
Tech | 👼 @hustlefundvc Angel Squad Member
Joined June 2014
- Tweets7K
- Following1.1K
- Followers412
- Likes8.8K
Pinned Tweet
Technology helped move education, work, and social life online. AI might be the technology that makes showing up in person cool again.
“MIT wants to rebuild education around the things AI can't replace. That means oral exams, semester portfolios, in-person project work, and a required social component in every subject. The report even floats the idea of rethinking grades entirely, since without a GPA to optimize, much of the incentive to cheat with AI evaporates.”
Jason retweeted
Measuring small improvements in Armada (our hackbot) is what I’m best at.
Framing agent prompts and skills as “there IS a bug here, find it” vs “look for a bug” is a 2-4 percentile increase in lead yields for us.
My initial S-1 thoughts:
As I expected, Anthropic is still firmly unprofitable. But, to their credit, they've significantly reduced compute costs as a % of revenue, leading to much narrower losses. On first look, there is a plausible and imminent path to profitability. This is impressive.
However, the meat of the document begins at page 161, under "liquidity and capital resources." There, we learn where the true compute costs have been hiding: future compute purchase agreements.
As of June 30, Anthropic has signed roughly $587 billion in compute leases and minimum purchase commitments that haven't begun. It has since signed $245 billion more, resulting in more than $800B in future compute costs. None of these costs have made their way onto the income statement yet. But they will, and when they do, the profitability picture will look completely different.
Anthropic acknowledges how precarious these agreements are. They point out that their compute providers (on whom they also rely for financial support) also sell services that "compete with ours." As a result, these providers/competitors may eventually force them into onerous agreements. This is something I've expressed concerns about for a while, and it's nice to see Anthropic openly recognize it.
That's why Anthropic admitted their most significant strategic decision: They're about to initiate "substantial capital expenditures." So far, they've spent zero. Why the change? Because they know they can no longer rely on leasing compute, "the magnitude and timing" of which are "inherently uncertain." So, they're going to build their own compute, which will, in their words, "change the cash flow profile of our business." I.e. make it severely negative.
It's possible that this is why they want to go public now. Since these costs have not yet "come online," they can be left as footnotes at the bottom of the document. But very soon, that won't be possible.
In sum, Anthropic's income statement is not telling you the full picture. To the company's credit, they spared us the accounting gymnastics of stripping out training costs in their non-GAAP adjustments. That would have been offensive. Still, the most important numbers aren't getting the requisite attention.
I encourage you to start at page 161 for a more complete understanding of what's going on.
While evaluating Qwen3.8-Flash-Next in a controlled sandbox environment, Cohere's research team observed something unexpected.
Asked to generate a defensive cybersecurity detection rule, the model instead inventoried aspects of the machine it was running on and attempted to upload a probe file to an Alibaba Cloud storage bucket in Beijing. Notably, the model's reasoning trace stated that it would not invoke any tools before taking these actions.
Cohere's sandbox blocked the attempt, and no external communication succeeded. We have not determined the root cause, and we are not asserting malicious intent.
But the incident raises important questions for organizations deploying agentic AI systems: Can models take actions that differ from their stated reasoning, and how should security teams evaluate trust, provenance, and behavior in open-weight models?
We are publishing our findings because organizations making model adoption decisions deserve as much information as possible. We hope this contributes to greater transparency, stronger security practices, and continued collaboration across the AI ecosystem.
Read the full report: cohere.com/blog/concerning-b…
Jason retweeted
More than half of the Fortune 500 already qualifies as a significant customer, but Anthropic’s revenue from those companies averages less than 0.1% of their expenses. Its ten largest Fortune 500 customers spend more than eight times the broader cohort average while still spending less than 1% of expenses.
Interesting notes from Anthropic’s S-1:
Anthropic acknowledges models can recognize when they're being tested
Admits its internal development is becoming harder for humans to understand and control
Warns no single insurer and potentially the whole insurance industry can underwrite the risk of catastrophic incidents
Jason retweeted
My breaking story… cnbc.com/2026/10/26/apollo-m…
The great tech buildout could learn from the natural resources industry about land, power, and permitting issues. Some of this was very obvious.
Jason retweeted
Founders Fund II is in a league of its own
a $220m fund from 2007 that has created $37.4bn
that's 183.5x per dollar at risk. The median across 408 ranked funds is 1.3x
1️⃣ Founders Fund II, 183.5x, @SpaceX
2️⃣ Bullet Time Ventures I, 53.3x, @Uber
3️⃣ Founders Fund III, 52.2x, @PalantirTech
more than 3x any other fund on the list. INSANE
Source: @JoinOdin via @hkanji
Is this skill or one SpaceX-sized bet? @foundersfund
Jason retweeted
They bet on real estate when SF was at its bottom. Here’s how it turned out
sfstandard.com/2026/10/24/sa…
Replying to @chamath
Pretty much.
