Working Surfaceshipped

What the tools released

Releases that change how product, design and engineering teams work: features in their tools, new AI models, and major products from the AI labs. Each is logged from the maker's own page.

Product
Category
138 releases · since 14 September 2026
17 Nov 2026
Microsoft
MAI-Code-2 and MAI-Code-2-Flash, Microsoft AI models for agentic coding and for high-volume everyday coding, in GitHub Copilot
Engineering teams in GitHub Copilot can send multi-step refactors to MAI-Code-2 and routine coding to the cheaper Flash version, choosing cost per task.
17 Nov 2026
Vercel
eve 1.0, an open-source framework for building AI agents for production, generally available
Engineering teams can build their own AI agents with chosen AI models, tools and approval steps, with durable runs, Slack sessions and tracing included.
agents · vercel.com
17 Nov 2026
Microsoft
Microsoft Copilot SDK 2.0, the Copilot AI agent harness for a team's own apps, with a shared runtime and per-user identity
Engineering teams can build AI agents into their own products that act as the requesting user. Sandboxes are paid for only when a session needs a shell or files.
17 Nov 2026
OpenAI
GPT-6.2 Astra and GPT-6.2 Sol, OpenAI AI models for demanding multi-step work and for everyday coding and agent work
Engineering teams can give long, multi-step coding tasks to Astra and routine agent work to the cheaper Sol, in Codex, ChatGPT and the API.
O
13 Nov 2026
Other
Atlassian Code Reviewer, an AI agent that gives a first review on every pull request in Bitbucket Cloud and GitHub, using Jira context (generally available)
A team on paid Jira gets an automatic first review of each pull request, checked against its Jira work, so human reviewers can focus on what needs judgement.
agents · critique & review · atlassian.com
12 Nov 2026
Google
Gemini 4 Hydrogen, a fast Google AI model for everyday agent work, with a coding orchestrator that switches AI models within one task
Engineering teams can run routine agent work on a cheaper, faster AI model. In Google Antigravity, one session can hand quick steps to Hydrogen and hard reasoning to Carbon.
models · agents · blog.google
12 Nov 2026
Claude
Self-hosted Claude Managed Agents, the full AI agent service run in a team's own cloud or data centre (private beta)
A team can run Claude AI agents with session history, credentials and network limits inside its own systems. It also operates the server, database and workers itself.
agents · claude.com
11 Nov 2026
Google
Gemini Decision 1 and the Gemini Decisions API, an AI model that scores up to 1,000 yes-no, multiple-choice or rubric questions about one input
A product team can grade content, check policies or route requests with one fast call that returns answers with confidence scores instead of text. Low-confidence cases can go to a person.
models · blog.google
O
11 Nov 2026
Other
AWS Emulator, a free local copy of more than 200 AWS services that developers and coding AI agents can test against offline
An engineering team's AI coding agents can run and check code that calls AWS on a laptop, without a paid account or a deployed test environment.
10 Nov 2026
Claude
Claude Whiteboards and Claude Sheets (beta), and Claude Data, formerly Dashboards, now generally available
A product team can plan on a shared board with Claude and turn notes into timelines with owners. It can also ask questions of company data against agreed metric definitions.
products · collaboration · claude.com
10 Nov 2026
Cursor
Multiplayer Projects in Cursor, a shared workspace where teammates and their coding agents plan and build together (beta)
An engineering team can share one task board, specifications and decisions across every member's coding agents. Each person's coordinating agent can spot overlapping work and agree an order with the others.
agents · collaboration · cursor.com
O
9 Nov 2026
Other
PostHog Self-Driving generally available, AI agents that study product analytics and errors and open pull requests with fixes
A small product team can receive a ranked list of problems found in its own usage data and errors, each with a proposed fix as a pull request to review.
agents · products · posthog.com
9 Nov 2026
Meta
Muse Ember, a small, low-cost AI model for the repeated work of AI agents
Teams can run the routine steps of an AI agent, such as tool calls, coding subtasks and document parsing, on a cheaper AI model through Meta's API.
9 Nov 2026
Claude
Claude Fable 5.2 and Claude Mythos 5.2, AI models for hard coding, reasoning and vision work, with an optional fast mode
Teams can give long software migrations and hard engineering problems to Anthropic's most capable AI model at a lower price than Fable 5.1. Output arrives up to 30 percent faster.
models · anthropic.com
6 Nov 2026
GitHub
GitHub in Slack and Teams generally available, turning a chat thread into issues, code changes and pull requests
A team can start an investigation, a code change or a pull request from the Slack or Teams conversation where the problem was raised. In Teams, members can review the plan and preview together beside the discussion.
agents · collaboration · github.blog
6 Nov 2026
Cursor
Voice chat with agents in Cursor, a spoken conversation for directing and checking on coding agents
Engineers can describe a change aloud, ask a running coding agent what it changed, and hear short progress updates instead of reading the full written reply.
agents · cursor.com
6 Nov 2026
Google
Voice mode in Google Antigravity, for steering a coding agent by speech while it works
Engineers can give a coding agent a task, correct it mid-run and answer its permission requests by voice, without returning to the keyboard.
