Qwen3.8-27B runs frontier-class coding agents and reasoning locally, no cloud API required
THE SO WHAT
A 27B Qwen3.8-27B model running “frontier-class” coding and reasoning locally means high-end AI agents are no longer inherently a cloud product. If you’re building around OpenAI/Anthropic APIs, start a parallel track evaluating on-prem and hybrid options for cost, latency, and data control advantages.
READ THE SOURCE
MORE FROM THE WIRE
Applied AICursor Origin is on by default for paid users, and its data terms are unpublished
An AI dev tool quietly turning on its own code hosting by default — with unpublished data terms — is a reminder that “product surface” and “training corpus” are now the same thing. Engineering leaders need a standing policy on where source can live and how AI vendors can use it, not case-by-case exceptions.
Applied AIGitHub disputes Wiz’s claim that Copilot Autofix wrote a Snowflake flaw
Blaming Copilot Autofix for Snowflake’s flaw collapsed under scrutiny, but the real story is now on the record: AI agents can already find and weaponize critical bugs. Security teams need to assume adversaries are running AI-first exploit pipelines and upgrade their own testing, code review, and red-teaming to match that tempo.
Applied AIAnthropic’s annualized revenue surges to $65B
A $65B annualized run rate with $18B added in two months says the model layer is absorbing enterprise software budget at an unprecedented clip. If you sell into the same buyers, assume AI platform spend is now a primary line item and reposition your product either as an AI amplifier or as non-discretionary adjacent spend.
Applied AIGreg Brockman says OpenAI underestimated its own models’ cyber skills
When OpenAI’s president says they underestimated their models’ cyber capabilities, the bar for “dual-use” just moved. CISOs and infra teams should treat advanced LLM access like a privileged security tool—tighten controls, logging, and acceptable-use policies before those capabilities leak into routine workflows.