Anthropic has been shipping at an insane pace that's honestly hard for most power users to even keep up with. A new release pretty much every day, and a major release roughly every two weeks since January. New models, new tools, new integrations, and entire new product categories. If you blinked at the wrong time or took a few weeks off, you probably missed more than you'd like to admit. And Claude impacts and changes how you work. Period.
This is the everything guide. Every major feature now live on Claude you can use as of March 23, 2026. How to actually set each one up, when to use what, and the best practices that separate people who think these tools are "super cool" from people who've actually rebuilt how they work around them.
You're gonna want to bookmark this and come back to it. Share it with your team or friends. This is the reference document I wish existed when I started.
The Models: What You're Working With
There's three model tiers in the Claude 4.6 family.
Here's what each one does and when to use it:
Claude Opus 4.6 is the ceiling. Launched February 5, 2026 with a 1 million token context window (more on the pricing change below). Scores 78.3% on MRCR v2 at 1M tokens, highest among frontier models at that length. Dominates benchmarks in legal, financial, and coding tasks. Anthropic reports a 14.5 hour task completion window, longest among frontier models. $5/$25 per million tokens on the API. 128K max output tokens. Supports adaptive thinking with a new "max" effort level for peak capability.
When to use Opus: complex analysis across large contexts, codebase refactoring, deep research, high stakes deliverables, serious content production, anything where quality matters more than cost.
When NOT to use Opus: anything you'll run more than a few times a day. At $5/$25 per million tokens, a heavy Opus workflow can burn $50-100/day. Default to Sonnet and escalate to Opus only when Sonnet's output isn't good enough.
Claude Sonnet 4.6 launched February 17, just 12 days later. This is the one most people should default to. 1M token context window (GA since March 13). Upgrades across coding, computer use, long context reasoning, agent planning, knowledge work, and design. Early testers preferred it over Sonnet 4.5 roughly 70% of the time. Users chose it over the previous flagship Opus 4.5 in 59% of cases. Default for Free and Pro plans on claude.ai. $3/$15 per million tokens. 64K max output. 30-50% faster than Sonnet 4.5.
When to use Sonnet: everyday work, quick drafts, standard coding tasks, agent workflows where you want speed without sacrificing too much intelligence. It matches Opus on many office tasks (Anthropic's OfficeQA benchmark shows it actually edging out Opus on some) at roughly 40% lower cost.
Claude Haiku 4.5 is the fast, cheap option for API developers running high volume pipelines and for subagent model assignments where you want a low cost read-only worker. One important caveat: Haiku has zero prompt injection protection. If you're using it in agentic setups where it's processing untrusted input, read the docs carefully and understand the risks before deploying it.
The 1M Context Window at Standard Pricing
Previously, requests over 200K tokens were billed at a premium ($10/$37.50 per million for Opus). As of March 13, that surcharge is completely gone. A 900K token request now costs the same per token rate as a 9K one. No multiplier, no fine print, no beta header required.
That's roughly 750,000 words of context. Entire codebases. Full legal contracts. Massive datasets. Months of documentation. All held in working memory simultaneously. Media limits also jumped to 600 images or PDF pages per request, up 6x from 100. Available on Claude Platform, Microsoft Foundry, and Google Cloud Vertex AI.
Teams that previously had to chunk context, build summarization pipelines, and manage rolling windows can now just load everything. One company reported that raising their context from 200K to 500K actually reduced total token usage because the model spent less time re-reading and re-processing earlier information.
Which Claude Mode to Use When
Claude has four modes and most people only know about one. Here's the quick breakdown:
Chat is the browser/mobile interface you probably already know. Good for quick questions, brainstorming, writing drafts. Every conversation starts blank. You're always driving.
Cowork is the desktop agent. It reads and writes to your actual files, executes multi-step tasks autonomously, and delivers finished work to your folder. Use it when you want to delegate work, not have a conversation.
Code is the developer tool. Runs in your terminal, sees your codebase, writes code, executes commands, manages git. If you write code, this is where the deepest leverage lives.