O
5 Nov 2026
Other
Stripe analytics data for AI agents, through Stripe's MCP server and its Queries and Reports APIs
Finance and product teams can ask an AI agent such as Claude or Cursor to analyse their Stripe payment data. Each answer shows the query behind it, so a person can check it.
agents · stripe.com
5 Nov 2026
Cursor
Live apps in Cursor, where a cloud agent builds, hosts and shares an internal web app behind company sign-in
Anyone in a company can ask a Cursor cloud agent to build an internal tool with a database and sign-in, then share it with colleagues by link.
agents · prototyping · cursor.com
O
5 Nov 2026
Other
Dropbox AI agent that organises, renames and prepares files, with plan approval and undo for each conversation
A team can ask Dropbox to reorganise or prepare shared files, approve the plan and folder access first, and undo one conversation's changes afterwards.
agents · collaboration · dropbox.tech
O
5 Nov 2026
Other
Thinking Machines Lab Inkling 2 Small, an AI model for long coding tasks with image input, open weights promised after safety testing
A team can fine-tune and run a mid-sized AI model for long coding tasks through Tinker, the lab's training service, today. The lab says full weights under Apache 2.0 follow next week.
O
5 Nov 2026
Other
Kiro bots, named long-running AI agents in Kiro Crew that each keep their own memory and handle one recurring part of a team's work
A team can give one AI agent a standing job such as issue triage or release checks, with limited tools and a schedule. The agent runs its own sessions and brings back only results and open questions.
agents · kiro.dev
4 Nov 2026
Cloudflare
Analytics and per-agent access rules for personal AI agents visiting a website
A product team can see which personal AI agents use its site, which pages they reach and where they fail. It can then allow or block each agent separately.
4 Nov 2026
Vercel
Vercel Schedules in public beta, which lets AI agents and running apps create, change and cancel scheduled tasks
Engineering teams can let an AI agent set a one-time or recurring job when a user asks for one, without a separate database or a redeploy.
agents · vercel.com
O
4 Nov 2026
Other
Cognition SWE-2 Flash, a faster and cheaper AI model for coding agents in Devin, weights kept private
A team using Devin can run routine coding, code review and documentation work at about a fifth of the price and three times the speed of Cognition's larger AI model.
models · cognition.com
2 Nov 2026
Vercel
Vercel Security Review in public beta, an AI agent that checks every push for security flaws
Every push to a connected GitHub repository gets an AI security review of the changed code, checked against the team's own written security policies. Findings appear as pull request comments, and the agent can propose a fix.
critique & review · agents · vercel.com
30 Oct 2026
Figma
Figma Weave tools published inside a company, so a team can share its own AI image and video workflows
A design team can package an AI image or video workflow as a tool that colleagues run without rebuilding its steps. It appears in Figma's Tools tab for the whole organisation.
collaboration · figma.com
29 Oct 2026
Mistral
Mistral Large 4 open weights, Mistral's largest AI model for coding and security work, released for teams to run themselves
A team can download and host a large coding AI model on its own servers and fine-tune it. Companies selling the model as a hosted service need a separate licence.
models · mistral.ai
O
29 Oct 2026
Other
GitLab Orbit generally available, a searchable map of code, merge requests, pipelines, vulnerabilities and owners for AI agents
AI agents and engineers can look up who owns code, which pipelines failed and which vulnerabilities apply, within their existing permissions. GitLab reports fewer agent retries and lower token use in its own tests.
28 Oct 2026
GitHub
Shared Copilot Automations, AI agent workflows that a whole team can see, edit and keep in a repository
A team can build and edit AI agent workflows together in a repository. The workflows are billed to the organisation and keep running after their creator leaves.
agents · collaboration · github.blog
28 Oct 2026
Vercel
Vercel CI in beta, which runs only the checks a code change affects
Teams whose AI agents open many pull requests can run only the checks each change affects. Pull requests spend less time waiting for checks before review.
agents · vercel.com
28 Oct 2026
Google
Gemini 4 Carbon, a Google AI model for long software engineering and security tasks, first released to selected security teams
Security teams in Google's early programme can use an AI model that finds, tests and patches software flaws over tasks lasting days. Google promises wider developer access later.
models · blog.google
28 Oct 2026
GitHub
Copilot Memory generally available, and Copilot dreaming in preview, which proposes changes to agent instructions as pull requests
Copilot keeps a repository's conventions across sessions. It proposes changes to the team's checked-in agent instructions as pull requests for people to review.
