the dam brief #16

The Headline

the week's biggest story

Anthropic launches Claude Opus 5 for long-running agent work

Anthropic launched Claude Opus 5 on July 24, calling it a daily-use Opus model that comes close to Claude Fable 5 at half the price. the company says Opus 5 is now the default model on Claude Max and the strongest model on Claude Pro, with improvements on coding, knowledge work, and long-running agents.

the practical test: give it one long job that normally falls apart across tabs, files, and follow-ups, then check whether the review work gets smaller or just moves to the end.

(Source)

The Build

tools, models, and dev releases

• code-review-graph gives coding agents a local map before review - the MCP and CLI project builds a graph of file relationships so an agent can inspect the part of a large repo it needs instead of reading everything (GitHub)
• Kimi Code CLI puts another coding agent in the terminal - the repo ships a runnable CLI that can read and edit code, run shell commands, search files, fetch pages, and use Kimi or compatible providers (GitHub)
• ui-skills routes agents through UI-specific skill sets - `npx ui-skills start` gives design and frontend agents task-specific guidance for UI work instead of another generic prompt file (GitHub)
• Claude HUD makes Claude Code less of a black box - the plugin keeps context usage, active tools, running agents, and todo progress visible under the input box, which is the kind of operating layer solo builders need (GitHub)
• Hallmark gives coding agents a design anti-slop gate - the skill package for Claude Code, Cursor, and Codex helps generated pages choose different structures and reject default AI-looking layouts (GitHub)
• OpenSEO is a self-hostable search-ops workbench - the repo packages rank tracking, audits, and agent hooks into a cheaper Semrush/Ahrefs alternative that small teams can inspect or run themselves (GitHub)

Founder Fuel

funding, acquisitions, and market moves

• Prentis is reportedly raising $100M for routine-computer-task agents - TechCrunch says the new AI lab from Reid Hoffman and Mark Pincus is betting that automating ordinary computer work grows faster than coding alone (More)
• Cognition buys Poke to shape Devin's personality layer - the acquisition gives the coding-agent company a team focused on conversational style, interaction design, and the product feel around agent work (More)
• AegisAI raises $36M to fight AI-driven spear phishing - former Google security executives built agents that inspect small message anomalies instead of relying only on static phishing checklists (More)
• AMD introduces Helios for rack-scale AI systems - the company says the system will ship later this year, putting more pressure on Nvidia's full-stack data-center story (More)

Longevity & Health

aging, biotech, and health research

• OpenAI launches Health in ChatGPT for eligible US users - the feature connects medical records and Apple Health so people can ask questions against their own data instead of generic symptoms alone (More)
• a sulfur compound may help aging muscle repair - ScienceDaily reports that LASSS protects a repair protein tied to damaged muscle, pointing at one route for preserving strength as people get older (More)
• APOE2 may protect the aging brain from Alzheimer's signals - researchers linked the variant to lower disease risk and better brain resilience, while the treatment path remains early biology rather than a pill (More)
• patient-grown mini brains may predict Alzheimer's treatment response - organoids built from patient cells showed different responses to an antidepressant, which could help sort therapy choices before human trial-and-error (More)

The World

regulation, policy, and geopolitics

• the European Commission says TikTok minors' accounts may breach the DSA - Brussels sent preliminary findings that the platform failed to meet required child-safety standards under the Digital Services Act (More)
• Google gets a €890M EU fine for favoring its own apps - the BBC reports that regulators say Google pushed its own services above rivals, keeping search ranking inside Europe's platform-power fight (More)
• Trump vows to investigate the EU over US tech fines - the BBC says the threat followed Brussels enforcement against American tech companies, turning European platform rules into a trade-politics fight (More)

Go Deeper

long reads, podcasts, and documentaries

• Silicon Valley is completely divided over Chinese AI - Wired | staff. a useful read on why smaller AI players and frontier-lab incumbents see open Chinese models through different business incentives. (Read)
• European researchers say Big Tech is blocking access to their data - Wired | staff. the DSA gives researchers new rights to study platform harms, but access still depends on the companies being studied. (Read)
• gassed up - Longreads | Shane Mitchell. tomato ripening rooms are a cleaner systems story than most AI operations posts: consolidation, timing, logistics, and the hidden cost of making freshness look simple. (Read)
• should we do more to fight lookism? - The New Yorker | staff. a status-and-appearance essay for anyone watching optimization culture move from software into bodies, hiring, and social life. (Read)

The Colony

events and community

• next friday: July 31 - build fridays in Berlin, Hamburg, Lüneburg, and Frankfurt - four workbenches, one rule: bring a laptop and move something real forward (Berlin, Hamburg, Lüneburg, Frankfurt)
• next saturday: August 1 - agentic coding studio, Hamburg - a focused room for builders working with coding agents, without publishing the venue before registration (RSVP)