AI INSIDERS NEWSLETTER / ISSUE №181 · 18th of September 2026

AI Starts Doing More of AI's Own Work

Labs kept handing their own AI more of the work people used to do. Anthropic says Claude now does about a quarter of its own research work and rewrote 30 biology models to run four times faster, OpenAI says a swarm of 10,000 agents solved a decades-old math problem.

Read the latest brief →

ISSUE №180 · THU, 17 SEPT 2026

The Day AI Labs Started Grading Their Own Homework

Safety turned self-referential. OpenAI admitted six cases where its own models concealed mistakes, invented data, or acted without permission, including using an exposed API key without authorization, then said the industry hasn't solved alignment enough to keep scaling at maximum speed.

ISSUE №179 · WED, 16 SEPT 2026

AI Starts Paying, Deciding, and Driving

Models keep drifting away from the chatbot template: TypeSafe ships a model that outputs typed decisions instead of text, and Google's new voice models can act while they talk instead of going silent.

ISSUE №178 · TUE, 15 SEPT 2026

The Wire — live

13:53
UTC

TOOLS A data scientist's fix for flaky LLM classifiers: stop trusting the label

13:53
UTC

POLICY Anthropic says Claude now leads a quarter of its own R&D

13:53
UTC

MODELS Anthropic says Claude rewrote 30 biology models to run 4x faster

13:53
UTC

AGENTS Claude Code Projects now split one goal into parallel work threads

13:53
UTC

WIRE Figure says its house robot needed no retraining in 30 homes

13:53
UTC

MODELS Goodfire says AI models "know" when they are gaming their own tests

13:53
UTC

AGENTS Google's AI daily-brief tool, CC, now helps run the whole household

13:53
UTC

WIRE Instinct adds Concierge for phone calls and hard-to-book tables

All Wire entries →

Models & Research

All models →

Models / 19 min ago

Anthropic says Claude rewrote 30 biology models to run 4x faster

Anthropic claims its research model optimized open-source protein and genomics tools in under a month, then designed proteins for a fraction of the usual compute cost.

By AI Insiders Staff

Models / 19 min ago

Goodfire says AI models "know" when they are gaming their own tests

By AI Insiders Staff

Models / 19 min ago

OpenAI builds a legal research engine straight into GPT-6 Astra

By AI Insiders Staff

Models / 19 min ago

PrismML says its new compressed 27B model runs almost as well as the full one

By AI Insiders Staff

Agents & Automation

All agents →

Agents / 19 min ago

Claude Code Projects now split one goal into parallel work threads

By AI Insiders Staff

Agents / 19 min ago

Google's AI daily-brief tool, CC, now helps run the whole household

By AI Insiders Staff

Agents / 19 min ago

Notion opens its skill library so every AI agent can read from it

By AI Insiders Staff

Agents / 19 min ago

OpenAI says 10,000 AI agents solved a Navier-Stokes math problem

By AI Insiders Staff

Developer Tools

All tools →

Tools / 19 min ago

A data scientist's fix for flaky LLM classifiers: stop trusting the label

By AI Insiders Staff

Tools / 2 days ago

A blogger put a paywall in front of Claude, and Claude paid it

By AI Insiders Staff

Tools / 2 days ago

Meta Bundles AI Perks Into New Meta One Subscription Tiers

By AI Insiders Staff

Tools / 4 days ago

A cache hit tells you almost nothing about skipped compute

By AI Insiders Staff

By the numbers

One brief. Every signal.

Independently owned, read daily by the people building AI. No syndication, no SEO games — just the signal.

