Feed, Agentic Levels

56 updates

Jul 2026Claude

Tool Update

A new way to reflect with Claude

Your monthly recap is a new feature at Settings > Reflect that shows you the topics you spent time on, your most active day and peak hour, and observations about how you work with Claude. It's in beta on Free, Pro, and Max plans on the web and Claude Desktop, and requires memory to be on. Alongside it, Settings > Time and focus lets you set optional break reminders and quiet hours. For more information, refer to See your monthly recap and Set break reminders and quiet hours .

Jul 2026Claude 5

Model Release

Claude Fable 5 returns after export controls

U.S. restrictions on Fable 5 were lifted after Anthropic added tighter safeguards for cyber misuse. The practical lesson is reliability: access to frontier models can change for policy reasons, not only product reasons.

Jun 2026Claude 5

Model Release

Claude Sonnet 5

Anthropic released Sonnet 5 as the default everyday Claude model, with stronger agentic work, coding, browsing, and tool use. The practical shift is that the model most people touch first is now built for delegated work, not only chat.

Jun 2026Gemini 3.5

AI News

Gemini 3.5 Flash gets computer use

Google added public-preview Computer Use to Gemini 3.5 Flash, with support for browser, mobile, and desktop environments plus stronger prompt-injection controls. Gemini can now act inside software, not only answer questions about it.

Jun 2026Codex

Tool Update

Record & Replay in Codex

Codex can now turn a recorded Mac workflow into a reusable skill for eligible users. For Levels 3-5, this is the clearest bridge from showing a workflow once to running it again with an agent.

Jun 2026Chatbot

Tool Update

Scheduled Tasks Get a Real Home in ChatGPT

Scheduled Tasks now have a sidebar page, pause and edit controls, broader time windows, and monitoring tasks that can check the web or connected apps before notifying you. For Levels 1-3, this moves recurring reminders and briefings into a visible workflow.

Jun 2026Claude

Tool Update

Access to Claude Fable 5 and Claude Mythos 5 suspended

We've suspended access to Claude Fable 5 and Claude Mythos 5. We aim to return access as quickly as we can. See our statement Edit Claude’s drafts in place When Claude writes a draft in chat or Claude Cowork (a report, a plan, a brief), it opens right beside your chat. Highlight the part you want changed, type the change, and Claude makes the edit right where you marked it. No switching apps, no re-describing paragraphs in chat. For more information, see Work with artifacts .

Jun 2026Claude

Tool Update

Introducing the AI Fluency Score

Built on Anthropic's published AI fluency research. Claude is building a scorecard that reads your Claude transcripts. We built one that works wherever you work, scores all four skills including the one theirs skips, and links every gap to a fix. Run our prompt in your own AI, free and with no account, and one conversation hands back your agentic Level, your Fluency Score across the 4 D's (Delegation, Description, Discernment, and Diligence, the habit a transcript can't see), and the one blind spot you didn't know you had. Get yours at /fluency.

Jun 2026Codex

Tool Update

Codex Sites and Role Plugins

Codex added Sites, which turns a task into a shareable web app or dashboard from a plain-language description, plus six role plugins spanning data, sales, creative, design, and finance that pre-connect Codex to the tools each job uses. It is rolling out first to business and enterprise. For Levels 3-4, this is the easiest way yet for non-coders to try agent-driven work.

Jun 2026Codex

AI News

Codex becomes a knowledge-work tool

OpenAI published new Codex usage data showing it is no longer only a coding tool: more than 5 million weekly active users, with knowledge workers now about 20% of usage. Reports, spreadsheets, presentations, research, data analysis, and workflow automation are becoming normal Codex work.

May 2026Claude

Tool Update

Claude Opus 4.8 launch

We’ve upgraded Claude Opus to a new version. Claude Opus 4.8 shows improvements over Opus 4.7 in coding, agentic skills, reasoning, and practical knowledge work tasks. For more information, see our blog post: Introducing Claude Opus 4.8 . Enterprise plans can manage connector access with custom roles We added connector permissions to extend the existing custom roles framework and allow administrators to control which connectors, and which individual tools on those connectors, are available to each custom role. For more information, see Manage custom roles on Enterprise plans .