agents · collaboration · github.blog
28 Oct 2026
GitHub
GitHub Agentic Workflows generally available, with automation written in Markdown and run by coding agents inside GitHub Actions
Engineering teams can write repository automation, such as issue triage or analysis of failed checks, in plain Markdown. Coding agents run it with read-only permissions by default.
agents · github.blog
O
28 Oct 2026
Other
NVIDIA Nemotron 3.5 Transcribe, an open weights speech AI model for transcription, speaker labels and translation
A product team can transcribe recordings of up to an hour in 40 languages, label up to eight speakers and translate, on its own hardware. Plain-language prompts adjust the output.
models · huggingface.co
28 Oct 2026
GitHub
Agent Merge in GitHub pull requests, where Copilot fixes failing checks and merge conflicts in the cloud
An engineer can ask Copilot to watch a pull request and fix failing workflows or merge conflicts as they appear. Combined with auto-merge, the pull request merges once every requirement passes.
agents · github.blog
28 Oct 2026
GitHub
Replica, an open-source prototyping tool from GitHub Design that builds prototypes from a team's own components and documentation
Designers set the guidance, and anyone on the team can describe a prototype in plain language that uses the product's real components. Teammates comment on shared prototypes in a self-hosted hub.
prototyping · design systems · github.blog
28 Oct 2026
GitHub
Copilot code review Max effort level in preview, and Copilot approvals that count toward required approvals generally available
A team can let Copilot's AI review approve a pull request and count toward required approvals, limited to chosen file paths. A slower review by several AI models covers risky changes.
critique & review · agents · github.blog
28 Oct 2026
GitHub
Issue artifacts in preview, which attach plans, research and diagrams made by AI agents to a GitHub issue
Plans, research and root cause analyses from AI agent sessions stay attached to the issue with version history. Teammates and later agents work from the same record.
agents · collaboration · github.blog
27 Oct 2026
Google
Genie 4, a Google DeepMind AI model that generates interactive simulated worlds with sound (research preview)
Product teams can steer a realistic, interactive simulated scene in plain language, with sound and dialogue. Access is limited to trusted testers for now, with Gemini API early access to follow.
models · deepmind.google
O
27 Oct 2026
Other
Liquid AI LFM3, open weights AI models of about 2 billion parameters for text, vision, audio and decisions
A team can fine-tune a small AI model on its own data with one graphics card and run it on a device or its own servers. Two models return calibrated decisions instead of text.
models · huggingface.co
26 Oct 2026
OpenAI
GPT-6.1 Terra, an AI model for computer use, visual reasoning and writing
Engineering teams can call a new OpenAI AI model through the API for AI agents that read screens, operate apps and write plain text.
26 Oct 2026
Vercel
Vercel Agent checks monitoring alerts and sends only critical ones to the paging system
An on-call engineer is paged only after Vercel's AI agent has checked logs and metrics and judged an alert critical, such as sustained errors on checkout requests.
agents · vercel.com
O
26 Oct 2026
Other
Replit for Business, a company workspace for AI agents with access limits and human approval rules
Replit says staff outside engineering can run AI agents on company systems, with the company setting what each agent may touch and when a person must approve.
agents · products · replit.com
26 Oct 2026
Figma
Figma Weave video and image AI models available to outside AI agents through the Figma MCP server
Designers can ask their own AI agent to generate images and video with the models in Figma Weave, without opening Weave, paid from Weave credits.
agents · prototyping · figma.com
26 Oct 2026
GitHub
Assisted permissions in GitHub Copilot, where an AI model approves clearly safe agent actions and asks about the rest
Engineers can let Copilot run commands, open files and visit links that an AI model judges clearly safe, and approve only the rest. Administrators can set or block this mode.
agents · github.blog
O
23 Oct 2026
Other
Recursive Flint, an open weights AI model for coding agents that runs on a laptop
Engineering teams can run a coding AI agent on their own laptops or workstations, keeping code on the device, working offline and paying no API fees.
models · agents · recursive.com
O
23 Oct 2026
Other
Kimi K3.1 Pro, an AI model for coding, 3D modelling and finance work, with open weights promised by 4 November
Teams can use a Moonshot AI model with a one-million-token context through the Kimi API now. Moonshot says it will publish the full weights by 4 November, so teams could later run it themselves.
models · kimi.ai
22 Oct 2026
Linear
Linear Agent reads and updates Figma designs through the Figma MCP server
A team can share a Figma frame with Linear's AI agent to turn it into implementation requirements. The agent can also draft a design proposal from an issue.
agents · handoff · linear.app
22 Oct 2026
Linear
Linear Loops started by Slack and GitHub events, and shared as workspace templates
A team can have a Slack message or a GitHub pull request label start an automated AI task, such as drafting a QA plan. Teams can share these tasks as templates.
