← Back to Post AI Company
Radar — Week 23
What the best AI newsletters published this week
Curated with opinion, not summarized.
Updated June 8, 2026
The teaser. The most interesting thing I read this week wasn't an article. It was a footnote in Jack Clark's newsletter: nominal AI GDP in the United States is now roughly $250 billion, growing at 2,600% per year in quality-adjusted terms, and almost none of it shows up in conventional statistics. The Jaws analogy: the economic data says the water is calm, but the people inside AI are screaming at the swimmer to turn around.
🎙️ From Post AI Sessions
Marco, Episode 1 — June 11
"I still think Midjourney is an outlier. But I think it's the kind of outlier that tells you where the mean is heading."
The debate
Marco and Lena spent 24 minutes arguing whether Midjourney's 11-person, $200M structure is a blueprint or a fluke. They never fully agreed. That's the point. New episodes every other Wednesday, starting June 11.
Listen to the trailer →
📊 Post AI Index — Stat of the week
$3.48M
AI-native avg rev/emp
vs
$610K
Traditional SaaS avg
5.7× gap across 47 companies tracked
🔭 Import AI
Jack Clark — Issue 459, June 1
Inside Anthropic: "Many engineers no longer write code directly." Humans are becoming a verification layer atop a "vast virtual organization of AI agents." Also quantifies the invisible AI economy: $250B in nominal AI GDP, growing ~2,600% per year in quality-adjusted terms.
Our take
This is the Post AI thesis in one issue. The "invisible economy" problem Clark describes is exactly why we build the Post AI Index — to measure what conventional metrics miss.
Read on Import AI →
🏗️ Ben's Bites
Ben Tossell — June 4
Codex launches Plugins and Sites. Cursor reveals 40% of internal PRs now come from cloud agents. New models: Gemma 4 12B, Ideogram 4.0, Miso One 8B TTS. Microsoft Scout ships as AI agent built on OpenClaw.
Our take
"Build tools to build more" is a productivity multiplier thesis. Companies that build internal AI tooling compound faster than those that only consume third-party AI.
Read on Ben's Bites →
📊 Interconnects
Nathan Lambert — June 1-2
Lambert leaves Ai2 (stealth new venture). Argues open and closed models are on diverging exponential curves — closed wins on marginal intelligence, open wins on cost, customizability, and ecosystem compound.
Our take
The split-exponential thesis explains why both sides keep winning simultaneously. Harvey (fine-tuned Kimi 2.6) beats Opus 4.7 on legal benchmarks at 11× lower cost. The question isn't which side wins — it's how fast the gap moves.
Read on Interconnects →
🧪 Latent Space
swyx & Alessio — June 2-4
Andon Labs on building frontier evals. GitHub's Kyle Daigle on agent infrastructure strain. xAI's Ethan He on building Grok Imagine in 3 months.
Our take
Evals are becoming product. As frontier models converge on capability, the eval layer becomes the new battleground. GitHub's story echoes what every platform is scrambling to build: when code is abundant, orchestration becomes the scarce resource.
Read on Latent Space →
📰 The Batch
Andrew Ng — June 5
Qwen3.7-Max challenges Google for third place. Fine-tuning breaks copyright alignment. Conservation AI shows real-world impact.
Our take
The Qwen story isn't "China catching up" — it's "open weights catching up." The fine-tuning/alignment breakage is the sleeper: as domain-specific models proliferate, safety is increasingly a per-instance problem.
Read on The Batch →
🗞️ The Rundown AI
Rowan Cheung — Early June
RSI (recursive self-improvement) framed as the defining AI story. Anthropic confronts the RSI clock. Coverage: Ideogram, Microsoft Build 2026, Nvidia agent stack consolidation.
Our take
If RSI is a matter of years, not decades, org design decisions made today — human-in-the-loop vs. human-as-verification-layer — compound dramatically.
Read on The Rundown →
😺 The Neuron
Pete Huang & Noah Edelman — June 5
"Anthropic: AI Is Building AI now." Plus: TSMC supply warnings, ChatGPT memory, Meta's $200/month AI agent, Robinhood giving AI agents wallets.
Our take
The headline captures the moment better than any other this week. And giving AI agents wallets before agreeing what they should and shouldn't do is simultaneously absurd and inevitable.
Read on The Neuron →
One Question
If most code inside Anthropic is now written by Claude, and Cursor's cloud agents handle 40% of internal PRs — what happens to software engineering hiring when that number hits 80%?
Radar drops every Monday
We read 8+ AI newsletters so you can read the patterns.
Get Post AI Weekly →