May 2026Claude 4

Model Release

Claude Opus 4.8

Anthropic released Opus 4.8 with stronger agentic work, coding, and practical task performance. The bigger shift is the surface around it: effort control in Claude and Cowork, plus dynamic workflows in Claude Code for longer supervised work.

May 2026Gemini 3.5

Model Release

Gemini 3.5 Flash

Google released Gemini 3.5 Flash as a fast frontier model for agentic workflows, coding, and multimodal work. It is available in the Gemini app, Search AI Mode, Antigravity, Gemini API, Android Studio, and Google's enterprise products.

May 2026Codex

Tool Update

Codex in the ChatGPT App, on Every Plan

OpenAI brought Codex into the ChatGPT mobile and desktop apps as a remote control: you can watch a running task, read its diffs (the exact before-and-after changes it proposes), and approve or deny steps from your phone. The work itself runs in OpenAI's cloud, or on your own computer if that is where you started it. Access reached every plan, including Free and Go, which lowers the bar to trying Codex without living in a terminal.

May 2026Claude

Tool Update

Google Labs Releases design.md

Google Labs published design.md, an open-source pipeline that analyzes any public URL and outputs a structured DESIGN.md file with color tokens, fonts, spacing values, and component patterns. For Levels 2-4, this turns the reference-site-to-design-brief workflow into a repeatable pipeline instead of a manual analysis session.

Apr 2026GPT-5

Model Release

GPT-5.5

With GPT-5.5, most people no longer have to pick a model at all. Its default mode, GPT-5.5 Instant, auto-routes between quick and careful answers for you. Under the hood it is OpenAI's first fully retrained base model in a while, built for multi-step work with a 1M-token context window, so it follows long instructions and does what you asked instead of guessing.

Apr 2026GPT Image

Model Release

GPT Image 2

OpenAI's image generator got a big upgrade. The model is called GPT Image 2, and in the app it shows up as ChatGPT Images 2.0 (it replaced DALL-E 3). It now puts readable text inside images, edits one part of a picture without redoing the whole thing, and keeps a character consistent across a set. The basic image mode is free; paid plans add higher resolution and a thinking mode that plans more carefully before it draws.

Apr 2026Claude 4

Model Release

Claude Opus 4.7

Anthropic released Opus 4.7, built for long-running agentic work, meaning AI that carries out multi-step tasks on its own. It adds stronger reasoning and a 1M-token context window, so if you hand off a big task, this is the model that holds the most in its head at once.

Mar 2026Codex

Tool Update

Official Codex Plugin for Claude Code

OpenAI shipped an official plugin that installs inside Claude Code and lets it call Codex, either for a second-opinion code review or to hand off a task. For Levels 3-5, it is the cleanest way to run the two leading coding agents together and let each check the other's work.

Mar 2026

AI News

The standalone Sora app shuts down

OpenAI closed the separate Sora video app. The Sora 2 model itself did not go away: it now lives inside ChatGPT on paid plans. If a guide tells you to download the Sora app, it is out of date, which is a good reminder to check the date on any AI how-to.

Mar 2026Claude

Tool Update

Computer use research preview in Cowork and Claude Code + Dispatch improvements

Anthropic brought computer use into Cowork and paired it with stronger Dispatch flows. For Levels 4-6, this is the moment Claude shifts from advising you to actually driving local work on your behalf when the task is scoped well.

Mar 2026Claude

Tool Update

Computer Use in Cowork and Claude Code plus Dispatch Improvements

Anthropic added computer use to Cowork and Claude Code and paired it with stronger Dispatch flows. For Levels 4-6, this is the shift from Claude advising you to Claude operating a machine on your behalf when the task is scoped well.

Mar 2026Agents

AI News

Computer Use Goes Mainstream

Anthropic brought computer use into Cowork and Claude Code, letting the assistant actually operate a machine on your behalf when a task is scoped well. The shift is from advice to action.

Mar 2026Claude

Tool Update

Control Cowork from your phone with a persistent thread (research preview for Pro/Max)

Anthropic extended Cowork with persistent remote control from mobile and Dispatch. For Levels 3-5, this matters because long-running delegated work no longer has to stay tied to one desktop session.

Mar 2026Claude

Tool Update

Control Cowork from Your Phone with a Persistent Thread

Anthropic added a persistent Cowork thread across desktop and mobile for Pro and Max plans. For Levels 3-5, this matters because delegated work can stay alive across surfaces instead of being trapped inside one session or device.

