Voice quality is table stakes now. The week Decagon, Amazon and Microsoft all shipped the fix for dead air mid-call, and the acceptance tests agencies should run before promising a resolution rate.
Last week Google did for enterprise AI budgeting what most agencies still have not done for themselves: it installed a meter. At Gemini at Work 2026 on October 8, Google Cloud introduced a single “universal” Gemini agent that plans work,…
On October 9, Google shipped two changes that look unrelated and are actually the same story. Demand Gen picked up a new “Overlay on videos” setting that lets Google write text onto your video creative — and it arrived switched…
This was the week agentic AI stopped being a conference talking point and started turning up in product documentation, media plans and client conversations. Google shipped a universal work agent and put a preview button on its automated ad copy.…
Google's EmbeddingGemma 2 puts embeddings on the device. How to decide what runs in production: local vs hosted models, pgvector storage, and why retrieval quality is still an architecture problem.
Text-to-speech stopped being the expensive part of a voice agent this year. Google’s Gemini 2.5 Flash TTS bills $10 per million audio output tokens, which converts to roughly $0.015 a minute of finished audio — about 90 cents per audio…
Two platform moves in the last week point in the same direction, and agency operators should read them together. Meta expanded its Muse AI agent to small businesses, wiring it into the software those businesses already run on. TikTok split…
On October 1, 2026, Google quietly revised its guidance on generative AI content. The edit added three sentences, but one of them lands squarely on agency workflows: the instruction to manually fact-check AI output now “also applies to metadata.” Titles,…
Google Ads automates campaign localization, ChatGPT turns chat into a fitting room, and hiring data reprices agent orchestration. What it means for agencies.
Two launches this week point the same way: retrieval quality is bought with fusion, not a bigger embedding model, and cost with quantization, not a bigger instance. The evidence, and the order of operations it implies.