Meta says interactions with Meta AI (text/voice/image) will start informing content and ad personalization on Dec 16, 2025. No global opt-out beyond not using Meta AI; UK/EU/South Korea are excluded at launch. For builders, treat assistant logs like ad-relevant telemetry: update privacy copy, consent flows, retention windows, and guardrails.

RTEB introduces a mixed open-and-private retrieval benchmark to measure real-world RAG generalization across domains like law, health, code, and finance; and ModernVBERT is a 250M-parameter, MIT-licensed visual document retriever that delivers SOTA-for-size performance with ready-to-use checkpoints and recipes for lightweight PDF/image RAG.

What developers need to know about Meta’s plan to fold AI-chat signals into ad targeting: scope, timeline, regions, and controls.

Starting Dec 16, 2025, Meta will use interactions with Meta AI (text/voice/image) to personalize content and ads; notifications begin Oct 7. No opt-out — don’t use Meta AI if you don’t want this.

If you build assistants or RAG features, treat conversational data like ad-relevant telemetry — update privacy copy, consent flows, and data-retention policies accordingly. (Rollout excludes UK/EU/South Korea initially.)

AI chat isn’t “outside” the ad stack — expect downstream effects on measurement, attribution, and user expectations across Meta’s surfaces.

“We will soon use your interactions with AI at Meta to personalize the content and ads you see.”

Applies only to people who use Meta AI; sensitive categories are excluded, but “sensitive” boundaries can be fuzzy — ship clear guardrails and disclosures.

Google ships a CLI and public API for its asynchronous coding agent (Gemini 2.5 Pro under the hood), targeting tightly scoped tasks and CI/CD/Slack integrations.

Hugging Face launches RTEB, a retrieval benchmark that mixes open + private datasets to better test real-world generalization for RAG/embedding models.

Multi-year Instinct roadmap (starting with MI450) with the first 1GW landing in H2 2026 — a real signal on accelerator supply, pricing, and co-design with top model customers.

A 250M-param vision-language retriever delivering strong doc-retrieval results for its size, with checkpoints and recipes released under MIT.

An MIT-licensed UI-element detector and evaluation methodology designed for agentic systems — pitched as an open, commercially-friendly alternative to restrictive licenses.

Stack Overflow’s convo with Kong’s CTO on agent-friendly APIs, the rise of MCP, and why security/observability must be first-class for agent integrations.

Another week of separating AI signal from noise. If we saved you from a demo that would've crashed prod, we've done our job.

📧 Got a story? Reply with your AI tool wins, fails, or war crimes. Best stories get featured (with credit).

📤 Share the skepticism: Forward to an engineer who needs saved from the hype. They'll thank you.

✍️ Who's behind this? The Augment Code team—we build AI agents that ship real code. Started this newsletter because we're tired of the BS too.

🚀 Try Augment: Ready for AI that gets your whole codebase?