Tracking AI capabilities inside social products, and the relationship, content, and community mechanics taking shape inside AI products.
Radar onlineLast added 2026.10.23 09:35 Beijing time
Status
Market
9 results
2026.10 monthly archive
Showing 9 items
Event dateSocial networksReleasedContent distribution · Audience insights
X rolls out new timeline recommendation architecture with AI/SI ranking for dwell time and retrieval for full reads
On October 20, Elon Musk announced two major recommendation model upgrades for X's For You timeline: a new AI/SI ranking model that significantly improves prediction of probable dwell time on individual posts, and a specialized retrieval model trained to surface content that users read all the way through with high completion rates. Musk also disclosed that total daily user active seconds on the platform reached all-time highs with exponential growth.
Why it matters: Marks a strategic shift in social media distribution away from shallow metrics like click-throughs, quick likes, and emotional bait toward AI-driven semantic evaluation of reading depth and dwell time. By filtering for full-read content at the retrieval layer and weighting dwell time in ranking, X directly penalizes shallow engagement bait while providing stronger distribution incentives for long-form creators and substantive analysis.
X · Global · First seen 2026.10.23 09:35 Beijing time
Event dateSocial networksTestingCreator tools · Content distribution
Bluesky Expands AI Platform Attie with Social Vibe Coding for Interactive Apps
At TechCrunch Disrupt 2026, Bluesky announced a major expansion of its experimental AI platform Attie, allowing creators to generate websites, quizzes, mini-games, and polls using natural language text prompts without traditional coding skills. Framed by CEO Toni Schneider as social vibe coding for communities, the generated micro-sites run on the decentralized AT Protocol with dedicated addresses (*.attie.site), linking directly to creators social identities and tapping into Bluesky network of 47 million registered users.
Why it matters: Breaks the isolation of current AI-assisted vibe coding as a solitary single-user developer workflow by tightly integrating natural language generation with decentralized social graphs and network distribution. By eliminating software development barriers and tying lightweight community experiences directly to open protocol identities, it introduces a novel paradigm for user-generated interactive experiences on the open social web.
Bluesky · Global · First seen 2026.10.18 09:35 Beijing time
Event dateCollaboration toolsTestingShared knowledge · Team memory
Microsoft Introduces Team Knowledge Base in SharePoint and Teams Curated by Copilot and Agents
Microsoft announced the Frontier Preview rollout of Team Knowledge Base in SharePoint, designed to empower human-agent teams across Microsoft Teams channels, M365 Groups, and the Copilot app. Team members can explicitly contribute emails, messages, meeting notes, and existing files, which Copilot and the Work IQ knowledge compilation engine synthesize into an interlinked library of agent-friendly markdown files, delivering up to 4x faster response times and 30% greater token efficiency for team teammates.
Why it matters: Enterprise knowledge bases have traditionally relied on manual documentation that quickly falls out of date and fails to provide contextual grounding for collaborative AI agents. By integrating automated knowledge compilation directly into Teams channels and SharePoint groups, Microsoft elevates the collaborative workspace into a dynamic context layer where human teammates and autonomous agents co-curate and share persistent organizational memory, advancing enterprise collaboration from single-user prompts to true multiplayer human-agent workflows.
Microsoft Teams · Global · First seen 2026.10.16 09:35 Beijing time
Event dateSocial networksReleasedContent distribution · Audience insights
Facebook Launches Your Algorithm to Let Users Tune Feed and Reels Recommendations Using Natural Language LLMs
Meta announced the launch of Your Algorithm on Facebook, a customization experience powered by large language models that is now available in the three-dot menu across Feed posts and Reels. Users can directly tell Facebook what topics they want to see in their own words, add custom preferences, or provide conversational feedback after tapping Interested or Not Interested, with recommendation ranking adapting in real time within the same session.
Why it matters: Social feed recommendations have traditionally relied on passive engagement signals like watch time and likes to infer user interest, often creating rigid algorithmic bubbles. By introducing conversational LLMs into direct ranking controls, Facebook shifts discovery from passive behavioral tracking to explicit user intent, establishing a two-way dialogue between users and the recommendation engine while reshaping niche content distribution for creators.
Facebook · Global · First seen 2026.10.15 09:35 Beijing time
WeCom Deeply Integrates with WorkBuddy Agent, Joining Company Directory and Unlocking Organizational Context
Tencent announced that WeCom has updated to version 5.0.12 and completed deep bidirectional integration with its native office agent WorkBuddy 5.8.0. WorkBuddy joins the WeCom contacts directory and sidebar as a conversational bot and supports unified WeCom account login. Under organizational authorization, it connects to ten core data capabilities including chat messages, emails, docs, calendars, meetings, and drive files to decompose complex tasks and return finished assets, while administrators centrally manage seat allocation, compute tokens, and data permissions in the admin console.