Mar 2026Codex

Tool Update

Codex App on Windows

OpenAI added a Windows desktop surface for Codex with isolated worktrees, reviewable diffs, and handoff between the app, CLI, and IDE. This matters most for Levels 3-6 where you want a durable review surface without losing repo context.

Mar 2026Claude

Tool Update

Memory for Free Users

Anthropic expanded memory from chat history to all Claude users, including free plans. For Levels 1-3, this lowers the setup burden because repeated work can build on prior context instead of restarting from zero each session.

Feb 2026Claude

Tool Update

Scheduled Tasks in Cowork

Cowork added recurring and on-demand scheduled tasks, plus a new Customize area that groups skills, plugins, and connectors together. For Levels 3-5, this matters because repeated work can move from remembered prompts into durable operating routines.

Feb 2026Claude

Tool Update

Cowork Plugins and Admin Controls

Anthropic added a Cowork plugin marketplace and org-level controls. For Levels 4-6, this is the point where reusable workflows become easier to distribute safely across a team instead of living as one-off personal setups.

Feb 2026Claude 4

Model Release

Claude Sonnet 4.6

Sonnet 4.6 landed as the everyday workhorse: faster than Opus, cheaper to run, and good enough for most real tasks. For day-to-day use, this is usually the model you actually want.

Feb 2026Codex

Tool Update

Harness Engineering Becomes Explicit

OpenAI published a detailed harness-engineering write-up showing how Codex benefits from repo instructions, trusted tools, and real execution environments. For Levels 5-8, the product lesson is to stop treating the agent as a chatbox and start treating the environment as part of the system.

Feb 2026Codex

Tool Update

GPT-5.3-Codex

GPT-5.3-Codex was introduced as a faster, more capable agentic coding model designed for long-running work with research, tool use, and complex execution. This is primarily a Level 5-8 update because better long-horizon behavior changes what work you can safely delegate.

Jan 2026Anti-Gravity

Tool Update

Trustable AI Case Study Using Antigravity

Google's trustable AI blueprint showed Antigravity in a real workflow where agentic development compressed a project cycle from roughly three months to two weeks. This is best read as a Level 6-8 operating example: autonomy rises only when oversight, artifacts, and trusted data are built in.

Jan 2026Claude

Tool Update

Cowork Expands to Pro Plans

Cowork moved beyond Max into Pro plans on Claude Desktop. For Levels 2-4, wider access matters because the chat-to-agent jump becomes practical for more users instead of staying an advanced-tier experiment.

Jan 2026Codex

Tool Update

GPT-5.2-Codex in the API

OpenAI added GPT-5.2-Codex to the Responses API. For Levels 6-8, this matters because Codex stops being just a user-facing coding tool and becomes something you can embed inside your own agent pipelines.

Jan 2026Claude

Tool Update

Cowork Research Preview on Claude Desktop for Max Plans

Anthropic brought Cowork to Claude Desktop for Max users, with local VM execution, local file access, and MCP integrations. For Levels 2-4, this is the clearest bridge from plain chat into agentic local work without jumping straight into the CLI.

Dec 2025Codex

Tool Update

GPT-5.2-Codex Launch

GPT-5.2-Codex improved code generation, code review, and agentic software engineering benchmarks. Level 5+ engineers should care because model quality changes how much scaffolding and review overhead is required for the same task.

Dec 2025Claude

Tool Update

Skills for Organizations, Partners, and the Ecosystem

Anthropic expanded skills with organization-wide management, a partner directory, and an open Agent Skills standard. For Levels 5-7, this is the moment skills become shared capability infrastructure instead of private prompt files.

Dec 2025Claude

Tool Update

Claude in Chrome Broadens Browser-Native Workflows

Claude in Chrome expanded to all paid plans with Claude Code integration, console and network inspection, recorded browser workflows, and org-level controls. This is a Level 3-6 shift because browser verification becomes a first-class part of the Claude workflow instead of a separate manual loop.

Dec 2025Skills

AI News

Agent Skills Become an Open Standard

Anthropic opened the Agent Skills format so Skills, which are saved instruction sets you can reuse, can be shared across tools and teams instead of being locked to one app. The ones you write become building blocks anyone on your team can pick up.