Projects are saved workspaces where you upload files and instructions once. Every new chat in that project starts with full context. Use them for recurring work like weekly newsletters or client deliverables where the context doesn't change much between sessions.
Quick rule: Chat for quick questions. Cowork for real work on your files. Code for development. Projects for recurring work with stable context.
Memory and Personalization
As of March 2, 2026, Claude's memory from chat history is available to all users including free tier. Claude remembers relevant context from your conversations and generates a memory summary that carries across sessions. You can view, edit, and delete memories in Settings > Capabilities. You can also import and export your entire memory, which is useful for backing up before you experiment with changes or for transferring context to a new account. Incognito chats let you exclude specific conversations from memory.
The move here: go to Settings > Memory right now and read what Claude already remembers about you. Edit anything that's wrong or outdated. Add context it should know. The more accurate your memory profile, the less you repeat yourself across sessions.
Cowork sessions don't carry memory between sessions. Your context files are the workaround (more on this in Limitations below).
Claude Cowork: The Knowledge Worker's Operating System
Cowork changed the game. It launched January 12 as a research preview for Claude Max subscribers on macOS. Expanded to Pro on January 16, Team and Enterprise on January 23. Windows followed. Investors freaked out. SaaS stocks dropped hundreds of billions in market cap within days. Wall Street saw what was coming.
Stop thinking about this like a chat interface. Cowork is task delegation. You describe what "done" looks like. Claude makes a plan, breaks it into subtasks, executes autonomously on your actual computer, and delivers finished files to your folder. You walk away. You come back to completed work.
Anthropic built Cowork in roughly 10 days using only Claude Code.
The Setup That Changes Everything
The people struggling with Cowork are writing long, detailed prompts for every task and getting inconsistent results. The people who figured it out spent an afternoon building their context setup (context files, global instructions, folder structure) and now write 10 word prompts that produce client ready deliverables.
ChatGPT trained you to write better prompts. Cowork rewards better files. One is a skill that gets less useful over time as models improve. The other compounds.
Step 1: Build your workspace folder.
Create a dedicated folder for Cowork on your computer. Don't point it at your entire Documents directory. If something goes wrong (it might), you want the damage contained. Cowork has real read/write access to whatever folder you share.
This keeps things organized and limits what Claude can see. Every guide from every power user converges on this same basic structure. The folder names don't matter. The separation does.
Step 2: Create your context files.
This is the step that kills the "generic AI output" problem. Create three markdown files in your CONTEXT folder:
about-me.md is who you are, what you do, your current priorities. Not your resume. What you actually work on day to day, who you serve, what matters right now. Include one or two examples of work you're proud of.
brand-voice.md is how you communicate. Tone descriptors, words you use, words you'd never use, formatting preferences, and 2-3 paragraphs of your actual writing as a reference sample. This is what separates "generic AI output" from "this actually sounds like me."
working-preferences.md is how Claude should behave. Ask questions before starting. Show a plan before executing. Never delete files without confirmation. Default output formats. Quality standards. Things to avoid.
These three files kill the cold start problem overnight. Without them, every session begins from zero. With them, Claude starts every session already knowing your voice, your standards, and your preferences.
The key insight most people miss: these files compound. Refine them weekly. Every time Claude produces something you don't like, ask yourself whether it's a prompt problem or a context problem. Nine times out of ten, it's context. Add one line to one file. Permanent fix.
I built my context folder in about 45 minutes. Three .md files covering who I am, what I build, and how Claude should operate. The next session I gave it a 10 word prompt for a project brief and the output matched my standards on the first pass. Before that, every session started with me re-explaining my whole setup from scratch.
Step 3: Set Global Instructions.
Go to Settings > Cowork > Edit Global Instructions.
Global Instructions load before everything else. Before your files. Before your prompt. Before Claude even looks at your folder. They're the baseline behavior that applies to every single session.
Here's a starting template:
This means even your laziest, most rushed prompt still produces calibrated output. Claude always knows who you are. Always reads the right files first. Always asks before guessing. Your prompt just handles the specific task.
