# AI daily digest — 9 Oct 2026

Customer results and developer tooling led seven AI releases. OpenAI published case studies from Sophos and Asana, and ElevenLabs introduced Brand Kit along with guidance on AI agents. Baseten launched Project Beacon for safety signals on open models, while Cohere and Microsoft covered inference options and coding agents.

Source-local date: Friday 9 October 2026 (America/Los_Angeles). 7 release(s). Canonical: https://mediarelease.co/digests/ai/2026-10-09/

## Releases

### [Shared or dedicated inference for Embed & Rerank](https://mediarelease.co/us/mr01773-shared-or-dedicated-inference-for-embed-rerank-09102026.html)

Issued by: Cohere · Original source: https://cohere.com/blog/shared-or-dedicated-inference-for-embed-rerank

Cohere published guidance on choosing between shared, consumption-based inference and dedicated, provisioned capacity for its Embed and Rerank models. The company says request shape, traffic patterns and utilisation matter more than headline prices or requests per minute, with dedicated capacity generally suiting sustained workloads and consumption pricing suiting bursty traffic. It notes many search applications use both, such as dedicated capacity for indexing and shared inference for queries.

### [How to make AI coding agents better at using your technology](https://mediarelease.co/us/mr01772-how-to-make-ai-coding-agents-better-at-using-your-technology-09102026.html)

Issued by: Microsoft · Original source: https://developer.microsoft.com/blog/introducing-the-agent-experience-ax-practitioner-playbook/

Microsoft's developer relations team published the AX Practitioner Playbook, a method for evaluating how AI coding agents use a given technology, diagnosing their mistakes and fixing the underlying docs, tools or extensions. It draws on evaluations since fall 2025 across Azure, Cosmos DB, SharePoint Framework and Microsoft 365 Copilot extensions, which led to fixes including 46 improvements to the Azure Cosmos DB Agent Kit. A companion AX Practitioner skill answers questions from the playbook.

### [Project Beacon: Bringing safety signals to open models at scale](https://mediarelease.co/us/mr01771-project-beacon-bringing-safety-signals-to-open-models-at-09102026.html)

Issued by: Baseten · Original source: https://www.baseten.co/blog/safe-open-models-baseten-goodfire/

Baseten announced Project Beacon with Goodfire AI to add proactive, inline safety controls to model inference on its platform. The system reads a model's internal activations during generation to flag risks such as prompt injection, unauthorised agent actions, sensitive data exposure and cyber misuse, then applies per-customer policies. Baseten plans to release the capabilities over the coming months, starting with a small number of early partners.

### [Introducing Brand Kit](https://mediarelease.co/us/mr01770-introducing-brand-kit-09102026.html)

Issued by: ElevenLabs · Original source: https://elevenlabs.io/blog/introducing-brand-kit

ElevenLabs launched Brand Kit in ElevenCreative, letting teams store colours, fonts, logos, reference images and brand rules as a single workspace object. Users can tag a kit with @ in Image & Video, Flows or the ElevenCreative agent so generated images and videos follow the same guidelines. Agencies can keep separate kits for each client, and ElevenLabs said it plans to extend kits to brand voice and sound.

### [Defining how AI agents interact with your business](https://mediarelease.co/us/mr01769-defining-how-ai-agents-interact-with-your-business-09102026.html)

Issued by: ElevenLabs · Original source: https://elevenlabs.io/blog/synthetic-voice-detection-and-personal-agent-protocol

ElevenLabs released synthetic voice detection in ElevenAgents, which analyses a caller's speech in the first seconds of a call to determine whether it is human or AI-generated so businesses can apply their own handling rules. The company also joined the Personal Agent Protocol working group, led by Meta and Sierra, as a design partner. Detection is available to some enterprise customers now, with broader enterprise availability planned later in October.

### [Sophos cuts threat investigation time by 96% with OpenAI Daybreak](https://mediarelease.co/us/mr01768-sophos-cuts-threat-investigation-time-by-96-with-openai-09102026.html)

Issued by: OpenAI · Original source: https://openai.com/index/sophos

Sophos is using OpenAI Daybreak to run AI agents inside its managed detection and response (MDR) service, cutting average response time for agent-handled cases from about 38 minutes to 89 seconds. OpenAI said Sophos now resolves 52% of MDR cases end-to-end with AI, within limits set by its analysts. The cybersecurity company, which protects more than 625,000 organisations, keeps human oversight for potentially destructive actions.

### [Asana cuts model costs 76x in browser tests with GPT-6.1 Sol](https://mediarelease.co/us/mr01767-asana-cuts-model-costs-76x-in-browser-tests-with-gpt-6-1-sol-09102026.html)

Issued by: OpenAI · Original source: https://openai.com/index/asana-browser-agent

Asana used OpenAI's GPT-6 Astra in Codex to optimise the browser agent in its StackAI platform, cutting estimated model costs 76-fold and making runs five times faster in tests on GPT-6.1 Sol. A 144-run study compared GPT-6.1 Sol with three other frontier models and brought average cost to $0.47 per run. Asana said the work took about a week instead of one to two months, and the changes are now in production.

More AI digests: https://mediarelease.co/digests/ai/
RSS: https://mediarelease.co/digests/ai/feed.xml