Dec 2025GPT-5

Model Release

GPT-5.2

OpenAI released GPT-5.2 as a general-purpose upgrade that follows instructions more closely and stays reliable on longer tasks. In practice that means it does what you actually asked, not what it guessed you meant.

Nov 2025Anti-Gravity

Tool Update

Google Antigravity Public Preview

Google launched Antigravity in public preview as an agentic development platform with an Editor View and a Manager Surface. This is the key Level 2-5 milestone because the product is designed around planning, execution, and review rather than autocomplete.

Nov 2025Anti-Gravity

Tool Update

Artifacts Become the Review Surface

Google framed Artifacts as the way Antigravity agents communicate progress through screenshots, recordings, and other review material. This is the main Level 4-7 lesson in the launch: demand inspectable proof, not just a fluent summary.

Nov 2025Anti-Gravity

Tool Update

Manager Surface for Multi-Agent Coordination

Antigravity's Manager Surface introduced an explicit place to launch and coordinate autonomous agents across the editor, terminal, and browser. This is most relevant for Levels 5-7, where decomposition and oversight become part of the job.

Nov 2025Anti-Gravity

Tool Update

Cross-Platform and Multi-Model by Default

Google shipped Antigravity across macOS, Linux, and Windows and positioned it to work with Gemini 3 Pro, Claude Sonnet 4.5, and GPT-OSS models. For Levels 2-4, the implication is that your workflow can be tool-first rather than model-first.

Nov 2025Gemini 3

Model Release

Gemini 3 Pro

Google shipped Gemini 3 Pro alongside Antigravity, with stronger coding and reasoning and a very large context window. If you use Google's new agentic tools, this is the model running underneath them.

Nov 2025MCP

AI News

MCP (Model Context Protocol) Becomes the Default Way to Connect Tools

The Model Context Protocol kept spreading across major AI products as the common way to plug models into outside tools and data. Learning MCP once now pays off across many apps.

Sep 2025Gemini 2.5

Model Release

Gemini 2.5 Flash

Google's fast, low-cost model for high-volume tasks where speed matters more than peak reasoning. A good fit for summarizing, drafting, and quick back-and-forth work.

Sep 2025Codex

Tool Update

Codex Upgrades Across Terminal, IDE, Web, and Phone

OpenAI expanded Codex into a fuller cross-surface workflow spanning terminal, IDE, web, and phone, while also sharpening task steering and agent persistence. For Levels 3-6, this changes how often you can keep a task alive across contexts instead of restarting from scratch.

Jun 2025Claude

Tool Update

Remote MCP Support and OAuth Resources

Anthropic added remote MCP server support plus OAuth resource handling. This is a real Level 4-6 unlock because it expands Claude Code from local repo work into connected systems without inventing bespoke integrations every time.

Jun 2025Claude

Tool Update

Claude Code SDK in TypeScript and Python

Anthropic added SDK support for building agentic applications and workflows on top of Claude Code. This is a Level 7-8 update because it turns the CLI operating model into something you can embed and orchestrate programmatically.

Jun 2025Claude

Tool Update

Claude Code on Pro and Max Plans

Anthropic expanded Claude Code access to Pro and Max. For Levels 1-3 this matters more than it sounds, because wider availability changes whether the CLI becomes a real daily surface instead of a niche advanced tool.

Jun 2025Codex

Tool Update

Codex for Plus Users and Internet Access During Tasks

OpenAI expanded Codex access to Plus and added internet access for task execution. For Levels 2-5, this is the point where Codex starts to become viable for real research-backed work instead of closed-box code generation.

May 2025Claude

Tool Update

Claude Code General Availability

Anthropic moved Claude Code from research preview to general availability and introduced an extensible CLI workflow with GitHub Actions support. This is the foundational Level 2-5 shift where Claude Code stops being an experiment and becomes part of real engineering workflow design.

May 2025Codex

Tool Update

Codex Research Preview

OpenAI introduced Codex as a software engineering agent that can run tasks in a cloud sandbox, read and edit files, and cite terminal logs and test output. This is the core Level 3-5 reference point for the product's original operating model.

Feb 2025Claude

Tool Update

Claude Code Research Preview

Anthropic first announced Claude Code as a terminal-native way to delegate substantial engineering work directly inside the repo. It remains a useful reference point for Levels 2-4 because it framed the core pattern: read the code, run commands, verify, then tighten the system.