AI Agents Daily Briefing: Friday 13 November 2026
Good morning, here's today's briefing. The lead: Anthropic will let companies run Claude Managed Agents on their own servers. Also today: OpenAI launches agent tools that hunt for software bugs and propose fixes, Nectar ships an agent that finds creators and can message them, ElevenLabs opens a marketplace of industry voice agents, and J.Crew puts a Claude stylist on its website.
LEAD: Anthropic lets companies self-host Claude Managed Agents

Background: Anthropic first offered Managed Agents as a hosted service, then added self-hosted sandboxes so a company's code could stay inside its own network. The control plane, credentials and session history still lived on Anthropic's side. As agents started fixing builds, testing releases and working through bug backlogs unattended, some teams wanted the whole service in-house.
What happened: Anthropic opened a private beta of self-hosted Claude Managed Agents on Thursday. Customers run the Managed Agents server, its database and the workers that operate each sandbox in their own cloud or data centre. Agents, credentials and session history stay on the customer's systems. Model calls still leave the network: by default they go to Anthropic, or to Amazon Bedrock, Google Cloud, Microsoft Foundry, or a company's own gateway. Access is by request through an Anthropic account team. Nothing is copied across from the hosted service, so agents, skills, vault credentials and memory stores have to be created again. Webhooks, scheduled runs and "dreams" are missing, and Anthropic has not published a price. Anthropic, docs: get started, docs: differences
Why it matters: This is self-hosted orchestration, not an on-premises copy of Claude. For companies that will let agents touch production systems, who holds the session log and the credentials may matter as much as which model answers.
OpenAI launches Daybreak Red Team and Defense Factory

Background: OpenAI already runs Daybreak, a programme that vets people and organisations before they can use its most capable cybersecurity models, including GPT-6 Cyber.
What happened: At a keynote on Thursday, OpenAI president Greg Brockman said companies will be secure "only to the extent of the frontier model that they use to defend themselves and how well they utilize that." The company launched two products on its Daybreak platform. Daybreak Red Team, in early access, uses a fleet of agents, plus source code and application context, to test web apps for weaknesses, check reported bugs and try out patches. Nancy Wang, who leads cyber-applied AI, called it "superhuman pen-test at the scale of automated continuous scanning." Daybreak Defense Factory then takes findings, traces them through a codebase, and prepares a patch and a test for a human engineer to review before anything is merged. The agents currently run on GPT-6 Cyber. OpenAI Daybreak, Defense Factory, The Register
Why it matters: OpenAI is selling the idea that only frontier models can now defend a company, and it is putting its own cyber-tuned agents in that loop, with a person still meant to approve the fix.
Nectar launches Creator Agent for brand outreach

Background: Nectar, founded by former Meta staff Misbah and Farah Uraizee, already sells AI tools for social media, customer chat and creator partnerships. It raised a $30 million Series A led by Menlo Ventures in May. US spending on creator advertising is expected to reach $44 billion in 2026, according to the Interactive Advertising Bureau.
What happened: On Thursday Nectar launched Creator Agent, which it says can search more than 300 million creator profiles, review videos and past brand deals, check brand safety in context rather than by keyword, and rank matches, including smaller creators who are just getting noticed. It can draft and send messages in a brand's voice, help with contracts, gifting and shipping, and watch how campaigns perform on Instagram and TikTok. Brands still approve outreach. The agent is available globally today, and Nectar plans to offer it through the Model Context Protocol so teams can call it from Claude, ChatGPT, Gemini, Copilot or their own agents. TechCrunch
Why it matters: Influencer work is a natural target for agents because it is repetitive and spread across huge catalogues of people. The useful test is whether brands will let an agent send the message, not just build the shortlist.
ElevenLabs opens a marketplace of industry voice agents

Background: ElevenLabs is widely used for synthetic voices and voice agents in customer service, healthcare and finance.
What happened: ElevenLabs opened a Marketplace of industry-specific agents and apps built by partners, including Salesforce, IBM, NiCE, Twilio and Zendesk. Listings at launch cover financial services, insurance, healthcare, telecoms, retail and government, and enterprises can contact the partner to evaluate a deployment. Buying through an existing ElevenLabs commitment is planned for next year. Examples include Salient for borrower servicing calls, Assort Health for appointment booking, and Consio for Shopify support calls in 70 languages. ElevenLabs, Marketplace
Why it matters: Voice agents are moving from custom projects to a catalogue. The marketplace does not yet complete the purchase, but it shows how quickly vendors want to turn a model platform into a store of ready-made agents.
J.Crew puts a Claude stylist on its website

Background: Retailers have been adding AI shopping assistants, while many shoppers still start with general chatbots rather than a store's own tool.
What happened: J.Crew launched Style Advisor, a consumer stylist built with Anthropic's Claude, on its US desktop site, with a mobile app version planned later. It is trained on thousands of outfit pairings from the company's own photo shoots and styling rules, and can take an occasion, such as a not-too-formal wedding in Miami, and suggest a full look from J.Crew's catalogue. Chief information officer Danielle Schmelkin says it will later use a shopper's purchase history. The tool uses Fable 5.1 for trickier styling judgements, with Sonnet 5.5 and Haiku 4.5 handling quicker back-and-forth, and the company watches costs through a back-end observation layer. Fortune
Why it matters: This is a concrete retail deployment: the agent recommends complete outfits from the brand's own catalogue, rather than sending shoppers out to the open web.