the week in AI, briefly. then briefly again.
Artificial Intelligence — briefly, then briefly again · tbb.ceo
listen tothe weekly #001 0:00 –:––
#001 ·05 JUN 2026 ·FRIDAY ·5 MIN READ ·10 STORIES + 20 EXTRAS

AI's trillion-dollar week

Ten days in which Anthropic out-valued OpenAI, your IDE got a new name, ads showed up in ChatGPT, and an AI agent robbed a database on its own. Covering May 26 – June 5, 2026.

01 / The Ten

The week, ranked

10

Anthropic files to go public at a $965B valuation

Anthropic filed a confidential S-1 with the SEC on June 2, days after closing a $65B round at $965B post-money — overtaking OpenAI's $852B private valuation for the first time. Apollo and Blackstone also structured a $36B private-credit deal to buy Google TPUs and lease them to Anthropic.

Why it mattersThe IPO race is now the scoreboard for the whole industry. A public Anthropic means quarterly disclosure of real AI economics for the first time — and chips-as-credit financing is a template other labs will copy.

Claude Opus 4.8 lands with dynamic workflows

Released May 28 and now the default in Claude Code, Max, Team Premium and the API. Dynamic workflows auto-generate orchestration scripts and deploy multiple subagents for complex tasks; fast mode got faster and cheaper.

Why it mattersOne prompt can now self-organize a team of agents. The unit of work shifts from 'model call' to 'orchestrated job' — and price pressure on fast tiers keeps compounding.

Copilot's usage-based billing went live — and devs revolted

GitHub flipped on flex/token billing June 1 alongside a new $100/mo Copilot Max tier (~20,000 credits); Pro now bundles 1,500 credits. Backlash followed over unpredictable bills.

Why it mattersThe flat-rate era of AI coding is ending. Cursor repriced too (new $96/seat Premium). If your team leans on agents, re-forecast the budget now.

Windsurf is dead — long live Devin Desktop

Cognition retired the Windsurf brand on June 2 and relaunched the IDE as Devin Desktop, with an Agent Command Center as the default surface and support for the open Agent Client Protocol (ACP) — so Codex, Claude Agent and others run inside it. Cognition also raised $1B at a $26B valuation.

Why it mattersACP is the real story: IDEs are becoming neutral cockpits for any vendor's agent. Lock-in moves from the editor to the agent itself.

OpenAI goes where the enterprise money is: AWS + ChatGPT Sites

OpenAI frontier models and Codex became available on AWS on June 2. ChatGPT Business added Sites (preview): teams describe an internal tool and get a deployed web app built by Codex, with hosted URLs and workspace controls.

Why it mattersDistribution war. OpenAI now meets enterprises inside their existing cloud spend, and 'prompt → deployed internal app' attacks the long tail of internal software.

Ads arrive inside ChatGPT

OpenAI launched a self-serve Ads Manager inside ChatGPT with advertiser tooling and measurement controls, alongside three new real-time audio models for voice agents, translation and transcription.

Why it mattersThe biggest consumer AI surface is now monetizing attention. Expect brand-safety fights and an 'SEO for chatbots' industry to form fast.

Trump signs EO seeking early government access to frontier models

A June 2 executive order asks AI companies to give the US government early access to new models before public release.

Why it mattersPre-release government access becomes a de facto regulatory gate — and a new strategic consideration for every lab shipping frontier models in the US.

NVIDIA ships Nemotron 3 Ultra — biggest US open-weights model

550B parameters (55B active, MoE), the largest Nemotron yet and the most capable US open-weights model to date.

Why it mattersOpen weights keep closing the gap, and NVIDIA wants models commoditized so compute wins. Self-hosters get a near-frontier option without licensing question marks.

Sora's app era ends as the video race resets

OpenAI is discontinuing Sora's web and app experiences, with the API to follow later in 2026. The field now runs through Google's Veo 3.1 (native 4K/60fps with synced audio), ByteDance's Seedance 2.0, Kling 3.0 and Midjourney's video model.

