Agents that pay, test, and triage their own issues

CASE STUDY

How we built a software factory to drive Astro's GitHub issue count to zero — replaces manual issue verification with isolated AI subagents running in GitHub Actions, and Astro's maintainers report an 85% cut in open issues. The valuable part isn't the number, it's the shape: reproduction, patch testing, and preview releases as CI steps you already know how to debug. — Matthew Phillips

Microsoft open sources code-testing-generator, a polyglot unit-test agent — claims 92.1% task completion versus 78.9% for stock Copilot. The trick is boring and correct: "it reads a repository before writing anything — detecting the language, test framework, existing conventions, and the real build and test commands — then plans, writes, runs and validates the tests it produces." Repository-aware planning beats vibes. — Michal Sutter

Announcing Cloudflare Wallets: the programmable wallet for the agentic Internet — gives agents payment rails and verifiable web identities via the x402 protocol, so "agents can autonomously purchase APIs and content within clear safety guardrails." If your agent already calls paid APIs, the interesting question is spending limits, not payments. — Will Papper

COMPLIANCE

EU will mandate labels on authentic-looking AI content starting August 2 — means anyone shipping generative features to EU users needs provenance metadata and a disclosure path in the UI. Worth scoping now rather than after the first complaint. — vrganj

DeepSeek V4 Flash 0731 Intelligence, Performance and Price Analysis — the usual Artificial Analysis triangle of capability, latency, and cost — handy if V4 Flash is on your shortlist and you'd rather compare numbers than vibes before you route production traffic to it. — theanonymousone

Governments are making a dangerous bet on the AI boom — The Economist argues states are leaning hard on AI-driven growth continuing. Thin on specifics, but a useful frame if your roadmap assumes subsidised compute stays subsidised. — andsoitis

AI financial advice is surprisingly good, especially if you ask right questions — MIT Sloan reports advice quality rises with question specificity — a prompt-design finding dressed as a finance finding. No model comparisons or study details in the excerpt, so treat it as a hint rather than evidence. — foxtrot8672