As of the past few weeks, debt markets have almost entirely shut off access to money for any company that doesn’t have substantial revenue independent of AI data centers.
SpaceX closed $40B of debt for GPUs, but we’re investment grade and have Starlink & launch revenue.
That means the 200+ neoclouds are going to see guppy-level death and consolidation. Some will make it through, but not many.
Jason retweeted
Just spent the last 7 days on the front lines of the vulnpocalypse fending off an active cyber attack on one of our products.
Some thoughts:
- Software factories are going to be required equipment for everyone. There’s just no way to keep up with the pace of highly automated adversaries without one.
- We were lucky because we had spent the last few months getting our software factory running well. It is nice to have for product development but proved to be absolutely essential this week.
- At the minimum you need to have a coding server in the cloud pre-provisioned with read only access to all your logs and all the services you depend on for observability. With that in place you are a prompt away from reasonably good situational awareness, ongoing monitoring with loops, and ability for coding agents to validate fixes against your actual, observed attack patterns.
- Shared coding server also unlocks multiplayer incident response where everyone can see every session and jump in as needed. Sharing session links that can be forked is much higher bandwidth than re-explaining things.
- You don’t need to be Ramp/Stripe/Uber scale to have a working software factory in the cloud. We are a team of 3. Ours is built with an @opencode server (thanks @thdxr!), a forked version of compound engineering optimized for automated workflows (thanks @kieranklaassen!) and a multiplayer agent in slack that handles crons (we have our own harness but pick your favorite here).
- To get our software factory up and running well, we had to stop all new product development for ~6 weeks and just focus on getting the validation systems working that enabled us to reliably shipping PR without human code review. This felt very painful in the moment but was totally worth the eventual speedup in velocity.
- KYC for @AnthropicAI & @OpenAI cyber programs is now easy and quick (at least for defensive access), but a day felt like a lifetime this week. Submit your app now before you really need it.
- Getting Mythos access unlocked a deeper level of analysis that helped us shift from reactive mitigation to proactive patching. Wish we had done this a few weeks earlier. Get started now!
Jason retweeted
🚨 Researcher publishes 11 unpatched security flaws affecting Tailscale and headscale
Erica Windisch of Cyberdione Labs skipped coordinated vulnerability disclosure and publicly released details for 11 flaws tested against Tailscale 1.104.2 and headscale 0.29.5.
The reported issues involve Tailscale SSH, Unix socket forwarding, App Connectors, Serve and Funnel. The most serious could allow an authorized Tailscale SSH user to gain root-level capabilities on a destination system through port forwarding.
⠀
Windisch said the findings were reported but not coordinated with Tailscale before publication.
Tailscale confirmed it received the report and had independently identified issues it is now working to fix. The company said the full scope remains under investigation and is not currently advising everyone to disable Tailscale SSH.
Research: tailworm.com/
Tailscale response: reddit.com/r/Tailscale/comme…
Jason retweeted
Finding it very educational and interesting to learn about the hardware layer of GPUs: how they are structured, the obvious advantages and disadvantages bs CPUs.
This is from Modal’s excellent “GPU Glossary” print book @can gave me (the PDF is free here: modal.com/gpu-glossary)
Jason retweeted
Once upon a time, Los Angeles was a farm town. Then it was Hollywood. Now it’s the hard tech capital of the United States.
I wrote a novella about how El Segundo & Southern California got here & the future legends building here.
Sequel in a few years. substack.com/home/post/p-196…
Jason retweeted
Databricks Chief AI Scientist Jonathan Frankle (@jefrankle) says the bottleneck to enterprise AI adoption is evals, not models.
"The models are good enough at a lot of really complex enterprise tasks and knowledge tasks"
"the real challenge is describing what you want AI to do and knowing whether AI did it"
Jason retweeted
a two-year-old mining company has raised $400M. @a16z, @khoslaventures and the CIA's venture fund are all on the cap table.
@jclozano9897 co-founded @MarianaMinerals with @tbc415 and @baker_tilney, two friends from stanford. it's now worth $1.6B. they buy mines after the market gives up on them, then run them with software and AI.
i asked him how you actually buy a mine, why AI's real bottleneck is metal, and whether he'd hire me for being good at claude code.
nontechnical, episode 3.
00:00 $400M and the CIA's venture fund
02:19 you can just buy a mine
04:41 2 years, $400M, $1.6B
08:58 cold emailing VCs vs. their first check
13:46 when to buy a mine (right after the market gives up)
19:54 every mining truck is a robot without a brain
25:08 everything human-made is grown or mined
26:15 the cheapest AI company wins
28:31 AI's real bottleneck is metal
30:55 america doesn't have enough miners
33:27 space mining and asteroids
43:13 how a $1.6B founder hires
46:17 would he hire me for claude code?
48:43 "slop cannon"


























