AI Engineering Signal #39
Anthropic deprecates Extended Thinking for Claude Opus 4.6 and Sonnet 4.6, replacing it with Adaptive Thinking enforced by default
Signals
Anthropic deprecates Extended Thinking for Claude Opus 4.6 and Sonnet 4.6, replacing it with Adaptive Thinking enforced by default
any pipeline that explicitly invokes extended_thinking mode will break; audit API calls and update system prompts before the cutover date.
OpenAI Codex ships to ChatGPT mobile
background coding agents now run from phones; review API cost exposure for any Codex-connected workflows.
Web
Ontario audit finds AI medical note-takers routinely hallucinate clinical facts
any deployment in regulated care settings needs human review gates before records are finalized.
Web
vLLM publishes first TurboQuant accuracy and performance study
quantization routing decisions for serving infrastructure need a recheck against these benchmarks.
Web
OpenAI employee devices caught in TanStack npm supply chain attack
any CI/CD pipeline pulling npm packages needs dependency lockfile audits now.
Web
ArXiv institutes one-year ban for hallucinated references
preprint pipelines using LLM-assisted citation generation need validation steps before submission.
Web
The Take
The Adaptive Thinking forced migration and the TurboQuant study land in the same week: inference behavior is shifting under production systems without operator opt-in, and the quantization tradeoffs that govern cost are finally getting rigorous public benchmarks. Teams that have not locked down both their API call signatures and their serving stack assumptions are running on stale ground.
Subscribe
Related Signals