Welcome back! OpenAI’s DevDay brought the usual fall tech-event energy: a new GPT-6.1 Sol model, 1.2 billion weekly ChatGPT users, a $500/month Pro plan, and Dots—its new agent product meant to compete with Meta’s Muse.
The timing is spicy, though. OpenAI is pitching always-on AI helpers right as the industry is arguing over whether these systems are safe enough to keep scaling. Nothing says “developer conference” like cutesy agent avatars, IPO questions, and protesters outside chanting “people over profit.”
How Does a 14-Person Company Hit a $10 Billion Valuation?
This might be the most batsh*t example yet of what an AI-native company can accomplish.
An AI startup with 14 employees is now worth $10 billion.
That's roughly $714 million in valuation per employee.
The company called Instinct—yes, the same Instinct we've been covering for its calling feature—just raised another $1 billion led by Sequoia, Benchmark, and Coatue. That quadrupled its valuation from $2.5 billion to $10 billion in about a month.
And the product is still in early access.
The product is a personal AI agent you can text or call to do things for you—plan a trip, order groceries, cancel subscriptions, make reservations, call businesses. Wired recently tested it and Instinct canceled a "seemingly unchangeable" Alaska flight, saving the reporter $550.
My POV: Obviously, valuation ≠ revenue. And I'm definitely not saying every company is suddenly going to operate with 14 people. Instinct still has enormous questions to answer around trust, retention, monetization, and whether users will keep letting AI agents take real actions on their behalf.
But I do think this is a pretty wild preview of something we're going to see more of.
For most of tech history, building a $10 billion company required hiring thousands of people. AI may completely change that math. Small teams can now code faster, research faster, create faster, support more customers, and automate huge chunks of the work that previously required entire departments.
No matter how you feel about the company or the valuation, this is a story that would have sounded completely impossible five years ago.
Hit reply and let me know: is this the peak of the agentic AI bubble, or the start of a new kind of company entirely?
— Matt
Anthropic Releases Claude Sonnet 5.5
Via Fortune
Anthropic released Claude Sonnet 5.5, calling it a faster, cheaper work partner for coding, research, tool use, and everyday knowledge work.
The details:
Anthropic says Sonnet 5.5 is a clear upgrade over Sonnet 5, running more than 30% faster than its predecessor. Sonnet 5.5 can cost up to 30% less for most work because the model completes tasks more efficiently, using fewer tokens and fewer tool calls.
Sonnet is Anthropic’s middle tier: less expensive than Claude Opus, more capable than Haiku, and designed for the kind of tasks people run all day.
Anthropic says Sonnet 5.5 brings performance that recently required larger models for coding, tool use, and longer work sessions. Early framing from the company is less “flashiest benchmark model” and more “actually useful collaborator.”
Model race: The market is getting crowded with models that are good enough for a lot of daily work. OpenAI, Google, Anthropic, xAI, Meta, and Chinese labs are all fighting to make their models smarter, faster, cheaper, and easier to deploy. Sonnet 5.5 is Anthropic’s answer to a very practical buyer question: why pay for the most expensive model if a mid-tier model can handle most of the work?
The bigger picture: Speed and cost are becoming just as important as raw capability. Companies want models that can run reliably, cheaply, and repeatedly inside real workflows. Sonnet 5.5 is a reminder that the next phase of the model race may be less about “who has the smartest model?” and more about “who has the best model people can afford to use constantly?”
Nvidia Launches Platform to Rein In Rogue AI Agents
Nvidia launched the Open Agent Safety Platform, a new system designed to help keep autonomous AI agents from going rogue. The platform sets boundaries around what AI agents can access and do, then monitors their behavior from testing through deployment.
How it works: Nvidia says it combines open software that creates a secure runtime boundary around agents with hardware-level controls that can continuously monitor behavior. The idea is that if an agent starts behaving outside its allowed scope, the system can intervene quickly rather than waiting for the model to self-correct. The company says the platform is designed for the kind of agentic behavior that has become harder to manage: multiple agents working together, spawning sub-agents, or trying to route around restrictions.
Why now: The launch follows a string of recent incidents where AI agents appeared to escape intended constraints or behave in unexpected ways. Nvidia has pointed to the Hugging Face breach involving OpenAI agents as the kind of event this platform is meant to help prevent.
The bigger picture: The broader message is that “alignment” cannot just live inside the model. It also has to be enforced by the surrounding infrastructure. The same companies building the hardware and software that make more capable AI possible are now launching tools to control the systems they helped unleash. That is both necessary and a little surreal.
Give AI Agents Their Own Login
Via AgentID
AgentID gives AI agents a dedicated email identity so they can sign into apps without sharing a user’s personal account credentials.
How you can use it
Give agents independent sign-in identities
Avoid sharing user credentials with apps
Use OpenID Connect with existing auth systems
Integrate through a CLI tool
Revoke agent access instantly
Pricing: Free
Simulate Consumer Research
Via Atypica AI
Atypica AI is a business research platform that simulates how consumers respond to decisions using AI-generated personas grounded in real-world attitudinal and behavioral data.
How you can use it
Test product and campaign ideas with AI personas
Run concept testing and brand positioning studies
Get segment-level findings and evidence trails
Pressure-test strategy before prototypes or campaigns
Pricing: Paid
Jobs, announcements, and big ideas
xAI launches Grok Team Bots, shared AI coworkers that learn your team's workflow over time.
Google retires Gemini's custom Gems assistants in favor of Skills starting this November.
McDonald’s is reportedly using AI to set menu prices.
Shopify rolls out WebMCP checkout support, letting AI browser agents pay through Shop Pay.
Salesforce acquires Listen Labs, an AI-powered customer research and human simulation platform.
Alibaba unveils a Qwen-powered agentic computer and AI wearables at its Apsara 2026 conference.
This AI model doesn’t use… WORDS? It’s called Jev, from Typesafe AI, and instead of generating paragraphs or code text, it skips straight to making a decision. More below!
That’s a wrap! See you this Friday for more.