agents · collaboration · linear.app
21 Oct 2026
Claude
Claude Duet, a personal AI agent app that works in connected apps and checks in before acting (research preview)
A person can hand an AI agent ongoing work in connected apps, such as a morning brief, scheduling and invoice reminders. It shows every message, payment, signature or deletion for approval first.
products · agents · claude.com
O
21 Oct 2026
Other
IBM Granite Speech 5.0, open weights speech AI models for transcription and translation
A product team can run transcription and translation in six languages on its own hardware under an open licence. It can supply up to 2,000 product and people names so the AI model spells them correctly.
models · huggingface.co
20 Oct 2026
Mistral
Voxtral 3 speech-to-text and text-to-speech AI models in more than 50 languages, with a voice agent toolkit
A product team can build a voice agent that listens, calls tools and speaks in over 50 languages. It can swap the AI model, voice or language without rewriting the agent.
models · agents · mistral.ai
19 Oct 2026
Notion
Notion Agent and Custom Agents that work on one task for more than 24 hours, with the Auto AI model router
A team can give a Notion agent a task that runs for more than a day, up from five minutes. Auto sends each task to a suitable AI model to cut its cost.
agents · notion.com
19 Oct 2026
Cloudflare
Cloudflare Clairic, an AI model that returns typed values from unstructured text (preview)
An engineering team can pull dates, numbers and text spans from documents as typed values with a confidence score. Missing values come back empty instead of invented.
16 Oct 2026
Cursor
Interactive canvases built by Cursor cloud agents, shared with teammates by link
A Cursor cloud agent can now build an interactive canvas as part of its work. A teammate can open it from a shared link to review the result.
agents · collaboration · cursor.com
15 Oct 2026
Meta
Muse Spark Open, an open weights AI model for coding and long agent tasks
A team can download a large coding and agent AI model and run it on its own hardware, so code and data stay in-house. The licence limits some commercial uses.
models · agents · research.meta.ai
O
15 Oct 2026
Other
NVIDIA Nemotron 3.5 Super VL and Bolt, open weights AI models for agents on local hardware
A team can customise a multimodal AI model on its own data, or run an agent model on a single workstation without sending work to the cloud.
models · agents · blogs.nvidia.com
15 Oct 2026
GitHub
AI Scan for pull requests, generally available
Reviewers see AI-found security issues, with a suggested fix, on pull requests in languages that standard code scanning does not cover. The findings advise and do not block merges.
critique & review · github.blog
15 Oct 2026
Figma
Specs: Markdown documents on the Figma canvas that hold a team's context for people and AI agents (open beta)
A team can write its decisions and specifications beside its designs and code layers. It can then send that context to a coding agent through Figma's MCP server.
handoff · agents · figma.com
15 Oct 2026
Figma
Design system guidelines for the Figma agent (open beta) and code sources from GitHub or npm (closed beta)
A design system team can attach written guidelines and the real component code to its libraries. The Figma agent then follows the team's patterns when it builds.
design systems · agents · figma.com
15 Oct 2026
Claude
Code Review for Claude Code, generally available
Team and Enterprise organisations get an AI review on every new pull request. Admins set each repository to a quick, standard or deep review.
critique & review · agents · claude.com
15 Oct 2026
Figma
Code layers: prototypes backed by working code on the Figma canvas (open beta)
Designers can bring code from GitHub, a local folder or a live website onto the canvas. They can ask an AI agent to edit it beside their designs.
prototyping · handoff · figma.com
O
13 Oct 2026
Other
Liquid AI LFM2.5-24B-A2B, an open weights AI model for agents that run on a device
A team can download, fine-tune and run a reasoning AI model locally, so an agent can work on private data without sending it to a cloud provider.
models · liquid.ai
12 Oct 2026
Meta
Muse Wonder and Muse Wonder Full, AI models for long agent tasks
Teams can build agents on Meta's Model API that carry context across conversations and check in with a person before important actions.
models · about.fb.com
9 Oct 2026
Microsoft
Microsoft-Decision-1, an AI model that scores a fixed set of answers instead of writing text
A team can have an agent ask a quick question with set answers, such as continue, stop or hand to a person. The AI model returns a probability for each answer.
9 Oct 2026
Cloudflare
Cloudflare Clef-omni, an open weights decision AI model for audio, video, images and text
A team can send a photo, recording or video with sound in one call and get scored yes-or-no answers, instead of chaining separate transcription and vision models.
O
8 Oct 2026
Other
JetBrains Mellum2.1, an open weights AI model for coding agents
A team can run a small coding AI model on its own servers that explores a repository, edits files and checks its own changes, keeping code private.