Step 4: Learn AskUserQuestion.
This feature flips the entire interaction model. Instead of you engineering the perfect prompt, Claude engineers the perfect questions.
When you add "Start by using AskUserQuestion" to any prompt, Cowork generates an interactive form. Multiple choice questions. Clickable options. Specific alternatives. A structured prompt that helps you think through what you actually want before Claude starts executing.
You stop writing long, carefully engineered prompts from scratch. You let Claude figure out what it needs to know. And if the first round of questions doesn't get you aligned, you say so. It generates a new set and you iterate.
The universal prompt template that works for nearly everything:
That's it. That template plus your context folder handles 80% of use cases. The workflow is always the same. Only the context changes.
Every Feature That's Shipped for Cowork
Connectors
Shipped February 24. Claude Cowork connects to Google Drive, Gmail, DocuSign, FactSet, Google Calendar, Slack, and more. These shipped alongside the enterprise update. These aren't shallow integrations. Claude can autonomously navigate your Drive, pull documents, synthesize data across sources, draft emails based on what it finds, and flag issues in contracts. Once connected, Claude can reference live data from that tool in every session. No copy pasting, no screenshots, no downloads.
Go to Settings > Connectors. Browse the directory (50+ integrations). Click "Add." Authenticate. Done. You do this once. Connectors are free on all plans including the free tier (since Feb 24) and probably the most underused feature in Cowork.
After connecting Slack, try: "Search my Slack messages from the last 7 days and give me a summary of anything I need to follow up on. Organize by urgency."
After connecting Google Drive: "Find the most recent document about [project name] in my Drive. Read it and tell me the three most important things I need to know."
After connecting Google Calendar: "Look at my calendar for this week, identify any meeting conflicts, and draft a reschedule email for the lowest priority one."
Plugins and Marketplace
Shipped February 24. Pre-built plugins for specific job functions. Each plugin bundles skills, slash commands, and connectors into role specific packages. Anthropic shipped official plugins across Sales, Marketing, Legal, Finance, Data Analysis, Product Management, Customer Support, Enterprise Search, Engineering, HR, Operations, Design, Brand Voice, Bio Research, and more.
Install from the Customize menu in the left sidebar > Browse plugins. Click Install. Type / in any chat to see available slash commands.
The plugins that most people should install first:
Productivity handles tasks, calendars, and daily workflows. Type /productivity:start and Claude reviews your day.
Data Analysis lets you drop a CSV, type /data:explore, and Claude summarizes columns, flags anomalies, suggests analyses, and writes SQL in plain English.
Then install one role specific plugin that matches your work. /marketing:draft-content pulls your brand voice and drafts content. /sales:call-prep researches accounts and preps talking points. /legal:review reads contracts and flags risky clauses.
Teams can build private marketplaces to distribute custom plugins across their organization with admin controls for Team and Enterprise plans. Build once. Deploy to hundreds. Anthropic also launched a public Marketplace and Ambassadors program for community built plugins, so the ecosystem is growing beyond first party offerings.
You can also customize each plugin for your specific company. After installing, tell Claude to "customize the [plugin-name] plugin for me based on my company." Claude asks you questions about your workflows, terminology, and preferences, then remembers the answers for every future session with that plugin. This turns a generic Sales plugin into one that knows your ICP, your pricing tiers, and your outreach style.
Scheduled Tasks
Shipped February 25. Tell Claude once to do something on a recurring basis. Morning email summaries. Friday metrics pulls. Weekly competitive intel briefings. It runs automatically as long as your computer is awake and Claude Desktop is open.
Real example that multiple power users have set up:
You wake up Monday to a briefing doc ready to read. Combined with connectors, scheduled tasks become genuinely autonomous. "Every Monday, pull all unread Slack messages from #product-feedback, categorize them by theme, and create a summary in Google Drive." The scheduled task runs. The connector pulls live data. Claude processes it. The output appears in your folder.
