Enterprise AI Adoption Crisis - 4 Big Issues - A long form essay (Substack) If you don't agree and feel like enterprise adoption is exploding here is some data and what needs to be fixed. More data and links in the article. Happy to see some counter views and data though!

Cheap and lazy deposits have been the quiet moat of banking, and agents can make deposits price-sensitive overnight. What if the moat goes ?

Agent payments are a permission problem not a rails prob... Denmark & Canada both did their first AI-agent card payment this week, on Mastercard Agent Pay. The Canadian one ran inside a budget cap the cardholder set. That cap is the product.

For agentic payment startups -- people who have handled chargebacks for a living will be in high demand

What ?! Anthropic is paying $1B to Accenture to stop AI from surpassing humanity 🙂‍↔️. You have to respect the honesty, most AI companies sell you software to replace the consultants, Anthropic is hiring the consultants to save you from the software 🤷

Hi internet! We went ahead and upgraded all @stripe hosted checkout pages for 7.8M businesses—0.45% of the world’s GDP—with WebMCP so agents can more efficiently buy. Our evals show this reduces checkout latency by 39% and token usage by 42%. Learn more: stripe.dev/blog/how-stripe-i…

So Google says Gemini hacked three companies in a security test, researchers used Claude to hack OpenAI, and someone found malware run by an AI hive mind with no humans involved 🫢. The models are now stress-testing each other, and you have to wonder who is holding the pen on the other side.

Unpopular opinion: “we’ll figure out GTM later” is the most expensive line in early-stage. Product without a wedge into a real buyer is just a demo with burn rate. What’s the GTM mistake you see most in seed / pre-seed?

If Anthropic and OpenAI only wanted a moat, they would demand rules for everyone else and keep the doors shut. Instead they volunteered badges, desks, and employee-level access for outside evaluators. That is expensive, leaky, and hard to fake. The real tell is not the essay. It is whether those evaluators ever publish something the labs did not want published. Until then, “sinister cartel” maybe just a vibe.

OpenAI dropped the price on GPT 5.6 Luna. and the usage went up 10x and it was verified by OR. Luna hit #1 on OpenRouter more than 2x the token usage of Anthropic's Opus 5 and Sonnet 5 combined 🤯 moral of the story. ppl want to switch models more than ever before (with open source, free models, etc.). We built an AI harness for finance with task level model routing. and further token savings via graph building at skills level (context)

Banking and finance AI - Work where 95% correct is effectively 0% correct

you can’t discuss the ocean with a frog it only knows the pond

It was Sep 2026 when we saw... early clashes between personal AI agents and platforms designed for human-scale use, raising questions about bots racing for finite resources like tables and tickets. Amex's quick action against the 'tech bros' using Instinct, amid warnings of chaotic race conditions in bookings.

Craziest fintech co - 13 different product lines that do over $100M a year. They started with stocks. Then they added crypto. Now sports betting. As Shaan said "it’s an app for pure male degeneracy. All they’re missing is an OnlyFans feature"

Anthropic IPO investors are pushing for disclosures that don't exist in any tech filing: Revenue per token. Revenue per gigawatt of compute. Neither has appeared in a public IPO before. But they're the real unit economics of an AI company. How much you earn for every inference, every watt of compute consumed. If Anthropic sets the precedent in October: Revenue per token = the new ARR Revenue per gigawatt = the new gross margin The accounting standards for the AI era get written this fall.

What if Astra or Mythos 😬