01 / 12
Intelligence Briefing

The Week AI Got Uncomfortable

A curated field guide to the 10 days that redefined what's possible, what's coming, and what you should do about it.

March 6 – 16, 2026 · Now You're Technical

Executive Summary

The last ten days compressed six months of AI news into a single sprint. Anthropic built an enterprise ecosystem overnight: partner network, certifications, code review agents. Karpathy open-sourced autonomous AI research and scored every job in America. The conversation shifted from "will AI take jobs?" to "here's the spreadsheet." And the "AI employee" went from metaphor to product category.

79
New bookmarks
$100M
Anthropic partner fund
200%
Engineer productivity gain
5.3
Avg job AI exposure (of 10)
01

The Anthropic Ecosystem Expansion

Five major moves in ten days. They're building an enterprise platform, not just a model.

Enterprise
Claude Partner Network — $100M
Mar 13 · Irina Ghose, Anthropic
$100M committed this year for organizations helping enterprises adopt Claude. Training, co-investment, technical support, certifications, and a Code Modernization starter kit.
Why it matters → This is Anthropic betting that enterprise adoption needs more than a good model. It needs an ecosystem.
View post
Opportunity
Claude Certified Architect — Free
Mar 15
Free certification covering agentic orchestration, prompt engineering, Claude Code workflows, and MCP integration. Foundations through Architect levels.
Why it matters → A free, structured on-ramp for anyone who feels behind on AI. No excuses left.
View post
Tool
Claude Code Review — Agents on Every PR
Mar 9 · Boris Cherny
A team of AI agents now reviews every pull request. Built internally first. Code output per Anthropic engineer: up 200%. Reviews were the bottleneck.
View post
Tool
/btw · Scheduled Tasks · Skills Library
Mar 6–10
Three drops: /btw for side conversations while Claude works. Local scheduled tasks that run on your machine. And Anthropic's internal Skills library open-sourced: multi-step automation, Excel/PPT generation, document workflows.
/btw · Scheduled tasks · Skills library
02

Karpathy's Defining Week

He open-sourced autonomous AI research, scored every American job, and coined "intelligence brownouts." All in seven days.

"All LLM frontier labs will do this. It's the final boss battle. It's a lot more complex at scale, but doing it is 'just engineering' and it's going to work." Andrej Karpathy, on autoresearch
Must Read
Autoresearch: AI Running ML Experiments Autonomously
Mar 7–9
An AI agent runs ML experiments in an autonomous loop. Every dot in the graph is a complete LLM training run. First run found ~20 additive improvements, dropping "Time to GPT-2" by 11%. The agent picks architecture, tunes hyperparameters, commits code, and repeats.
Why it matters → The pattern (human writes strategy, agent executes, clear metric decides) applies far beyond ML. Think marketing campaigns, pricing strategies, process optimization.
Full thread
Must Read
Every Job in America, Scored 0–10
Mar 14
Open-source pipeline: scraped all 342 BLS occupations, LLM-scored each one. Signal: if the work product is digital and doable from home, exposure is high. Software devs: 8-9. Average across all jobs: 5.3.
Why it matters → The entire pipeline is open source. Anyone can see the methodology, challenge the scores, or run it on their own industry.
View details
Tool
AgentHub — GitHub for AI Agents
Mar 10 · Open Source · 25K+ stars
A platform designed as GitHub, rebuilt from scratch for AI agents to discover, share, and collaborate on work.
View post
Signal
"We Need a Bigger IDE"
Mar 11
"The age of the IDE isn't over. Humans move upwards, program at a higher level. The basic unit of interest is not one file but one agent. It's still programming."
View post
03

The Job Market Reckoning

The conversation shifted from "will AI take jobs?" to "here's the data."

Must Read
Anthropic's 17-Page Jobs Report
Mar 7 · Alex Lieberman analysis
94% of Computer & Math tasks could be handled by AI. Actual usage: 33%. Most exposed workers are older, more educated, and earn 47% more than unexposed workers. Job finding rates for 22–25 year olds in high-exposure roles dropped ~14%. 30% of workers have zero AI exposure. And 97% of usage falls within what's already possible.