I run 3-4 scheduled tasks throughout the day. Morning AI news brief saved to my content folder. Midday competitor check across X and product launches vs. what I'm building. Afternoon community digest from Discord and Telegram. Evening content performance recap. Each one saves 20-30 minutes of manual work. That's close to two hours a day I got back without thinking about it.
This also shipped alongside a new Customize section in Claude Desktop that groups skills, plugins, and connectors in one place.
Dispatch
Shipped March 17. The phone to desktop bridge. Available to Pro and Max subscribers. You access a persistent agent thread via Claude Desktop or Claude for iOS/Android to manage tasks in Cowork from anywhere.
The setup is simple: Open Claude Desktop > Cowork > click Dispatch in the left sidebar. Toggle "Keep awake" on (without it, Dispatch stops when your computer sleeps). Then open the Claude mobile app > tap Dispatch in the sidebar.
One continuous conversation that syncs across devices. You're on the train and you tell Claude to compile a report from three spreadsheets on your desktop. By the time you get to the office, it's done. Stack multiple tasks in one Dispatch message and Claude works through them sequentially while you're away.
Key detail from the Product Compass guide that most people miss: the Dispatch orchestrator doesn't read your CLAUDE.md. It formulates task prompts based on assumptions. Sub-tasks do read it, but the prompt already carries imprecise framing. Fix: include "read CLAUDE.md" in your Dispatch message.
You can't add connectors from mobile. Set up Gmail, Slack, Notion on desktop before you leave. Dispatch inherits them all. You also can't attach files from mobile yet. Workaround: email them to yourself and tell Dispatch to pull them via the Gmail connector.
Projects
Shipped March 20. Organize related tasks into persistent workspaces with their own files, links, instructions, and memory. Import existing folders or start fresh. This means you can have a project for "Q1 Financial Reports" and another for "Product Launch Materials" and Claude remembers the context for each one.
Projects turn Cowork from a one-off agent session into a persistent workspace. That matters for research heavy work where you keep losing context across chats and have to restate your goals over and over.
Computer Use
Shipped March 23 (today). Research preview, macOS only, for Claude Pro and Max subscribers. Available in both Cowork and Claude Code. Claude can now point, click, and navigate your screen. It opens apps, uses the browser, fills in forms, and operates any tool on your computer. When there's a direct connector (like for Slack or Google Calendar), Claude uses that first. When there isn't one, it uses your mouse and keyboard.
Claude asks for permission before taking significant actions. Anthropic still recommends not using this for sensitive information as a precaution. The biggest risk to be aware of: prompt injection through screen content. If Claude navigates to an untrusted site, the page content enters the context window and can influence Claude's behavior. Stick to trusted applications and known sites. Pair it with Dispatch and you can tell Claude from your phone to do something that requires navigating your desktop, your browser, or any application that doesn't have a connector yet.
**Claude in Chrome
**The Chrome extension lets Claude read pages, click elements, fill forms, and navigate sites alongside you. The features most people miss:
Record a workflow by showing Claude the steps once, and it learns to repeat them. Any repetitive browser task you do more than twice a week should become a recorded workflow. Claude Code integration lets you build in your terminal and test in the browser simultaneously. The extension reads console errors, network requests, and DOM state, so when your frontend breaks, Claude already knows why before you ask. You can control browser actions from Claude Desktop without switching windows. Admins on Team/Enterprise can manage the extension org wide with allowlists and blocklists for site access.
Practical example: record a workflow for checking competitor pricing pages weekly. Claude navigates to each site, extracts the pricing data, and drops it into a comparison spreadsheet in your Cowork folder. What used to take 45 minutes of clicking around becomes a one click replay.
Be cautious about which sites you give access to. Web content is a primary vector for prompt injection. Limit access to trusted sites.
Real Use Cases You Can Run Today
Organize months of accumulated files. Point Cowork at a folder with 6 months of dumped files. Receipts, contracts, notes, screenshots.
Claude reads every file, categorizes them, renames with dates, creates the structure, gives you a log. 10 minutes instead of 2 hours.
