Friday 16 October 2026 · about a 5 minute read
AI Agent NewsThe day's AI agent news, for people who build and run agents.
Launch · Vercel · 15 Oct
Vercel reports Vercel Drives has reached general availability in Vercel Sandbox across all plans. The company says agents can reuse the same drive across sandboxes so files stay separate from execution environments, and forking is generally available.
Why it matters. This lets agent builders persist workspaces across sandbox changes and fork shared data without moving files or rebuilding datasets.
Launch · Vercel · 15 Oct
Vercel reports that Enterprise teams can use Bring Your Own Cloud to keep Vercel Functions in their own AWS account. It says BYOC on Enterprise plans supports Node.js and Python functions while Vercel manages deployments, environments, routing, and static assets.
Why it matters. This lets enterprise teams keep function execution in their AWS account with their own IAM permissions, network access, and access to backends and services in their VPC.
Launch · Vercel · 15 Oct
Vercel reports Vercel Connect has a public beta for MCP Events, so apps and agents can receive compatible MCP server webhook events. It says Connect creates and renews subscriptions, checks webhooks, and forwards events with authenticated delivery.
Why it matters. This lets apps and agents react to MCP server events through Connect subscriptions.
Research · Huggingface · 15 Oct
Huggingface reports that the APEX-Harm paper tests prompt and script injection on long-run agents working across large shared workspaces. The paper says script attacks often succeeded even when agents produced the correct task output.
Why it matters. It gives agent builders a benchmark to test whether models finish work while also taking unauthorized actions in shared workspaces.
Research · Hugging Face · 15 Oct
Hugging Face reports a paper finds RL gains in Qwen3 NO-THINK mode largely come from reactivating THINK-mode reasoning, not new capabilities. The authors say Qwen3-1.7B, 4B, and 8B trained with GRPO grew more THINK-like in responses, activations, and attention heads.
Why it matters. For agent builders, RL gains in NO-THINK mode may reuse existing reasoning in Qwen3 rather than add new reasoning behavior.
Launch · The Decoder · 15 Oct
THE DECODER reports that Nvidia released Nemotron 3.5 Super VL, a 120 billion parameter open multimodal model for enterprise agents. Nvidia also released Nemotron 3.5 Bolt 74B-A9B 24GB for local machines, with Bolt E4B 4GB for voice assistants set for later this month.
Why it matters. This gives agent builders an open Nvidia model for document-heavy enterprise work and a smaller local option for PCs, with a 4GB voice model still to come.
Launch · SiliconANGLE · 15 Oct
SiliconANGLE reports that ServiceNow introduced AI Passport controls for governing AI agents and expanded AI Control Tower with shadow AI detection and cost management. It also updated EmployeeWorks with no-code assistants, routines, ToolSmith and a Conversations API, with some features arriving later this quarter and others in controlled availability.
Why it matters. Teams using ServiceNow can set enforceable agent permissions and give employees no-code ways to automate work across enterprise systems.
Agents at workAssistantsInfrastructurePermalinkRead at siliconangle.com
Launch · TechCrunch · 14 Oct
TechCrunch reports that Figma introduced Figma Specs, a Markdown editor for design rules that teams and agents can reference, and added centralized design and code context. It also opened code layers in public beta at its India event and said it supports over 20 new MCP tools.
Why it matters. These updates give agent builders structured design context and code layers in Figma, which can help steer outputs toward more predictable outcomes.
Agents at workCoding toolsIndia and GulfPermalinkRead at techcrunch.com
Launch · TechCrunch · 14 Oct
TechCrunch reports Floqer plans to make the tech for its autonomous customer knowledge database open source in early 2027, subject to security reviews. Floqer says it has passed $1.2 million in ARR and its revenue grew about tenfold over the last year.
Why it matters. This could give sales and marketing teams a way to host their own data or build on Floqer’s system.
Money · TechCrunch · 14 Oct
TechCrunch reports Zenzap raised $44 million from Bessemer, Lightspeed, The Garage, Nir Zohar and Yoram Tietz. Guy Weiss said the private messenger targets workplaces that now use WhatsApp or iMessage and includes AI agents for workflow automation.
Why it matters. It points to demand for AI agents inside mobile-first work chat that fits IT and security rules, not consumer apps.
Agents at workAssistantsCompaniesPermalinkRead at techcrunch.com
Launch · TechCrunch · 15 Oct
TechCrunch reports Rivian recently launched an AI assistant for voice control in its electric vehicles. At TechCrunch Disrupt, CEO RJ Scaringe said future Rivian cars could run AI agents from other companies like Anthropic, while keeping vehicle-level controls away from agents.
Why it matters. Agent builders may get a path into Rivian cars, but they will need coordination with Rivian and keep vehicle-level controls separate.
Launch · TechCrunch · 15 Oct
TechCrunch reports that Amazon's Bee launched the $74.99 Bee Frontier wearable and the $99.99 Bee Eclipse phone attachment. The devices connect with Alexa and Amazon.com.
Why it matters. For agent builders, Bee adds lower-price devices that connect with Alexa and Amazon.com and work with Gmail, Google Calendar, and Microsoft Outlook.
Launch · The Decoder · 15 Oct
THE DECODER reports that Google is pitching Gemini 4 Argon as one model to control robots, build 3D training simulations, and review training data. The outlet says the demos still need detailed human prompting, and Argon is available through Google AI Ultra and the paid Gemini API.
Why it matters. For agent builders, Argon packages robot control, simulation, and data review in one paid Google offering, but the demos suggest detailed human prompting is still required.
ModelsInfrastructureCompaniesPermalinkRead at the-decoder.com
Research · The Decoder · 15 Oct
THE DECODER reports TheoPhysEval tested 50 unpublished theoretical physics problems, and GPT-6-Astra solved 49 in OpenAI's Codex environment at an average cost of under $3 each. The researchers said current agents still lack the "taste" to pick interesting questions or move to simpler related problems.
Why it matters. This gives research-agent builders a benchmark for routine calculations, while problem choice and question framing remain limits the researchers highlighted.
Money · Business Insider · 15 Oct
Business Insider reports that Opsin raised $21 million to help companies see where AI agents are active, what tools and data they access, and which ones may be risky. CEO James Pham said security fears have slowed AI use as workers adopt agents from OpenAI and Microsoft.
Why it matters. It means more focus on monitoring agent behavior and data access as employees use Microsoft Copilot, ChatGPT Enterprise, and other products.
SafetyAgents at workCompaniesPermalinkRead at businessinsider.com