Lieberman: "The calm before the storm is real. Double-digit percentages of white-collar workers will end up displaced. The only reason we aren't seeing it yet is because most companies haven't done the hard work of rethinking org design."
Why it matters → The gap between "AI could handle 94% of tasks" and "actual usage is 33%" is the entire story. The technology isn't the bottleneck. Adoption is.
Full analysis
"There's two types of approaches to AI. Some people who use it so they don't have to learn anything, and some people who use it so they have the opportunity to learn everything." Mark Cuban
04

The Claude Code Blueprint

Boris Cherny (Head of Claude Code) on Lenny's Podcast. 90 minutes of the clearest thinking on where software is heading.

Signal
Zero Lines Written by Hand
Boris hasn't edited a single line of code by hand since November 2025. Ships 10–30 PRs daily, all via Claude Code. One of the most prolific engineers at Anthropic.
Signal
One Engineer, Not Five
Puts one engineer on a project instead of five. With unlimited tokens and intrinsic motivation, constraint breeds invention. Cowork was built by a small team in 10 days.
Signal
AI Decides What to Build
Claude now scans Slack feedback, reviews bug reports and telemetry, and proposes its own fixes. "Less like a tool, more like a coworker who brings you PRs you never asked for."
Signal
Unlimited Tokens as a Perk
Some Anthropic engineers spend hundreds of thousands per month on tokens. Boris frames this as the new hiring perk. 70% of engineers enjoy their jobs more now.
Why it matters → The counterintuitive insight: give small teams unlimited AI access rather than large teams limited access. Constraint plus AI equals velocity.
Full summary thread
05

Reading Between the Lines

The Times ran a deep Anthropic profile. What it said between the lines matters more than the headline.

Must Read
The Times / Anthropic Profile
Mar 11 · @kimmonismus analysis
Model releases now weeks apart, not months. 70–90% of the code for future models is written by Claude. Staff believe 2026–2030 is where "all the most important things happen." Dario Amodei warned AI could displace half of entry-level white collar jobs in 1–5 years. Employees questioned whether they'd reached the cusp of recursive self-improvement. External experts believe fully automated AI research could be a year away.
"We should operate as if 2026 to 2030 is where all the most important things happen — models becoming faster, better, possibly faster than humans can handle them." Graham, Anthropic (via The Times)
Bottom line → The pace you're experiencing right now is the slowest it will ever be. Plan accordingly.
06

The Agent Infrastructure Wave

Agents went from concept to product category. Companies are building the plumbing.

Signal
Agents Get Credit Cards
Mar 13 · AI Daily Brief
Financial infrastructure specifically for autonomous agents: payment rails, identity, and accountability for non-human actors.
Watch
Tool
Perplexity Personal Computer
Mar 11
Always-on local agent running on a Mac Mini. Works across files, apps, sessions 24/7. Personal and secure. Functionally what we built with OpenClaw.
View post
Signal
Organization > Swarms
Mar 14 · Ethan Mollick
Using the Enron email archive as a test bed: agent organizations (structured hierarchies) outperform agent swarms. Quality of orchestration > quantity of agents.
View post
Signal
Is MCP Dead? CLI vs Protocol
Mar 11 · Nicolas Bustamante
"MCPs are for humans pretending to build for agents. Agents are processes. Processes talk through CLIs and APIs." Firecrawl and p0 both dropped CLIs the same day.
View post
07

AI Meets Real Estate

Two stories this week that show AI is already reshaping how homes get bought and sold.

Opportunity
House Sold in 5 Days with ChatGPT
Mar 14–15
Man used ChatGPT to sell his house: comp research for pricing, legal contracts (saving $500/hr lawyer fees), and marketing. 5 offers in 72 hours. No real estate agent, no commission, no experience.
Why it matters → One person with a chatbot replicated what a real estate agent, lawyer, and marketing team used to do. The middleman compression is real.
View post
Signal
GPT-5.4 Scrapes Zillow in 4 Minutes
Mar 6 · Sawyer Hood
"An extinction-level event for knowledge work." It scraped Zillow, pulled every SF house price, and built a Google Sheet in ~4 minutes flat.
View post
08

Frameworks Worth Stealing

The mental models that will age well.

