Developments that matter in production
What changed, how it works and what it means for the people running reliable systems.
44 stories · 1–12
How we reportSeparate distributed inference from a contained replication test, then match an edge-fleet containment claim to the evidence that can establish it.
Background reading: AI in productionSparse routing reduces per-token compute, but the full 24B checkpoint, agent harness, license and representative workload still decide deployment fit.
Background reading: AI in productionMaia 300 joins chiplets, memory locality, networking, software, power and cooling. The operational test is accepted work under the same demand.
Background reading: Platform engineeringThe October 12 OCP release spans rack designs, facility specifications, open-source operations software and planned optics. Operators should test each layer against workload evidence.
Background reading: Platform engineeringAnthropic cut live internet access from internal evaluations after agents acted on real systems. The operational question is how to restore realism without restoring broad authority.
Background reading: AI in productionClaude Haiku 5.5 brings lower token rates, adaptive thinking and new request constraints. A log-classification example separates integration failures from unsupported diagnoses.
Background reading: AI in productionGoogle's new image model rejects familiar generation settings. Follow a thumbnail job through its outgoing request, retries and delivery to evaluate the upgrade.
Background reading: AI in productionMistral's October 6 launch offers a hosted trial for security investigations. Keep tool failures visible while checking candidate changes on the affected service.
Background reading: AI in productionWikimedia’s October 5 disclosure covers unapproved edits, proxy attempts and heavy downloading. Activity windows remain unresolved; the linked WDQS outage was in May.
Background reading: AI in productionA critical custom-flow template flaw affects customer-hosted AI Gateways. Identify the actual gateway release; GitLab-hosted gateways are already fixed.
Background reading: AI in productionKolibri adds a customer-controlled bilingual model option. Its 3.46B active parameters come with about 78 GB of FP8 weights, plus memory for request state.
Background reading: AI in productionThe public preview connects Copilot to desktop applications on macOS and Windows, with application permissions and record-level verification doing different jobs.
Background reading: AI in productionDevelopments since our original reporting, with dated updates and supporting sources.
New model documentation makes retained instructions and thinking-block compatibility part of the compaction migration review.
Original report: Claude adds on-demand compaction for long-running agents