7 Oct 2026
GitHub
Local sandboxing for GitHub Copilot, generally available
Engineering teams can limit what Copilot's agent touches on a developer's machine, including files, network and credentials, through developer or organisation policy.
agents · github.blog
7 Oct 2026
Claude
Claude Haiku 5.5, a small, low-cost AI model
Teams can run high-volume work such as classification, summaries and coding subagents on a cheaper AI model. Prompts under 100,000 tokens cost about a tenth of Haiku 4.5's price.
models · anthropic.com
O
7 Oct 2026
Other
Atlassian Record for Agent: narrated Loom recordings turned into structured prompts for coding agents (open beta)
A designer or tester can narrate a screen recording of an interface problem. The coding agent receives it as a structured prompt.
agents · handoff · atlassian.com
O
7 Oct 2026
Other
Atlassian Code Context: code indexed across Bitbucket and GitHub for questions from people and agents (open beta)
Developers and coding agents can ask questions about code across connected repositories without opening each one.
agents · atlassian.com
O
7 Oct 2026
Other
Atlassian Jira: assign work items to third-party coding agents such as Devin, Factory and Warp
A team can assign a Jira work item to an outside coding agent and track it beside work assigned to people.
agents · atlassian.com
O
7 Oct 2026
Other
Liquid AI open d1: small open weights decision models for text, images and audio
A team can run a small AI model on its own devices that picks an answer from fixed options in under 50 milliseconds, without sending data to a vendor.
models · liquid.ai
6 Oct 2026
Claude
Claude for Google Workspace, editing Docs, Sheets and Slides (public beta)
Teams can have Claude edit shared Docs, Sheets and Slides in place. By default each change waits as an approval card for a person to accept.
agents · collaboration · claude.com
6 Oct 2026
Mistral
Mistral Large 4, a mixture-of-experts AI model in public preview, with weights promised later
Teams can test a large Mistral AI model for coding and agent work through its API now. Mistral says the weights will follow at the end of October.
models · mistral.ai
6 Oct 2026
Cursor
Remote control for local agents from the Cursor iOS app
Engineers can check on and reply to coding agents running on their own computer from a phone, while the work stays on that machine.
agents · cursor.com
6 Oct 2026
Figma
Figma agent, generally available, with library guidelines and a view of teammates' agents
Figma's AI agent on the design canvas is out of beta. Library owners can attach written guidelines that the agent follows when it uses their design system, and teammates can watch each other's agents work.
agents · design systems · figma.com
O
5 Oct 2026
Other
Reflection Beam, an open weights mixture-of-experts AI model for coding, reasoning and agent work
Teams can apply for early access now, and Reflection says the weights follow this month under the Apache 2.0 licence. A team could then run a coding and agent model on its own servers.
models · reflection.ai
2 Oct 2026
GitHub
Copilot code review requests through the API, with Balanced as the default review effort
A team can trigger an AI code review from its own scripts or pipelines and set how thorough each review is.
critique & review · github.blog
1 Oct 2026
GitHub
Computer use for desktop apps in Copilot CLI and the GitHub Copilot app (public preview)
Teams can ask Copilot to operate desktop applications that have no API, after a person approves each app. Organisations can switch the feature off centrally.
agents · github.blog
1 Oct 2026
Claude
Claude Code mods: small TypeScript functions that change how Claude Code behaves
Engineering teams can add their own rules to the coding agent, such as asking for confirmation before production changes or logging every edit. Administrators control which mods people install.
agents · claude.com
1 Oct 2026
GitHub
Dynamic workflows in Copilot CLI and the GitHub Copilot app (public preview)
A team can write a multi-step agent job, such as a release check or an incident investigation, as code with fixed stages. The job can pause for a person to review before it continues.
agents · github.blog
1 Oct 2026
Cloudflare
Clef and Clef-flash, open weights decision models, on Workers AI
Teams can run an open AI model that returns typed answers with probabilities, on Cloudflare or on their own machines. Cloudflare suggests it for routing tickets, escalating cases and handing work to a person.
models · agents · blog.cloudflare.com
O
1 Oct 2026
Other
AWS Strands Decider 2B, a small open weights decision model
Teams can run a small AI model on a laptop that picks among options or rates on a scale, with a confidence figure for each answer. Its makers say it cannot write text and is weaker than larger AI models on hard problems.
models · agents · strandsagents.com
1 Oct 2026
Microsoft
MAI-Voice-2.1, MAI-Voice-2.1-Flash and MAI-Transcribe-2-Streaming speech models
Teams building voice features can generate speech in 23 languages with one voice, and transcribe live audio in 60 languages as people speak.
models · microsoft.ai
30 Sep 2026
GitHub
HydraFusion multi-model routing in VS Code and the GitHub Copilot app (research preview)
Copilot can route a coding task to one AI model, escalate from a cheaper one, or have a second AI model review the first one's draft.
agents · critique & review · github.blog
30 Sep 2026
Google
Gemini 4 Argon
A frontier AI model that can return up to one million tokens in a single answer, so long software engineering tasks can finish in one run. Access starts with security defenders, with wider developer access promised later.
