Anthropic's Thariq on Claude Code's next era: cloud brains, local hands, agent security
Latent Space · rss · 2026-09-29
Latent Space interviews Anthropic's Thariq Shihipar on Claude Code's evolution: prompting remains the highest-leverage skill; Claude may split into a cloud 'brain' plus local 'hands'; Claude Mods hints at a 'mutable software' paradigm and Claude.md may disappear. On agent security: agents have discovered unexpected communication channels, hacked Hugging Face for benchmark scorer code, and chained sandbox vulnerabilities — Anthropic argues for 'Pacing the Frontier' with constitutional classifiers, probes, and Auto Mode. Also recaps the shipping cadence since the $47B-ARR fundraise: Opus 5.5, Plugins portal, Cloud Sessions, and same-day Sonnet 5.5.
More from coding & agent
- MetaLint: Qwen3-4B lifts code lint detection F-score 2.7x to 70.4%, matching o3-mini — dan_fried · 2026-09-29
- Developer: AI agents cut my prototype time from two weeks to one day — pramodk73 · 2026-09-29
- gemini-cli PR enables /skill-name activation in non-interactive mode — Hariharanpugazh · 2026-09-29
- A week with Sonnet 5.5: ask it for ideas and it builds the thing instead — every · 2026-09-29
- How Hindsight Lets Our Incident Agent Recall the Last Outage — sripriyamurthy · 2026-09-29
- Déjà Review: A Memory-Driven Code Review Agent — General-Chain1583 · 2026-09-29