This user used Cowork to organize 317 Disney World videos by GPS location. It extracted GPS coordinates from the metadata, determined which Disney location each was from, and organized them into folders accordingly.
Client deliverable from raw materials. You have meeting notes, a transcript, and some research links. You need a polished report.
Claude reads all your raw materials, synthesizes into a structured report, matches your template formatting, saves it ready to send. 15 minutes instead of 90.
Weekly research brief on autopilot. Set up a scheduled task for competitive intelligence. Every Monday at 7am, Cowork researches competitors, checks industry publications, and saves a formatted brief. You review when you're ready. Combine with connectors to pull live data from Slack, Gmail, and Drive.
Financial modeling. One creator asked Cowork to build a social media exit valuation model. It planned itself, found formula errors itself, fixed itself, and delivered a Wall Street style Excel file with 129 formulas across four valuation methods. Blended valuation range included revenue multiples, EBITDA multiples, audience/subscriber value, and a 5 year DCF. Honestly, that's pretty nuts.
The Honest Limitations
Cowork eats your usage fast. A single complex session can burn through what would normally be dozens of regular chat conversations. On the Pro plan ($20/month), you'll feel the limit within a week of daily use. Community reports suggest heavy Cowork users hit rate limits in 3-4 days on Pro, which is honestly brutal if you're in the middle of something important. Multi-step tasks with file reading, document creation, and parallel subtasks are compute-intensive. If Cowork becomes your main workflow, Max ($100/month for 5x Pro usage, or $200/month for 20x) is probably the move. Monitor Settings > Usage to see where you are before you get cut off mid-task.
Context compaction will hit you on long sessions. When a Cowork session approaches its context limit, Claude automatically summarizes earlier parts of the conversation to make room. This keeps the session alive but you lose detail on earlier stuff. Numbers get rounded, specific file references get vague, decisions you made early on get compressed into summaries. If you notice Claude starting to reference "typical patterns" instead of the specific files it read earlier, compaction already happened. And yeah, it's as annoying as it sounds. For critical details, tell Claude to write key findings to a file you can reference later. That way the information persists even when the context gets compressed.
It's still a research preview. Anthropic is explicit about this. It can misread files. It sometimes takes an odd approach when a simpler one would work. Agents can go in a weird direction about 10% of the time on complex multi-step tasks, and the final output has a section that doesn't match the rest. Review what it produces before sending anything to a client.
No cross session memory. Every new Cowork session starts completely fresh. No knowledge of who you are, no memory of what you discussed yesterday. This is the biggest friction point until you discover that your context files are the workaround. That's why the setup matters so much. Your preferences live in files. Your project plans live in documents. Your standards live in instructions. If you want continuity, build it into files. But the upside is massive: a well-documented workflow is portable, shareable, and version-controlled.
Tasks stop if you close the app. Cowork runs as an active session inside Claude Desktop. Close the window, the task stops. Put your computer to sleep instead. The session survives that.
Desktop only. No mobile Cowork, no browser version, no syncing between devices (Dispatch bridges this partially but it's not the same). Build your context files in a cloud synced folder (iCloud, Dropbox, OneDrive) so at least your files are consistent across machines.
Claude Code: The Developer's Operating System
If Cowork is the knowledge worker's tool, Claude Code is the developer's.
Claude Code started as a command line coding tool in February 2025. It's now a full extensible platform for orchestrating AI agents across your entire development workflow, pulling $2.5 billion in run-rate revenue.
You install it via npm (npm install -g @anthropic-ai/claude-code), cd into your project, type claude, and you're talking to an agent that can see your entire codebase and actually do things with it. Read files, write files, run commands, search the web, execute tests, commit code, the works. There's also a web version at claude.ai that got a major upgrade in February with multi repo sessions, better diff and git status visualization, and slash commands. The terminal version is still where the deepest features live.
But the coding itself isn't what separates Claude Code from everything else. The extension layer is what turns it from a fancy autocomplete into a configurable platform.
