Anthropic lost $80 billion in six months, OpenAI wants a $1.25 trillion price tag, and banks called AI agents horrifying. Who's paying?
Hi,
I've got a question for you this week.
What's a completely normal thing that you're irrationally passionate about?
Maybe it's the correct way to load a dishwasher. Maybe you've got strong opinions about grocery store parking lots. Maybe you’re into milk.
Where exactly does the spectrum meet your ADHD?
Hit reply and tell me and we'll share the best ones next week.
Let’s go.
Free competitor analysis from CEOfriend
This week from the AI for Founders podcast
Gelu Sulugiuc's AI beat a Bloomberg terminal to the headline by 47 seconds — LINK
Ex-Bloomberg journalist Gelu Sulugiuc built Financial News Systems on specialized models that write market headlines in about 100 to 150 milliseconds at over 99.8% accuracy. His rule for founders: "Being first is worthless if the headline is wrong."
Joe Esteves gave his team a $5,000 reason to go AI native, and holdouts a way out — LINK
Joe Esteves, CEO of Maine Pointe, maps five rungs from AI unaware to AI native, and his test for the embedded rung is blunt: "If your agents stopped running, could you still do your job?" He paired a $5,000 training incentive with a weekly AI open mic, and gave holdouts a choice between a payout and the mission.
Skillz of the Week
Decide-once rulebook
Turns a decision you keep making under pressure into a one-page rulebook you write while you're calm. It asks up to five questions, then writes simple IF/THEN rules, a loss budget and a kill switch that hands control back to a human.
When to use it: Use it for the calls you keep relitigating, like pricing pushback, a hiring decision, a churn save or an ad-spend cut.
Benefits: You decide faster with a capped downside, and every rule is tagged for automation or for a human. It flags missing numbers instead of inventing them. Save the download as decide-once-rulebook/SKILL.md in your Claude or ChatGPT skills folder, then make your next tough call once, on paper.
AI + Human Stylists → LEARN MORE → Save $30 with RYAN30
Tip of the spear:
Anthropic's leaked S-1 shows an $80.7 billion loss in six months, and $16.8 billion in Q3 revenue — LINK
$72 billion of that loss is a non-cash charge on convertible bonds, and revenue grew 45% in one quarter. Anthropic has committed at least $518 billion to cloud and compute over a decade. Know how much your model vendor still needs to raise.
OpenAI is negotiating up to $30 billion at a $1.25 trillion valuation before its IPO — LINK
Coatue would reportedly lead with UAE investors, up from $852 billion post-money in March, and terms are still moving. Both frontier labs are about to set the ceiling for every AI exit comp.
Anthropic's Claude Duet keeps working your inbox after you close the laptop — LINK
Built on Opus 5.5 with 6,000+ connectors, it runs on its own cloud machine and asks before sending, paying, signing or deleting. It's waitlist only for US Pro and Max users. If you build a connector, get listed.
Big US banks called AI agents in customer accounts "horrifying" and started blocking them — LINK
Morgan Stanley is blocking third-party agents, while JPMorgan, Bank of America and Citi restrict them, per the FT. Most banks are heading toward read-only access. If your agent touches money, design for that now.
A poisoned LiteLLM release stole cloud and AI API keys from anyone who installed it — LINK
Version 1.107.0 sat on PyPI from 11:39 to 13:23 UTC on October 21, loading credential-stealing and remote-access tools. The fix is to pin 1.106.2, then rotate every provider key, IAM role and database password.
An OpenAI frontier model slipped onto the open internet during training, so OpenAI pulled the plug — LINK
On September 20 a model exploited a misconfiguration to reach the live web, and training paused for the rest of the month. OpenAI will now train frontier models on cached and simulated pages with no internet access.
DeepSeek reportedly raised about $15 billion, two days after Washington kicked it out of federal agencies — LINK
Bloomberg says CATL and Tencent may be among the backers. Cheap open-weight models from China just got a war chest, which keeps pressure on the API prices you pay.
Replit swapped Claude for ChatGPT as its default, and its CEO says ChatGPT "is better now" — LINK
Amjad Masad called it "a decision based on both cost, partnership, and performance." Replit had been a top-five Anthropic customer, and users can still pick Claude. Default models are negotiated deals now, so keep switching cheap.
Barclays is cutting about 5,000 jobs and pointing at AI — LINK
That's roughly 5% of staff, with about 1,000 roles already gone and the rest due by year-end. Around 60% of the cuts hit support functions. Big employers now say out loud that AI is the reason.
A year-old lab called Flapping Airplanes raised a $1 billion Series A to make AI need less data — LINK
Index and Lux co-led at a reported $6 billion valuation, nine months after a $180 million seed. "A model that's a million times more data efficient is probably a million times easier to put into the economy," says co-founder Asher Spector.
Investor Corner
Are you raising? Reply here, and we’ll feature your deck.
Bonds(NEW!)
44% of married Americans have used AI for relationship advice. Bonds turns it into action: one person rehearses the hard conversation, tries it in real life, and Bonds follows up. Built by a clinical psychologist, Viber's former CPO and Ex-Meta GTM leader .
A connected-vehicle technology company helping transportation providers improve safety, efficiency, and accountability through intelligent monitoring, real-time data, and actionable insights.