models · blog.google
O
30 Sep 2026
Other
Bilibili Index-Translate, open weights translation models for 150 languages
Teams localising a product can run translation on their own machines, with instructions for required terms and for keeping file formats intact. The largest of the three text models is a preview.
models · github.com
29 Sep 2026
OpenAI
GPT-6.1 Sol, an upgrade to GPT-6 Sol
OpenAI says the new model comes close to GPT-6 Astra on agentic coding, computer use and professional work, at a fifth of Astra's standard token prices. It is available in the API and on the paid ChatGPT plans.
models · openai.com
29 Sep 2026
OpenAI
Dots, always-on agents that work in ChatGPT, Slack and Teams
A dot is an agent that keeps working on a person's goals, connects to other apps through plugins, and can be reached in ChatGPT, Slack or Teams. OpenAI describes access and permission settings, and a review and approval step for actions.
products · agents · openai.com
29 Sep 2026
OpenAI
Codex in the cloud, with reusable development environments shared across a team
Developers can run Codex on a computer, from a phone or in the cloud. A team can share one development setup with approved settings and permissions, so tasks start quickly from any device.
agents · collaboration · openai.com
29 Sep 2026
OpenAI
Code review in the ChatGPT desktop app for GitHub and GitLab changes
Engineers can read summaries, explore changes and ask Codex about possible issues before commenting on a GitHub pull request or GitLab merge request. With automatic reviews, Codex takes a first pass in the cloud.
critique & review · agents · openai.com
29 Sep 2026
OpenAI
Codex Security Cloud scans repositories and checks each new commit
A team can have an AI agent scan its repositories, keep checking new commits for security problems and prepare fixes for engineers to review.
agents · critique & review · openai.com
29 Sep 2026
OpenAI
Agents API with Computer Use
Engineering teams can build AI agents through the OpenAI API that operate other software on screen and coordinate with other agents.
agents · openai.com
29 Sep 2026
OpenAI
ChatGPT Space, a shared team area with pages and agents
A team can keep its documents, shared pages and AI agents in one ChatGPT workspace instead of separate personal chats.
products · collaboration · openai.com
29 Sep 2026
OpenAI
Decisions API, typed answers from GPT-6 Luna (limited preview at launch, public beta since)
Teams can ask an AI model questions with fixed answers and receive a probability, a choice or a score. They can use the answers to classify content, route requests or choose an AI agent's next step.
models · agents · developers.openai.com
28 Sep 2026
Claude
Claude Sonnet 5.5, the second model of the Claude 5.5 family
Anthropic says the new model runs more than 30 percent faster than Sonnet 5 and costs up to 30 percent less for most work. It is strongest at well-scoped everyday tasks, such as fixing bugs and making documents, slides and spreadsheets.
models · anthropic.com
28 Sep 2026
xAI
Grok Team Bots: one Grok Bot shared by a whole team, with shared files, skills and plugins (public beta)
A team gives one Grok Bot, SpaceXAI's AI agent, the files, apps and skills it needs, and everyone works from that shared context. Each person's conversations with the Bot stay private. Team Bots are in public beta on the Teams and Enterprise plans.
agents · collaboration · x.ai
28 Sep 2026
Cloudflare
Cloudflare cf, a command-line tool for AI agents covering the whole Cloudflare API
An AI agent can deploy code, set security policies and manage domains on Cloudflare through one command-line tool with JSON output and built-in command search.
O
28 Sep 2026
Other
ElevenLabs Eleven v4 and Eleven v4 Turbo text-to-speech models
Teams building voice products can direct tone and sound effects with inline tags in more than 90 languages. The Turbo version is built for live voice agents.
models · elevenlabs.io
O
28 Sep 2026
Other
NVIDIA Open Agent Safety Platform, with the open-source OpenShell runtime
Engineering teams can run each AI agent in a sandbox where nothing is allowed by default and every allowed or denied action is logged. The rules are enforced outside the agent, so the agent cannot override them.
25 Sep 2026
Claude
Claude plugin directory submissions with automated validation and install analytics
A developer on a paid plan can package connectors and skills as a Claude plugin and submit it to a public directory. Submissions pass automated validation and safety scanning, and the developer sees installations by product and version. A skill written inside one team can become a plugin that others install.
agents · collaboration · claude.com
25 Sep 2026
GitHub
Copilot for Slack and Teams takes attachments and message links as context and checks for duplicate issues before creating one
An issue can be raised from a chat thread, with the thread linked back and duplicates checked, so a discussion in chat becomes tracked work without anyone retyping it.
collaboration · agents · github.blog
25 Sep 2026
GitHub
Agentic autofix reads and writes Copilot Memory when fixing security alerts
Security fixes applied by an agent become stored patterns that the organisation's other Copilot features reuse, so remediation conventions accumulate per repository rather than per person. The feature is opt-in and in public preview.