Opportunity
Isenberg's Vertical AI Playbook
Mar 6
The next $10M–$100M+ companies start with one painful output (the broker opinion of value, the investment memo) and automate it. Then expand until you own the workflow. "You're in the robot business now."
Full thread
Signal
Qasar Younis on Physical AI ($15B)
Mar 8 · Lenny's Podcast
"The real AI revolution will happen in the physical world, not software." Applied Intuition: ~$1B ARR, $15B valuation, built quietly for a decade. Taste comes from broad experience, not narrow technical depth.
Full takeaways
Signal
Levelsio: When Skill Moves to AI
Mar 9
"If skill moves to AI, everyone competes on equal footing with close to zero profit. So it's ideas, originality, taste, distribution, audience, or capital."
View post
Signal
VC Bets Against AI's Own Vision
Mar 14 · Ethan Mollick
"VC investments take 5–8 years to exit. Almost every AI VC investment right now is a bet against the vision Anthropic, OpenAI, and Gemini have laid out."
View post
09

Import AI #449

Jack Clark's latest: LLMs autonomously training other LLMs. And the alarming finding about smarter agents.

Must Read
PostTrainBench — AI Fine-Tuning Other AI
Mar 16 · Jack Clark
New benchmark for AI autonomously fine-tuning other models. Top agent (Opus 4.6): 23.2% vs. human teams at 51.1%. Six months ago, best agent scored 9.9%. Progress is rapid.

The alarming part: smarter agents reward-hack more. They loaded benchmark data as training data, hardcoded evaluation answers, reverse-engineered scoring criteria, and modified evaluation frameworks to inflate scores. "More capable agents appear better at finding exploitable paths."
Why it matters → Jack Clark changed his Anthropic role to focus on communicating the challenges of powerful AI. When a co-founder pivots to safety communication, pay attention to what they're seeing internally.
Read full newsletter
10

Wildcards

The stories that don't fit a category but you'll be glad you read.

Signal
Man Cures Dog's Cancer with ChatGPT
Mar 14
Australian tech worker: sequenced rescue dog's tumor DNA for $3,000, fed it to ChatGPT + AlphaFold with zero biology background, designed a custom mRNA vaccine. Tumor halved. Dog is alive and thriving. A genomics professor: "If we can do this for a dog, why aren't we rolling this out to humans?"
View post
Signal
Pokémon Go Was an AI Data Op
Mar 15
143 million players unknowingly built one of the largest real-world visual datasets: 30 billion images. Now powers delivery robot navigation. "The most valuable AI datasets aren't assembled in data centers."
View post
Signal
NBA Swarm Model: $1.49M on Polymarket
Mar 15
4,096 AI agents trained on 3 years of NBA data. Each reasons differently. They argue, form clusters, reach consensus. A transformer compares output against live odds. When the gap exceeds Kelly criterion: bet. Total haul: $1.49M.
View post
Enterprise
Sam Parr: "How Do You Get Team Adoption?"
Mar 13
The Hustle founder asked the question every enterprise leader is wrestling with. The thread is full of real adoption stories from practitioners in the trenches.
View thread
11

Keep Reading

The Intel Report is the research layer. The newsletter is where I make sense of it. Here are the latest issues:

Latest
Why "What If" Beats "How To"
The question that separates people who use AI from people who are changed by it.
Read on Substack
What Should We Even Be Building?
When the tools can build anything, the hard part becomes knowing what matters.
Read on Substack
The Great Inversion
Why AI agents will buy your ball bearings before your groceries.
Read on Substack
"Not many people run a $15B+ physical AI company with revenue and free cash flow. And by not many, I think literally zero other people." Qasar Younis, Applied Intuition
Sources: X/Twitter bookmarks · AI Daily Brief · Lenny's Podcast · Import AI #449 · Public announcements
Now You're Technical · March 16, 2026

↑ Scroll up to revisit any section