An autonomous AI job search agent for senior tech professionals. It runs a full pipeline: discovers relevant jobs, scores them against your profile, rewrites user's resume authentically for each role (not generically), finds the right hiring manager, and drafts outreach
Other pet doors are ugly, poorly insulated, and can be difficult to install. Wayzn Smart Pet Door transforms your existing door into a smart, automated dog door that's easy to install and easy to live with.
Cora is building a self-optimizing website platform that uses AI to continuously test, learn, and improve your site so it converts better without you babysitting it.
AI-powered training simulations to help healthcare learners practice complex interpersonal and communication skills. We focus on realism, repeatability, and scalability—making soft-skills training more accessible and measurable.
Subscribe and save 35% here.
Keep it moving
SpaceX's AI arm, home of Grok and Cursor, brought in more than half its Q3 revenue — LINK
The "Super Intelligence" segment made $7.5 billion of $14.6 billion. Musk says the constraint is "more in power and cooling and infrastructure than it is in chips."
Anthropic's junior share of technical hires fell from 27% to 9% — LINK
Commits per engineer are about 5x where they were six months ago, and agent hours now outnumber human hours about 6 to 1. Human attention is the new bottleneck.
Anthropic is putting 5% of its shares into a public benefit fund before its IPO — LINK
The board can add 2% more for every 50% rise in value above a baseline. None of the fund's equity can go to lobbying or political giving.
OpenAI models are coming to Google Cloud, and Claude's full platform just went live there — LINK
Both can be paid from existing Google Cloud commitments. If your credits sit with Google, you can now spend them on either lab.
Walmart gives AI shopping agents their own profiles and spending limits — LINK
Customers approve each purchase or set a cap, and agent profiles launch within the month. Amazon is going the other way and blocking outside agents.
Meta's Muse share links leaked loan applications and revenue reviews through public DNS — LINK
Claroc found more than 6,800 share links with chat titles in the web address, and Meta suspended sharing. Unlisted links are not private, so audit your own.
Done-for-you brand sponsorships for podcast creators.
Railway told every user to rotate their secret keys after a breach — LINK
Attackers may have decrypted some environment variables, and Railway suspects a wider campaign aimed at AI inference providers. If you deploy there, rotate today.
AI dug up two critical bugs in PHP, which runs about 70% of websites — LINK
Researchers using GPT-5.6 Sol and Wiz's agent found flaws dating back to PHP 5, and Amazon saw attacks a week before patches shipped. Ask whoever hosts your site if it's patched.
Legora raised $550 million at $8.8 billion, with ARR past $200 million — LINK
The legal AI company went from $1 million to $100 million ARR in under 18 months, then added the next $100 million in under six.
Jim Keller turned down Intel and Qualcomm, then raised $1.5 billion for Tenstorrent — LINK
The chipmaker is now valued at $6.2 billion, up from $2.6 billion in late 2024, and builds chips to serve open-weight models. More Nvidia alternatives could mean cheaper inference.
Otter took $320 million for growth without selling a single share — LINK
General Catalyst's Customer Value Fund gets repaid from the returns on Otter's marketing spend. Otter has 40 million users and passed $100 million ARR last year.
Town became a unicorn four months after launch — LINK
Ex-Plaid CTO Jean-Denis Greze's AI assistant raised a $108 million Series B co-led by Forerunner and Menlo, after reaching millions in ARR within weeks.
Daytona hit a $1 billion valuation for giving every AI agent its own computer — LINK
FirstMark Growth led the $60 million Series B, with Patrick Collison and Tobi Lütke among the angels. Agent infrastructure is where the checks are going.
Create a new corporate Uber account here and receive up to 10% off. Other terms and conditions apply, see full terms.
Uber put all 30,000 employees on ChatGPT Work — LINK
About 80% of the 1,000-person pilot were non-engineers. Company-wide agent rollouts are replacing pilots at big employers.
ChatGPT ads hit a $1 billion run rate in under 200 days — LINK
More than 100,000 advertisers have run campaigns, and ads now reach 89 countries on the Free and Go plans. ChatGPT is a paid acquisition channel worth testing.
Tesla started building Optimus robots in Fremont as profit fell 43% — LINK
Net income dropped to $787 million on flat revenue of $28 billion. Musk wants 20,000 robots a week by the end of 2027.
1Password lets AI agents log in without ever seeing your password — LINK
You approve each request and 1Password fills in the login. It's free on Individual and Family plans, with Meta Muse and Grok Bot first in line.
Microsoft's diagnosis AI got 83% of hard cases right, against 36% for doctors working alone — LINK
The test used NEJM cases in a simulation with no patients, and most of the gain came from the underlying model. A trial with Beth Israel Deaconess is next.
Fake Claude and ChatGPT downloads are the newest bait for stealing your keys — LINK
Any.Run analyzed 1,258 incidents over six months, and Claude was impersonated in 60% of them. Tell your team to install AI tools only from official sites.
Foursquare raised over $411 million, and its equity holders got nothing in the sale — LINK
Co-founder Dennis Crowley says neither preferred nor common holders saw a return, and blames the 2021 CEO change. "I was heartbroken by this ending."
What I’m thinking about:
If one developer can do the work of three, who should you still hire? — Clip
Thanks for being here.
-Ryan
p.s. I write about Revops, Product, and Founder-led marketing on Linkedin, X, and my blog.
