The AI Friday Brief
Good Thursday, NOLA. May 7th is all about the big compute deals and what they mean for the AI economy. Anthropic just locked in a compute partnership with SpaceX worth $5B/year, massive infrastructure plays are reshaping who wins, and the conversation around agentic coding is getting serious. Plus, one popular take on why vibe coding and agentic engineering might be closer than we'd like.
Big Moves: The Compute Kingmakers
Anthropic and SpaceX ink massive compute deal: $5B/year for Colossus I access
Anthropic just signed up for 300MW of computing power from SpaceX's Colossus cluster, plus higher usage limits on Claude. This is the infrastructure story of the moment: companies that can secure massive compute wins entire product categories. The deal signals where the real competition is happening—not in model weights, but in who has access to the GPUs that run them. Discussion on HN.
Anthropic
Five architects of the AI economy explain where the wheels are coming off
TechCrunch sat down with five insiders who touch every layer of the AI supply chain at the Milken Global Conference. The conversation: what happens when compute gets cheaper, when every company has AI, and the easy wins are gone. Worth reading if you're thinking about product strategy in 2026.
TechCrunch
On Agentic Coding: Hype vs. Reality
Vibe coding and agentic engineering are getting closer than I'd like
Simon Willison's take on the blurring line between "I'll let the AI figure it out" and actual engineering discipline. The worry: as agentic systems get better, builders might stop thinking deeply about what they're building. It's a thoughtful read on the second-order effects of cheap code generation. Popular discussion on HN.
Simon Willison's blog
Code with Claude: The 5 biggest updates explained
Claire Vo from How I AI breaks down Anthropic's latest coding announcements and what they actually mean for builders shipping products. Less hype, more practical: what changed, why it matters, and what you can do with it today.
How I AI (podcast)
Tools, Releases & What You Can Build
Visualize any Hugging Face model in your browser
New tool that lets you instantly explore and visualize any open-source model from Hugging Face—see layers, parameters, architecture. Handy if you're picking a model or just curious what's under the hood. No signup, no waiting.
Hacker News
Making LLM training faster with Unsloth and NVIDIA
Unsloth (a library for speeding up fine-tuning) and NVIDIA are collaborating on optimizations. If you've been thinking about fine-tuning a model but got spooked by training time, this tooling is making it more accessible.
Hacker News
ProgramBench: Can language models rebuild programs from scratch?
A new benchmark that tests whether LLMs can reconstruct entire programs from minimal specs—relevant to the agentic coding conversation. The results? Better than you'd think, but there's still a gap between "it works" and "it's production-ready."
Hacker News
Industry & Platform Updates
VS Code's Copilot co-author attribution gets clarified
Microsoft has updated how Copilot contributions show up in commit messages. If you've been seeing "Co-authored-by: Copilot" in your repo, there's now more clarity on what that means and how to control it.
Hacker News
Telus uses AI to alter call-agent accents
Canadian telecom Telus is deploying voice transformation to modify agent accents in real time. It's a polarizing move—interesting from a technical standpoint (real-time voice style transfer works now), but raises questions about labor and authenticity.
Hacker News
FFmpeg developer calls out OxideAV for AI license laundering
A reminder that open-source licensing still matters: FFmpeg dev flagged a company using AI to rewrite their code and relicense it under a commercial license. The tooling is getting cheaper, but the law and ethics aren't moving as fast.
Hacker News
Worth a Listen
Code with Claude: The 5 biggest updates explained
Claire Vo from How I AI walks through Anthropic's latest announcements with a focus on what builders can actually ship. Fast-moving conversation, no fluff.
How I AI
Also
- Learning the Integral of a Diffusion Model — New approach to understanding how diffusion models learn. Technical but cool if you're into the mechanics.
- Anthropic's $1.5B joint venture with Wall Street firms (from yesterday)
- Why the A.I. Job Apocalypse (Probably) Won't Happen – Ezra Klein
- Specsmaxxing: Writing better specs in YAML to tame AI — Practical guide on how to give AI clearer instructions. Shorter specs = better outputs.
- A blueprint for using AI to strengthen democracy
- Marc Lore's Wonder: AI-powered restaurant factories (mentioned yesterday)
Pass it on
