The AI Friday Brief
Good Thursday, NOLA. May 28th brings a reality check wrapped in some genuine wins. Anthropic and OpenAI have found product-market fit — but that doesn't mean the hype cycle has stopped. Meanwhile, Cognition just raised $1B at a $26B valuation, and there's growing chatter about what happens when the shininess wears off and you actually have to ship.
Reality Check: The Market Settles
Anthropic and OpenAI Have Found Product-Market Fit
Simon Willison's thoughtful take: both companies have clear, defensible use cases and real adoption. This isn't hype — it's the market voting with actual spend. The flipside? It also means we're past the "everything is possible" phase and into the "here's what actually works" phase. Worth reading for a grounded take on where we are in the cycle.
Simon Willison
Using AI to Write Better Code More Slowly
Nolan Lawson's honest reflection on AI-assisted coding: it doesn't always make you faster, but it can make your code better. The real skill is knowing when to let AI lead and when to take the wheel back. This feels like the natural follow-up to last week's hype cycle — the actual workflow patterns are settling in.
Nolan Lawson
"I'm Tired of Talking to AI"
A candid essay on search fatigue and the creeping realization that not every question needs an AI response. Some interesting observations on what happens when quantity of answers exceeds quality of answers. Discussion on HN.
Hacker News
Big Money Moves
Cognition Raises $1B at $26B Valuation
The AI coding agent startup just hit the decacorn club with a fresh $1B raise. The bet? Coding is an uncapped TAM market and agents are the way to own it. This is the kind of signal that moves entire sectors — expect a wave of coding-focused agent startups over the next 6 months.
Latent Space
OpenAI Foundation Commits $250M to AI Disruption
OpenAI is putting its money where its mouth is with a quarter-billion dollars in grants for AI research and development. The foundation angle gives them room to think beyond quarterly earnings — a smart move for a company that needs to look long-term while shipping features at breakneck speed.
The Rundown AI
Tools & Releases
YouTube Custom AI Feed: Describe What You Want to Watch
YouTube is rolling out a new feature that builds a personalized video feed from plain English descriptions. It's a small thing on the surface, but it signals how search and discovery are fundamentally shifting — no more scrolling, just asking.
The Verge
YouTube's AI Label System Gets an Upgrade
YouTube is expanding its AI-generated content labeling. More transparency on what's synthetic, clearer signals for creators about disclosure requirements. This is the unsexy but essential infrastructure work that builds trust at scale. HN thread.
YouTube Blog
Claude Code as a Daily Driver: Full Skills Breakdown
A comprehensive walkthrough on actually using Claude Code in production. Covers subagents, plugins, MCPs (multi-turn conversation protocols), and real workflows. This is the kind of practical guide that's worth bookmarking if you're integrating Claude into your stack. Popular on HN.
Hacker News
Evolving Webflow for the Agentic Web
Webflow is rethinking its platform for a world where AI agents can interact with it directly. This is a quiet but important shift — moving from tools humans use to tools agents can operate. If you're building AI-powered workflows, this is worth paying attention to.
Webflow Blog
Search, Discovery, and Where the Audience Goes
DuckDuckGo Saw 28% More Visits After Google Praised AI Mode
When Google announced people love AI-powered search, DuckDuckGo's traffic spiked 28% — because people searching for a no-AI alternative suddenly had a rallying point. It's a good reminder that not everyone wants AI in their search, and there's a real market for opting out. Discussion on HN.
PC Gamer
Also
- Token Supply Growing at 3x/Year, Demand at 10x — Market Still Overheated — The gap between token availability and actual usage is still massive — good news for anyone building on APIs, but signals potential consolidation ahead.
- OpenRouter Doubles Valuation to $1.3B in One Year — The API aggregation play is paying off — developers like having model choice without vendor lock-in.
- DeepSWE: A Benchmark for Long-Horizon Coding Agents — Mentioned in previous brief; worth revisiting as the research effort matures.
- Vertu's AI Foldable for CEOs (Starting at $6,880) — Built on Hermes (open-source), enterprise integrations, ultra-premium finishes. Niche product, but a clear signal: AI-first mobile is coming for the high-end market.
- On "Agent Debt" in Coding: When Token Spend Doesn't Translate to Shipped Features — Real companies are burning through AI budgets and getting 20% ROI in shipped features. The reckoning is starting.
- Codex NPM Installs Up 10x Since January — Real Growth Signal — Developers are actually downloading and using Codex. This is adoption, not just hype.
- Local LLMs Perform Better When Taught to Ask Clarifying Questions First — Simple technique, meaningful impact: asking the model to request clarification before responding actually improves accuracy. Worth testing on your own systems.
- Apple macOS Kernel Vulnerability (CVE-2026-28952) Found by Claude — Claude discovered a real security flaw in macOS. Not the first time an AI model has found vulnerabilities, but a good reminder that these tools can do real work beyond writing.
- Ripgrep's No-AI Policy — The maintainer of ripgrep (a beloved CLI tool) is explicitly rejecting AI-generated code contributions. Interesting stance on maintainability and quality. <a href='https://news.ycombinator.com/item?id=48292718'>HN discussion</a>.
- A Reality Check on AI Jobs Hysteria — MIT Tech Review cutting through the noise: displacement is real, but the timeline and scope are often exaggerated. Worth a read if you're tired of doom.
- Everyone (Including Google) Is Navigating AI Security in Real Time — Security best practices for AI systems are still being figured out. If even Google is learning as it goes, your company probably is too.
- Prompt Politeness Affects LLM Accuracy (Research Finding) — A short paper showing that nicer prompts actually get better results. Small effect, but real. Worth knowing if you're optimizing prompts.
Pass it on