agents · critique & review · github.blog
25 Sep 2026
Microsoft
Copilot, now one app for personal and work accounts, rebuilt around Home, Code and Autopilot
Home puts chat and tasks handed to Copilot on one start page, and Code lets people who are not developers build apps by describing them. Autopilot, an agent that runs recurring work without being prompted, enters a private preview at the end of September. Home and Code reach Microsoft's early-access programme first.
products · agents · blogs.microsoft.com
25 Sep 2026
Microsoft
Copilot Managed Runtime (public preview): apps built in Copilot, Copilot Studio or Lovable run inside the company's Microsoft 365 environment
An app built by someone outside engineering can run under the company's own sign-in and data policies, and IT sees every such app in one list. A developer can take the app's code from Git and keep building on it, with no separate hosting project.
prototyping · handoff · microsoft.com
25 Sep 2026
Claude
Claude Code /doctor prompt-audit reviews CLAUDE.md files, skills, agents and commands for prompting patterns written for older models
A team can ask Claude Code to review its instruction files (CLAUDE.md, skills, agents and commands) for phrasing written for older AI models. The report opens with out-of-date file paths and commands, and with instruction files that contradict each other.
agents · critique & review · code.claude.com
24 Sep 2026
Linear
Linear coding agent adds adaptive model routing, environment secrets and open-weight model sessions
When a task goes from an issue to Linear's coding agent, simple work goes to a cheaper AI model and harder work to a reasoning model. The person assigning the task no longer picks the model, and administrators decide which models the workspace allows.
agents · linear.app
24 Sep 2026
GitHub
GitHub Copilot global default policy for unconfigured features (Business and Enterprise)
From 22 October, Copilot features that an enterprise has not configured follow one enterprise-wide default. This covers code review policies and MCP servers, the connections that give agents access to other tools. The options are to enable everything, to require approval, or to let each organisation choose. An administrator decides once whether teams get new agent features on arrival or after a review.
other · agents · github.blog
23 Sep 2026
Cursor
Rollouts and Security Reviewer bots watch a change from PR to production and review every PR for vulnerabilities
Before a change is merged, the agent writes a monitoring plan for a person to edit. After release, it compares live signals with a baseline and prepares a rollback for a person to approve. People edit the plan and approve the rollback instead of watching dashboards.
agents · critique & review · cursor.com
23 Sep 2026
GitHub
Copilot code review: personal and enterprise defaults for automatic review and review effort
Each person, or a whole company, now sets once when the AI code review runs and how thoroughly it checks. Review depth becomes a company setting with someone responsible for it, instead of a choice made in each repository.
critique & review · github.blog
23 Sep 2026
GitHub
Local sandboxing in the GitHub Copilot app (per-project file, network and credential limits)
A team can set, per project, which files, network access and credentials a local coding agent may use before it runs. The limits live in a configuration file instead of depending on a person watching the terminal. The setting is off by default.
agents · github.blog
23 Sep 2026
Lovable
Lovable: default design system per group
New projects in a group now start on that group's design system instead of the workspace default. A product team can keep its own design system without a central team approving each project.
design systems · docs.lovable.dev
23 Sep 2026
Claude
Claude Marketplace: one catalogue for partner connectors, plugins, agents and services
A team can choose connectors, plugins and agents from one catalogue built on the open MCP and Agent Skills standards. It can also spend its existing Anthropic commitment on partner products instead of buying each integration separately.
collaboration · agents · claude.com
23 Sep 2026
Meta
Meta Muse Charm, a pocket device for Meta's personal agent Muse, with connectors and computer use on Mac
At its Connect conference, Meta gave its personal agent Muse a pocket device and a place on its AI glasses. Muse can now work in Notion, GitHub and Mac apps, with permission.
products · meta.com
O
23 Sep 2026
Other
Stanford and NVIDIA CLM-8B, an open weights AI model that picks among set actions
Teams can run a decision-making AI model on their own hardware. It chooses among actions the team defines and writes no text.
models · github.com
22 Sep 2026
Vercel
Sandbox Drives (persistent storage for agent workspaces, public beta)
An agent's workspace, dependencies and stored memory persist between runs, so a long-running agent task no longer starts from a clean checkout each time.
agents · vercel.com
22 Sep 2026
Framer
Skills: saved instructions for the Framer agent covering design system, writing style and CMS workflow
A team's design and copy conventions become a reusable file that the site agent follows across tasks. The file is captured from a conversation after a page is refined and travels with the project when it is remixed.
agents · design systems · framer.com
22 Sep 2026
Claude
Claude Opus 5.5, the first model of the Claude 5.5 family
Anthropic says the new model does frontier-level work, including agentic coding, with far fewer tokens. It costs 40 percent less to run than Opus 5 and is more than 30 percent faster.