Why it matters: Enterprise AI tools have often operated as fragmented personal productivity software lacking organizational context and compliance boundaries. By turning a desktop agent into an authenticated organizational contact inside an enterprise messaging backbone, WeCom closes the loop between enterprise IM directory relationships, contextual work assets, task execution, and centralized governance, advancing workplace AI from individual personal use into coordinated team collaboration.
WeCom · China · First seen 2026.10.15 09:35 Beijing time
Google Cloud Unveils Gemini Agent and Persistent Coworker Agents With Enterprise Identities
Google Cloud introduced the universal enterprise Gemini agent in private preview at Gemini at Work 2026. The platform allows organizations to deploy persistent coworker agents with their own organizational identity, including a dedicated company email address, calendar, Drive storage, and directory listing, enabling team members to collaborate directly via Google Chat mentions and document comments.
Why it matters: Transforming AI agents from individual productivity tools into persistent digital teammates with dedicated corporate identities redefines multiplayer and human-AI collaboration in enterprise networks. Operating under distinct identities with audit trails and version histories establishes a new baseline for shared multi-agent systems embedded in workplace relationship graphs.
Gemini · Global · First seen 2026.10.10 09:35 Beijing time
Event dateCollaboration toolsReleasedMulti-agent collaboration · Group collaboration · Cross-app actions · Team sharing
Cisco Unveils Agentic Collaboration at WebexOne 2026, Bringing Claude into Webex Spaces and Meetings as Digital Teammates
At WebexOne 2026 on October 7, Cisco officially introduced its Agentic Collaboration architecture, shifting enterprise AI from reactive single-player chatbots into proactive digital teammates embedded in everyday team workflows. Powered by Anthropic, Claude Managed Agents can be invited into Webex spaces, meetings, and calls via @mentions to execute multi-step tasks across Webex and third-party applications using shared organizational context. The release also adds Model Context Protocol (MCP) integration with OpenAI dots and the Dialog harness for long-horizon execution and persistent relationship context.
Why it matters: Traditional enterprise collaboration tools have largely confined AI to meeting transcript summaries or isolated sidebar chat widgets, leaving a gap between group decisions and subsequent execution. By embedding autonomous agents directly into team spaces, video meetings, and organizational context, Cisco gives agents collaborative identities and cross-app action capabilities that continue executing after teammates log off, marking another key milestone in the transition of enterprise IM platforms into human-agent multiplayer hubs.
Webex · Global · First seen 2026.10.09 09:35 Beijing time
Atlassian unveils Agentic Multiplayer Protocol (AMP) and Rovo Work for human-agent multiplayer teamwork
Atlassian officially announced the Agentic Multiplayer Protocol (AMP) and Rovo Work mode at Team '26 Europe on October 7. Designed to shift AI from isolated single-player chat into multiplayer teamwork, AMP gives AI agents distinct identities, designated owners, scoped authority, and Teamwork Graph institutional context, allowing them to collaborate alongside humans in Jira, Confluence, Loom, and Slack channels with live presence and cursors. The new Artifacts app provides durable URLs for agent-generated assets across team workflows.
Why it matters: Marks the evolution of AI inside enterprise collaboration platforms from personal productivity assistants into a hybrid human-agent multiplayer network. By giving agents team identities, shared cursors, permission boundaries, and durable asset URLs, organizations can integrate autonomous agents as accountable digital teammates into daily collaborative workflows.
Atlassian · Global · First seen 2026.10.08 09:35 Beijing time
Instinct Expands AI Agents to Group Chats with Account-Free Guest Access and Dual-Layer Privacy
Personal AI agent startup Instinct has expanded its autonomous agents into group chats, rolling out to early-access users. The feature allows users to invite an agent into group threads to coordinate travel itineraries, organize carpools, split bills, and purchase tickets, with non-Instinct users able to participate without creating an account. The system uses a dual-agent architecture where the group agent is isolated from individual personal accounts, personal agents require explicit user permission before connecting or sharing context, and pending responses are automatically held when new participants join.
Why it matters: Represents a critical step for consumer autonomous agents transitioning from single-user personal assistants to multiplayer social threads and group collaboration. By introducing a group-specific shared agent, zero-signup guest interaction, and strict privacy boundaries separating personal context from group threads, Instinct provides a concrete architectural model for embedding autonomous AI into multi-party social coordination while safeguarding personal data.
Instinct · Global · First seen 2026.10.07 09:35 Beijing time