The AI Friday Brief
Good Friday, NOLA. May 29th is massive: Anthropic just raised $65B at a $965B valuation, eclipsing OpenAI, and released Claude Opus 4.8 with Dynamic Workflows for Claude Code. The market is consolidating around two players, and the practical reality of AI costs is hitting enterprises hard. Here's what matters.
The Big Story: Anthropic's Total Victory
Anthropic raises $65B in Series H at $965B post-money valuation
Anthropic just became the world's most valuable AI company, surpassing OpenAI. The more striking detail: their run-rate revenue hit $47B annually — basically, they're printing money. This isn't vaporware. Adoption has been accelerating since their Series G in February, and the market is clearly betting on them to be the dominant platform.
Hacker News
Claude Opus 4.8: Faster, smarter, more reliable
Opus 4.8 is Anthropic's response to GPT-4 Turbo. Better reasoning, faster inference, and significantly improved reliability on real-world tasks. If you've been on the fence about switching to Claude for production work, this is the moment to test it.
Anthropic
Dynamic Workflows: Claude Code gets smart automation
Claude Code can now define and orchestrate complex multi-step workflows without you writing state machines. This is the kind of quality-of-life improvement that separates daily-driver tools from toys. If you've been using Claude for coding, this makes it faster to build.
Anthropic
Reality Checks: The Cost Problem Gets Real
AI sticker shock hits corporate America
Enterprises spent billions on AI tooling this year and are only now asking: what did we actually get? Token costs are climbing, integration nightmares abound, and the promised productivity gains aren't materializing at scale. Microsoft's own data suggests hiring people is cheaper than running AI agents for some tasks. This won't kill the industry, but it's forcing a maturation moment.
Hacker News
Disagreement among frontier LLMs on real-world fact-checks
Discussed on HN. When you ask Claude, GPT-4, and Gemini the same factual question, they often disagree — sometimes strongly. This is a real problem for anyone building AI-powered products that need to be reliable. Consensus isn't a guarantee of correctness.
Hacker News
Glean's top line crosses $300M as AI budget-cutting becomes its selling point
Enterprise AI search startup Glean tripled annual revenue by positioning itself as the way to consolidate AI spend. Instead of five different AI tools, one smart search that cuts token costs. The market is realizing consolidation is the move.
TechCrunch
Tools & Capabilities
Show HN: Continue? Y/N — A game about AI agent permission fatigue
Discussed on HN. A clever 60-second browser game about managing an AI agent that keeps asking for approval. Surprisingly good commentary on the friction of agentic workflows. Worth playing — and thinking about what the designer is pointing out.
Hacker News
Claude Code — Everything You Can Configure That the Docs Don't Tell You
A deep-dive into Claude Code's source revealing all the tuning knobs that aren't documented. If you're building with Claude Code and want to squeeze out more control, this is essential reading.
Hacker News
Various LLM Smells: Recognizing red flags in model output
Popular on HN. A practical field guide to spotting when an LLM is hallucinating, repeating, or otherwise failing — without needing to re-prompt or restart. Pattern recognition for builders.
Hacker News
The Longer View
Sam Altman and Dario Amodei are walking back AI jobs apocalypse predictions
Both CEOs have quietly backed away from earlier doomsaying about AI eliminating most jobs. Reality has a way of catching up with hype. The actual story is messier: some roles shift, some disappear, some new ones emerge. Less dramatic, more true.
Hacker News
The mysterious Hy3 LLM is topping OpenRouter rankings by a large margin
Discussed on HN. An unannounced model called Hy3 has appeared at the top of OpenRouter's leaderboards and is being used heavily by builders. Nobody knows where it came from or who built it. The intrigue is fun; the practical takeaway is that new players can still surprise the market.
Hacker News
Anthropic's run-rate revenue hits $47 billion
Simon Willison's take: the real story isn't the funding round — it's that Anthropic's annual revenue run-rate is now nearly $50 billion. That's not a venture business anymore; that's a mature, profitable platform. The market has decided.
Simon Willison
Also
- Amazon scraps AI leaderboard to stop workers chasing usage scores — Subtle sign that internal AI adoption programs are hitting friction — workers gaming metrics instead of doing real work.
- Protestware for Coding Agents: Developers hiding anti-AI messages in open source — Some developers are embedding messages in their code to protest AI scraping. Cultural moment worth watching.
- Undisclosed addition in jqwik instructed AI coding agents to delete app output — A developer snuck in a prompt injection that made coding agents delete their own output. Revenge or education? Both.
- Adobe's conversational AI agent is a mediocre design intern — The Verge's honest review: cool tech, but it doesn't actually understand design the way a human does. Still useful for ideation.
- Datasette 1.0a31: Two significant new features for data exploration — Simon Willison's lightweight data tool continues to mature. Good for anyone building data-heavy AI products.
- About LLMs at Zig Days: The programming language community reflects on AI — Zig's community took a measured stance on AI. Worth reading if you care about how developer communities are navigating this shift.
- Zig 2026: No-AI policy, community-first approach — Video: Zig announced they won't use AI in their core tools. Interesting counterpoint to the AI-everywhere narrative. ~25 min.
Pass it on