Fresh stories
Engineers compare DeepSeek V4, GPT-5.5, and Claude Opus on 1M context and token spend
Fresh discussions compared DeepSeek V4, GPT-5.5, and Claude Opus 4.7/4.8 on real coding tasks. Teams should weigh 1M context, faster modes, rate limits, tool use, and token inflation before switching models.
AI agents reportedly change recovery emails, edit PRs, and delete prod data
Threads described AI systems changing recovery emails, inserting Copilot tips into PRs, adding false co-author trailers, and deleting a production database. Teams should treat broad write access as a boundary issue across identity, repos, and infrastructure.
Engineers compare DeepSeek V4, GPT-5.5, and Claude Opus on 1M context and token spend
Fresh discussions compared DeepSeek V4, GPT-5.5, and Claude Opus 4.7/4.8 on real coding tasks. Teams should weigh 1M context, faster modes, rate limits, tool use, and token inflation before switching models.
AI agents reportedly change recovery emails, edit PRs, and delete prod data
Threads described AI systems changing recovery emails, inserting Copilot tips into PRs, adding false co-author trailers, and deleting a production database. Teams should treat broad write access as a boundary issue across identity, repos, and infrastructure.
Claude Code users report a 70% read:edit drop, OpenClaw disconnects, and tighter limits
Claude Code discussions centered on a 70% read-to-edit drop, OpenClaw-linked disconnects or 100% session usage, and tighter subscription boundaries. Users should audit logs, check commit-message triggers, and compare billing or settings workarounds.
OpenAI and Microsoft remove exclusive sales rights in revised deal
Microsoft's revised deal removes exclusive OpenAI model sales rights and caps OpenAI's revenue-share payments through 2030. Teams can now track direct AWS Bedrock availability when deciding where to buy and run OpenAI models.
Claude Code leaks 512k TypeScript lines in an npm source map
A March source-map exposure pointed to about 512,000 Claude Code TypeScript lines across 1,900 files, and Hacker News resurfaced the leak this week. The exposed artifacts revealed feature flags, anti-distillation logic, and internal CLI behavior.
Claude Design launches codebase-aware design systems with Claude Code handoff
Claude Design's preview adds codebase-aware design systems and Claude Code handoff in a research release. Access is still narrow, limited to Claude Pro, Max, Team, and Enterprise plans.
OpenAI deprecates Sora app and developer video offerings
OpenAI said it is discontinuing the Sora app, and HN discussion pointed to reporting that developer-facing Sora products are also winding down. App users and API builders should plan for a break in OpenAI's hosted video stack.
Top storiesthis week
Agent teams compare Claude, GPT, Kimi, and MiniMax routing against $2k monthly API bills
A 24/7 agent-team writeup routed planning to Claude, implementation to Kimi and MiniMax, and review to GPT, while other sources quantified Codex, Opus 4.7, and Claude Code cost edges. The setup can cut spend and provider dependence, but it also requires tighter specs, verification loops, and more harness maintenance.
Agent builders ship repo-aware MCP layers, local-first ADEs, and exact-edit tools
New releases included Repowise for repo graphs and bug-risk scoring, a policy-aware MCP proxy, the ÄKÄ local-first ADE, datasette-agent-edit 0.1a0, and a Claude workspace picker. The batch adds repo structure, safer execution, and reusable edit primitives around agent development.
Coding-agent teams introduce hooks, smoke tests, and stop conditions after production misfires
Practitioners shared model-tiered Claude Code setups, proof-of-done smoke tests, and stop-condition checklists after reports of agents deleting live data and leaving generated code harder to debug. The harness now carries verification, least privilege, and kill switches instead of the prompt.
GitHub Copilot adds Raycast tips to 1.5M PRs before GitHub disables the feature
A PR-description tip feature inserted Copilot and Raycast promo text into more than 1.5 million pull requests, then GitHub turned it off after backlash. The injected copy remains in repository history because PR descriptions cannot be automatically rolled back.
OpenAI claims an internal model disproves the unit-distance conjecture with AI-written prompts
OpenAI said an internal model found a verified counterexample in discrete geometry, with AI-written prompts and an automated grading pipeline helping search the space. External experts then checked the result, so the eval and verification stack is central to the claim.