Microsoft PDD Kit: agents write Rust, Verus proves it correct

A prototype pairs coding agents with the Verus prover, so you review intent, not code — while Copilot cuts three models Nov 16 and Vercel kills double-runs.

Nowline NOV 6 6:00 PM banner

Top AI stories from the last hour

Top AI stories from the last hour

Copy markdown

  • Proof-driven dev, not vibe-checking

    PDD Kit runs inside GitHub Next's Chopin: you state intent in plain English — "you only get a token with the right password" — the agent writes Rust plus Verus proofs, and Verus checks them against every input before opening a PR. On a 5,000-line app it hardened the login handler and surfaced existing bugs while the developer never read the generated code. It's a research prototype with no public download yet, but Verus itself is open, so you can start proving your own auth logic today.

  • Copilot drops Opus 4.8, Opus 5 and GPT-5.6 Terra Nov 16

    GitHub deprecates all three across every Copilot surface — Chat, inline edits, ask and agent modes, and completions — on November 16, steering you to Claude Opus 5.5 and GPT-6.1 Sol or GPT-6 Luna. Annual individual subscribers keep Opus 4.8; everyone else should update model policies and integrations now so pinned requests don't silently fall back.

  • Vercel's Workflow SDK gets idempotent starts

    start() can now atomically claim a hook token, so a duplicate start with the same token throws HookConflictError instead of spawning a second run. If you trigger workflows from webhooks or retries, that kills double-execution without you maintaining a dedup table.

  • Kimi K3.1 Pro lands on Vercel AI Gateway with ZDR

    Moonshot's Kimi K3.1 Pro and K3.1 Pro Highspeed are now callable through Vercel AI Gateway on Zero-Data-Retention providers, with automatic failover and request traces. A privacy-safe way to route to a strong Kimi model without wiring up Moonshot directly.

  • Copilot reaches JetBrains Air; usage API splits out BYOK

    Copilot now runs as an agent inside JetBrains Air over the Agent Client Protocol, with tasks executing locally or in the cloud on your Copilot plan. Separately, the usage-metrics API now reports bring-your-own-key and enterprise-custom-model usage on their own (used_local_byok, is_custom_enterprise_model), so you can finally see what your team actually runs.