CLAUDE.md: Your Project's Instruction Manual
When you start a session, the first thing Claude does is read your CLAUDE.md. It loads straight into the system prompt and stays in mind for the entire conversation. Whatever you write here, Claude follows.
Most people either ignore it completely or stuff it with garbage that makes Claude worse instead of better. There's a threshold where too much or too little information means worse output.
What to put in it:
Build, test, and lint commands (the actual bash commands that matter). Key architectural decisions ("we use a monorepo with Turborepo"). Non obvious gotchas ("TypeScript strict mode is on, unused variables are errors"). Import conventions, naming patterns, error handling styles. File and folder structure for the main modules.
What NOT to put:
Anything that belongs in a linter or formatter config. Full documentation you can already link to. Long paragraphs explaining theory. Keep it under 200 lines. Files longer than that start eating too much context, and Claude's instruction adherence actually drops because it's competing with Claude Code's own system prompt for attention.
Tell it why, not just what. "Use TypeScript strict mode" is okay. "Use TypeScript strict mode because we've had production bugs from implicit any types" is better. The why gives Claude context for making judgment calls you didn't anticipate.
Update it constantly. Press # while working and Claude adds instructions to your CLAUDE.md automatically. Every time you correct Claude on the same thing twice, that's a signal it should be in the file. Over time it becomes a living document of how your codebase actually works.
Bad CLAUDE.md looks like documentation written for a new hire. Good CLAUDE.md looks like notes you'd leave yourself if you knew you'd have amnesia tomorrow.
The CLAUDE.md Hierarchy
Most people miss this entirely. There isn't one CLAUDE.md. There are four layers that merge at session start:
1. Managed Policy (org-wide): IT deployed. Cannot be overridden. Organization-wide rules.
1. ~/.claude/CLAUDE.md (global): Your personal preferences across all projects. Not version controlled.
1. ./CLAUDE.md (project root): Team instructions. Committed to git. Everyone who clones the repo gets these.
1. CLAUDE.local.md (personal project overrides): Your personal tweaks for this specific project. Automatically gitignored.
Higher layers win when there's a conflict. This layering is what makes Claude Code work for teams instead of just individuals.
The most common team issue: a developer puts instructions in their user level config (~/.claude/CLAUDE.md) instead of the project level file. They get perfect behavior. A new team member clones the repo, doesn't have those user-level instructions, and gets inconsistent results. Everyone blames the model when the actual problem is configuration. I've watched a team debug "random Claude behavior" for two days before someone realized the senior dev's personal config was doing all the heavy lifting. Always put team standards in the project-level file.
Rules Directory: Modular Instructions That Scale
Once your CLAUDE.md gets crowded (and it will), split instructions into .claude/rules/ files. Every markdown file in this directory loads alongside your CLAUDE.md automatically.
Each file stays focused. The person who owns API conventions edits api-conventions.md. The person who owns testing edits testing.md. Nobody stomps on each other.
Path scoped rules are where this pays off. Add YAML frontmatter with glob patterns and the rules only activate when Claude is working with matching files:
This catches every test file regardless of directory. Directory level CLAUDE.md files only apply to files in that one directory. For conventions that must apply across 50+ directories of test files, path specific rules are the correct approach. They also reduce token usage because they only load when relevant.
Commands vs Skills vs Agents
These three extension types work differently. Using the wrong one creates friction.
Commands (.claude/commands/) are slash commands you trigger manually. A file named review.md becomes /project:review. Write markdown with instructions. The ! backtick syntax runs shell commands and embeds output before Claude sees the prompt.
Now /project:review injects the real git diff into the prompt automatically. Use $ARGUMENTS to pass parameters: /project:fix-issue 234 feeds issue 234's content directly in.
Project commands (.claude/commands/) are committed and shared. Personal commands (~/.claude/commands/) show up as /user:command-name.
Skills (.claude/skills/) are workflows Claude invokes on its own when the task matches. You don't type a slash command. Claude reads the skill's description, recognizes the current task matches, and activates automatically. Commands wait for you. Skills watch for the right moment and act on their own.