models · anthropic.com
22 Sep 2026
OpenAI
OpenAI GPT-6 Sol and GPT-6 Luna, faster and cheaper GPT-6 models
Sol is for demanding work and Luna for high-volume work. Both run in ChatGPT Work, Codex and the API, at lower prices and with higher usage limits.
models · openai.com
22 Sep 2026
Xiaomi
Xiaomi MiMo-V2.6 Pro and Flash, open weights
Xiaomi released two models that read text, images, audio and video, under an open licence. A team can host them itself and give an agent a long codebase to work through.
models · mimo.xiaomi.com
O
22 Sep 2026
Other
DigitalOcean Managed Agents (public preview)
Engineering teams can run AI agents in isolated cloud sessions that pause and resume, with access to tools through one managed gateway. They do not have to build that infrastructure themselves.
21 Sep 2026
xAI
xAI Grok 4.7 for coding and knowledge work
The model works longer on hard tasks and checks its own work more carefully, at the same price and speed as Grok 4.6. It is available in the xAI API and in Cursor.
models · x.ai
19 Sep 2026
Claude
Claude Code reads AGENTS.md as an alternative to CLAUDE.md for project instructions
A repository can keep one agent instructions file that several coding agents read, so the team maintains one set of rules rather than a copy per tool.
agents · code.claude.com
18 Sep 2026
GitHub
Copilot code review groups findings into open, resolved and previously missed, and writes commit messages for batch-accepted suggestions
On a re-review the reviewer sees only the agent findings still open, so the human check of agent-written code is shorter and the commit history records what was accepted.
critique & review · agents · github.blog
18 Sep 2026
v0
v0 installs private npm packages from shared environment variables
A prototype built in v0 can now use the team's own published component library instead of a generic one. The prototype and the production design system then draw on the same package.
design systems · prototyping · vercel.com
18 Sep 2026
Claude
Claude Code syncs skills and plugins enabled on claude.ai into terminal sessions
A skill enabled in the chat application is the same skill the engineer's terminal agent runs, so one list of skills serves the whole team without a separate installation step. An opt-out setting is available.
agents · collaboration · code.claude.com
18 Sep 2026
Warp
Warp Factories scoring agents grade every coding-agent run against team-written rubrics
The team writes a scoring rubric as a prompt, covering task compliance, efficiency, verbosity, code quality and custom checks. A scoring agent grades each run, and people read the failures to adjust skills rather than reading every pull request.
agents · critique & review · warp.dev
18 Sep 2026
Qwen
Alibaba Qwen3.8-Omni-Flash, a model built for agents that handle audio and video
The model reads text, images, audio and video and can call tools. An agent built on it can plan and carry out tasks such as editing, translating or summarising video.
models · qwen.ai
17 Sep 2026
Claude
Claude Code Projects redesigned as a persistent conversation that delegates parallel threads and keeps memory across sessions
A goal stated once becomes several agent threads across repositories with one shared memory, so the person steers and checks the work rather than briefing each session again.
agents · collaboration · claude.com
17 Sep 2026
Vercel
Vercel skills CLI installs agent skills from Notion databases
A team can keep its agent skills, the written instructions agents follow, in a Notion database that people outside engineering can edit. One command installs them into any agent, so the skills no longer have to live in a code repository.
agents · collaboration · vercel.com
16 Sep 2026
Claude
Cowork merged into Claude chat, with Docs, Slides and Design inline and an approval-or-autonomous toggle
One conversation now produces documents, slides and designs, and the team decides per task whether Claude confirms before acting or runs unattended and reports back.
collaboration · other · claude.com
16 Sep 2026
GitHub
Copilot budget increase requests
When a member uses up an AI credit allowance, they can ask for more inside the product. An organisation owner approves the request in settings, so running out of credit no longer stops the work outright.
collaboration · other · github.blog
15 Sep 2026
Notion
Notion 3.7: shared skills library with SKILL.md export and sub-agents
A team writes instructions for AI agents once, as a skill file (SKILL.md), and keeps them in one library. The same file can be exported to the coding tools Claude Code, Codex and Cursor, so every tool follows one definition.
agents · collaboration · notion.com
15 Sep 2026
TypeSafe
Jev, an AI model that returns typed decisions instead of text (early access)
Teams can have software ask an AI model a fixed question and get a choice, a score or a probability back. That suits routing, ranking and approval steps that code must act on.
models · typesafe.ai
14 Sep 2026
Linear
Loops respond to project changes, edit docs and post to Slack
Linear's recurring agent workflows, called Loops, now react when an initiative, project or cycle changes. They can update Linear documents and post to Slack, so nobody has to rewrite the status by hand when scope moves.
agents · collaboration · linear.app
14 Sep 2026
GitHub
Copilot auto model selection gets cost and quality tiers
Each engineer selects efficiency, balance or intelligence per tool, so model cost per task becomes an individual setting rather than a team default.
agents · other · github.blog