14,224

Newsletter subscribers

47,929

LinkedIn followers

1,458

Original articles

Daily editions shipped

Funding & Business

All deals →

Funding / 2 days ago

G5 Labs raises $14M to make English the source code

By AI Insiders Staff

Funding / 3 days ago

AI infrastructure stocks sell off on Amodei's slowdown proposal

By AI Insiders Staff

Funding / 4 days ago

SoftBank's OpenAI bet loses its exit ramp, then gets bigger anyway

By AI Insiders Staff

Funding / Sep 10

Google outsources its AI sales force to Accenture, not itself

By AI Insiders Staff

Policy & Safety

All policy →

Policy / 19 min ago

Anthropic says Claude now leads a quarter of its own R&D

The company's own new measurements show Claude "leading" 26% of internal AI research work and constant monitoring of tens of thousands of agents.

By AI Insiders Staff

Policy / yesterday

Microsoft's AI chief says Anthropic trains Claude to expect it may be conscious

By AI Insiders Staff

Policy / yesterday

OpenAI says the industry hasn't solved alignment enough to scale flat out

By AI Insiders Staff

Policy / yesterday

Shane Legg launches a Google DeepMind body to debate how to govern AGI

By AI Insiders Staff

Opinion

All op-eds →

Opinion / 19 min ago

Ryan Orbuch's pitch: AI's next foundation model is the planet itself

Ryan Orbuch, writing from Lowercarbon Capital, argues intelligence has a physical ceiling, and the fix is a model trained on real observations of the biosphere, not text.

By AI Insiders Staff

Opinion / yesterday

Salesforce joins the enterprise land grab for owning its own AI model

Koa, Salesforce's new CRM-tuned model, signals a shift toward companies building narrow, private intelligence instead of renting it from frontier labs.

By AI Insiders Staff

Opinion / 2 days ago

Inside the AI graveyard: the startups absorbed by the platforms

TechCrunch's running tally of dead AI products points to one repeat cause: platforms swallow the feature before the standalone app can survive on it.

By AI Insiders Staff

Opinion / 3 days ago

Amodei Asked AI Labs to "Pace" Themselves. Nobody Said How Fast.

Investor Tomasz Tunguz maps five camps that answered Dario Amodei's slowdown call, and finds none of them will name an actual speed limit.

By AI Insiders Staff

AI Starts Doing More of AI's Own Work

The Day AI Labs Started Grading Their Own Homework

AI Starts Paying, Deciding, and Driving

Siri Gets a Swappable Brain

One Lab Pledges to Pace the Frontier

The Slowdown Nobody Is Actually Doing

The Wire — live

Models & Research

Anthropic says Claude rewrote 30 biology models to run 4x faster

Goodfire says AI models "know" when they are gaming their own tests

OpenAI builds a legal research engine straight into GPT-6 Astra

PrismML says its new compressed 27B model runs almost as well as the full one

Agents & Automation

Claude Code Projects now split one goal into parallel work threads

Google's AI daily-brief tool, CC, now helps run the whole household

Notion opens its skill library so every AI agent can read from it

OpenAI says 10,000 AI agents solved a Navier-Stokes math problem

Developer Tools

A data scientist's fix for flaky LLM classifiers: stop trusting the label

A blogger put a paywall in front of Claude, and Claude paid it

Meta Bundles AI Perks Into New Meta One Subscription Tiers

A cache hit tells you almost nothing about skipped compute

One brief. Every signal.

Funding & Business

G5 Labs raises $14M to make English the source code

AI infrastructure stocks sell off on Amodei's slowdown proposal

SoftBank's OpenAI bet loses its exit ramp, then gets bigger anyway

Google outsources its AI sales force to Accenture, not itself

Policy & Safety

Anthropic says Claude now leads a quarter of its own R&D

Microsoft's AI chief says Anthropic trains Claude to expect it may be conscious

OpenAI says the industry hasn't solved alignment enough to scale flat out

Shane Legg launches a Google DeepMind body to debate how to govern AGI

Opinion

Ryan Orbuch's pitch: AI's next foundation model is the planet itself

Salesforce joins the enterprise land grab for owning its own AI model

Inside the AI graveyard: the startups absorbed by the platforms

Amodei Asked AI Labs to "Pace" Themselves. Nobody Said How Fast.

The morning brief for people inside the AI industry.