Voice agents and enterprise integrations moved to the foreground: Google introduced Gemini 3.8 Live with extended reasoning and background tool use, while Anthropic put Salesforce workflows inside Claude with 37 pre-built sales skills. These launches point toward conversational systems that do work across business applications rather than merely answer questions.
Agent-assisted engineering also produced unusually concrete operating data. Nous Research says 1,393 Hermes subagents reduced a million-line Python codebase by 34.4% in nineteen hours, while Perplexity says hundreds of coding agents helped two engineers build its CobbleDB serving infrastructure in two months. AI safety remained contested at Dreamforce, where Jensen Huang and Dario Amodei offered different views on how the industry should manage frontier risk.
OpenClaw's official account announced a practical local-assistant session with Hugging Face, NVIDIA AI and Ant Ling AGI, demonstrating Ling-3.0-flash on DGX Spark. This is a new ecosystem event rather than a release or channel change.
Google DeepMind introduced two conversational models designed to reason, understand visual input in near real time, recognize 97 languages automatically, and call tools in the background. The combination matters because it moves voice interfaces beyond turn-by-turn chat toward agents that can continue handling tasks without interrupting the user's conversation.
Anthropic made Salesforce in Claude available in beta with 37 pre-built sales skills covering account research, call preparation, deal review, dashboards, and forecasts. Bringing governed business data and actions into the conversation could reduce context switching for sales teams, while the beta label signals that enterprises should still evaluate permissions and operational controls carefully.
Nous Research says Hermes Agent dispatched 1,393 subagents over nineteen hours to clean up a million-line Python codebase, reducing it by 34.4%. The company estimates the work represented nearly $2 million in engineering hours, offering a large-scale example of autonomous coding coordination while leaving the quality and savings claims dependent on Nous's own measurement.
Perplexity published research on CobbleDB, a key-value database serving prepared web content for its search product. It says two engineers, assisted by hundreds of proactive always-on coding agents, built the core infrastructure in two months—a concrete case study in humans retaining architecture, review, and production authority while agents perform sustained implementation and inspection.
At Salesforce's Dreamforce conference, NVIDIA CEO Jensen Huang framed safety primarily as an engineering, testing, and accountability problem, while Anthropic CEO Dario Amodei emphasized frontier risks and stronger safeguards. The disagreement matters because infrastructure suppliers and model labs are helping shape whether emerging safety practices remain voluntary, become independently evaluated, or turn into formal regulation.
Selected posts from the edition, with links to the originals.
“Build your own local AI assistant. Join OpenClaw, Hugging Face, NVIDIA AI and Ant Ling AGI for a practical session on Ling-3.0-flash and DGX Spark, with a workflow demo and Q&A.”
View post on X ↗“We’re introducing Gemini 3.8 Live and 3.8 Live Extended Thinking—our best conversational AI. The models talk, think, and handle tasks in the background without breaking your flow.”
View post on X ↗“AI safety matters. Strong engineering, rigorous testing, and accountability are how we get it right. Jensen Huang on moving quickly, innovating, and leading responsibly.”
View post on X ↗“We had a million lines of Python to clean up. 1,393 subagents and nineteen hours later, the codebase was 34.4% smaller, saving us nearly $2m in engineering hours.”
View post on X ↗“We’re publishing research on how we built CobbleDB, our key-value database that serves web content for Perplexity search. Two engineers and a team of hundreds of proactive, always-on AI agents built the core infrastructure in two months.”
View post on X ↗“Salesforce in Claude is now available in beta. It brings your accounts, opportunities, and pipeline into Claude, with 37 pre-built sales skills.”
View post on X ↗“A few thoughts on our Humanist AI Code of Conduct that we released for consultation yesterday.”
View post on X ↗“Can we allocate more compute to the harder problems? If your GRPO group has all wrong completions, sample more in search of batches with nonzero gradient. It works. Called ‘Never Give Up.’”
View post on X ↗“The hardest part of an ‘org harness’ or enterprise AI is auth, and managing auth in a multiplayer channel like Slack. We’re thinking a lot about how to do this right for managed deep agents.”
View post on X ↗“Enterprises must stay sovereign, keep and control their alpha, and pay for AI based on value delivered—not compute consumed.”
View post on X ↗