporation. the opening line is about securing benefits and mitigating risks. the live question is whether that duty still binds once Class A trades on Nasdaq as ANT.
2. Q3 revenue is $17.0B, more than 12x year over year. net loss is $23.3B, and $21.9B of that is a non-cash fair-value charge on convertibles, derivatives, and SAFEs after the valuation rose. operating margin is -3%; adjusted is 15%.
3. more than 11,900 customers spend at least $100K trailing twelve months, also more than 12x year over year, as of Sept 30, 2026.
4. they claim a >4x compute-efficiency gain since Q1 2025, defined as revenue divided by compute and infrastructure expense. i'd want that definition held constant after the listing.
5. this is still a proposed offering. the registration statement is not effective. they expect to set the price on Nov 18 (EST) in a $70–$78 range, and to start trading on Nov 19, for 825 million Class A shares, 801 million of them primary.
Our initial public offering: ipo.anthropic.com
Legal disclaimer: Anthropic has filed a registration statement (including a prospectus) with the SEC for the offering to which this communication relates. Before you invest, you should read the prospectus in that registration statement (available at: sec.gov/Archives/edgar/data/…) and other documents Anthropic has filed with the SEC for more complete information about Anthropic and this offering. You may get these documents for free by visiting EDGAR on the SEC Web site at sec.gov. Alternatively, copies of the preliminary prospectus may be obtained from: Goldman Sachs & Co. LLC, Attention: Prospectus Department, 200 West Street, New York, NY 10282, by telephone at (866) 471-2526, or by email at [email protected]; or Morgan Stanley & Co. LLC, Attention: Prospectus Department, 180 Varick Street, 2nd Floor, New York, NY 10014.
The securities may not be sold, nor may offers to buy be accepted, prior to the time the registration statement becomes effective. This post is not an offer to sell, or a solicitation of an offer to buy, the securities in any jurisdiction where the offer or sale is not permitted.
4
77
Scion inverts Git's client model.
One authoritative server. Thin, always-connected clients. Changesets created server-side.
Moving main = one compare-and-swap on one Spanner row.
No merge queue. 200x smaller footprint than their Git mirrors. 7 people, 5 months.
1
28
Perplexity, the day before, described nearly the same system for documents instead of code.
Agent gets a sandbox checkout. Submits. Server accepts only if built on current head, else "stale, replay."
<2% stale. Median submit 1.7s. p99 8s.
Neither post mentions the other.
1
1
13
Anthropic merged ~2,500 changes a day in May.
By September it was 30,000+ a day.
Git did not break. The assumption under Git broke: one author, human speed, sometimes offline.
Between Sep 16 and Oct 15 every major lab rebuilt the same thing. A thread.
1
3
91
What they built, in four weeks:
Sep 16 Claude Docs + Cowork merge
Sep 17 Claude Projects: one cloud branch per thread
Sep 29 OpenAI Dots, Space, Pages
Oct 8 Gemini coworker agents with their own Drive
Oct 12 Perplexity's Git-backed workspaces
Oct 13 Anthropic's Scion
Oct 15 Perplexity open-sources SPACE
1
2
60
I used coding agents to research a farming-game policy.
Live peak: 3,056.
Final submission: 1,647.
Better local tests didn’t guarantee better live results.
The research loop, failures, and interactive replays: subramanya.ai/2026/09/30/kag…
2
201
Their whole world. One little app. 🐾
Meet Paww - the all-in-one app for dogs and their people. Everyday care, important details, and little wins, together.
Sign up for beta → heypaww.com
check it out
I rewrote NumPy's core in Rust.
Validated against NumPy's own unmodified test suite: 46,330/48,511 passing (95.5%) linalg and fft at 100% and bit-for-bit identical across 73,879 differential checks. 0 divergences.
Faster than NumPy on most large-array ops (sin/exp/log 4–6x, int matmul 8x).
github.com/subramanya1997/rn…
2
2
208
I rewrote NumPy's core in Rust.
Validated against NumPy's own unmodified test suite: 46,330/48,511 passing (95.5%) linalg and fft at 100% and bit-for-bit identical across 73,879 differential checks. 0 divergences.
Faster than NumPy on most large-array ops (sin/exp/log 4–6x, int matmul 8x).
github.com/subramanya1997/rn…