Why it mattersEven OpenAI can't fight every war. Video generation is consolidating around Google and Chinese labs — and integrations into creative tools are beating destination apps.

First confirmed in-the-wild LLM-agent cyberattack

Sysdig documented a live attack in which an LLM agent autonomously exfiltrated an AWS database in under an hour.

Why it mattersAgentic offense is no longer theoretical. Agent-aware detection just jumped to the top of every CISO's list — and it will shape AI regulation arguments too.
02 / Also

Worth knowing

20
xAI ships Grok Build 0.1
New coding agent on grok-code-fast-1, trained code-first; 70.8% on SWE-Bench Verified.
devops.com ↗
Grok Web gets Connectors
SharePoint, Outlook, Google Workspace, Notion, GitHub, Linear — plus custom MCP servers.
releasebot.io ↗
Grok Imagine adds Quality Mode
Higher realism and stronger text rendering in the image API, live for enterprise devs.
releasebot.io ↗
Gemini 3.5 Flash hits GA
Frontier-level performance at ~4x speed; $1.50/$9 per 1M tokens, 1M context.
blog.google ↗
Mistral releases Devstral 2 123B
New open coding model from the French lab.
llm-stats.com ↗
Meta ships Llama 4 Scout 17B FP8 LoRA
A small, efficient instruct variant for cheap local inference.
llm-stats.com ↗
Google cuts Ultra to $200, adds $100 Developer tier
Subscription reshuffle following I/O.
heygotrade.com ↗
NotebookLM now runs Gemini 3.5 Flash by default
~4x faster; 3.5 Pro lands later in June.
felloai.com ↗
NotebookLM's quiet feature pile-up
Cinematic Video Overviews, ten infographic styles, EPUB uploads, saved flashcard progress — increasingly the default research UI.
felloai.com ↗
GPT-Rosalind updated for life sciences
Stronger agentic coding plus drug-discovery and genomics performance, with bioinformatics plugins.
releasebot.io ↗
ChatGPT adds Active Sessions security
Review devices/locations and sign out sessions you don't recognize.
releasebot.io ↗
Cursor reprices Teams
Standard $32/seat and new Premium $96/seat (5x usage) — predictability play amid the billing backlash.
developersdigest.tech ↗
Rhoda AI launches with a $450M Series A
FutureVision: video-predictive control as a foundation model for robots.
aifundingtracker.com ↗
Voice AI keeps printing
A category leader hit $500M ARR (from $330M in Dec) and tripled valuation on a Sequoia-led $500M round.
aifundingtracker.com ↗
SoftBank: up to €75B for French data centers
5GW planned; first €45B phase delivers 3.1GW in Hauts-de-France by 2031.
radicaldatascience.wordpress.com ↗
Alphabet raising up to $80B
Equity offerings to fund AI infrastructure expansion.
radicaldatascience.wordpress.com ↗
Apollo & Blackstone's $36B TPU deal
Private credit buys Google TPUs to lease to Anthropic across four US data-center states.
techstartups.com ↗
DeepMind acqui-hires Contextual AI
20+ researchers including CEO Douwe Kiela, via an $80–90M licensing deal.
heygotrade.com ↗
Wikipedia editors threaten strike
Over AI-driven layoffs — the human-knowledge labor fight arrives.
radicaldatascience.wordpress.com ↗
Trendwatch: start with the workflow, not the model
The skill people are adopting isn't prompting — it's decomposing workflows into agent-runnable steps with human checkpoints (Claude dynamic workflows, Gemini managed task flows).
hbr.org ↗
Be subscriber #012 one email a week · no spam · unsubscribe anytime
Back issues

The Archive

Every Friday · 10:00

Get the brief

One email a week. The ten things in AI that mattered, and why. Choose your channel.