Skills are folders, not single files. They can include scripts, reference docs, data, templates. A SKILL.md with YAML frontmatter defines the trigger:
Skills now also support effort frontmatter, letting you override the model's thinking effort level when invoked. And they support on-demand hooks that only activate when the skill is called. /careful blocks destructive commands. /freeze blocks edits outside a specific directory.
Anthropic's own engineering team uses hundreds of skills internally across nine categories: library/API reference, product verification, data fetching, business process automation, code scaffolding, code quality review, CI/CD deployment, runbooks, and infrastructure operations. On March 7, they open sourced 17 of these skills at anthropics/skills on GitHub, covering creative design, document creation, technical development, and enterprise communication. The highest value content in any skill is the gotchas section. Build it from the stuff that burned you.
Agents (.claude/agents/) are specialized subagent personas with their own system prompts, tool access, and model preferences.
The tools field restricts what the agent can do. A security auditor only gets Read, Grep, and Glob. No write access. That's intentional. The model field lets you use cheaper models for focused tasks. Haiku handles most read only exploration well.
Subagents keep your main context clean. The main agent's context window fills up with exploration. Subagents do the messy work in isolated context and return compressed findings. Your main conversation only gets the answer.
Tasks
Shipped January 22. Anthropic upgraded the old Todos system into Tasks, a proper project management primitive. Tasks have dependencies, are stored in the filesystem (~/.claude/tasks), and multiple subagents or sessions can collaborate on the same task list. When one session updates a task, the change broadcasts to all sessions working on that list. Set a task list as an environment variable and spin up parallel agents that coordinate through it. This is the backbone of multi-session workflows and how Agent Teams stay organized.
Agent Teams
Shipped February 5 alongside Opus 4.6 as an experimental feature. Where subagents are isolated workers reporting back to a coordinator, Agent Teams are a collaborative squad. Teammates message each other directly through an inbox based system, divide work through a shared task list with dependency tracking, and coordinate in parallel.
Up to 10 simultaneous teammates. A lead session coordinates work, assigns tasks, and synthesizes results. Teammates work independently, each in their own context window. Unlike subagents, teammates can share findings, challenge each other, and coordinate without going through the lead.
Enable with CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1 in settings.json.
Where this actually shines: say you're adding a new feature that touches the API, the frontend, and the test suite. Spin up three teammates. One handles the API endpoint. One builds the React component. One writes the integration tests. They coordinate through the shared task list so the test teammate knows what endpoints and components to test. All three work in parallel instead of sequentially. A task that takes one session 90 minutes finishes in 30.
They add coordination overhead and consume significantly more tokens than a single session. Use them when parallel exploration adds real value and teammates can operate independently. For sequential tasks or same file edits, stick to single sessions or subagents.
Remote Control
Shipped February 24 for Max users, later expanded to Pro. The predecessor to Channels. Run claude rc in your terminal, then open the Claude mobile app or claude.ai on another device to continue controlling that session remotely. You start a task on your desktop, walk away, and keep steering it from your phone. Channels (below) superseded this for most use cases by adding Telegram and Discord as interfaces, but Remote Control is still the simplest way to get phone to terminal access without setting up a messaging bot.
Claude Code Channels: Always On Messaging
Shipped March 20 as a research preview. If you've ever wanted to text your AI agent from your phone and have it execute on your local machine, this is it.
Channels hooks up a running Claude Code session to Telegram or Discord. Send a message from your phone. Claude processes it using your full local dev environment (files, tools, git, everything). Replies through the same chat app. Your session sits in the background. Events come in from the outside world. Claude handles each one with full project context. Your presence at the terminal is optional.
This is the same interaction model that made OpenClaw go viral after its November 2025 release. A persistent AI worker you can message 24/7 through common chat apps. Except Channels is native to Claude Code, backed by Anthropic's safety infrastructure, and built on MCP so the architecture is extensible.
Setup:
Open Telegram, send any message to your bot. It replies with a pairing code. Use /telegram:access pair <code> to complete the link. The pairing locks the bot to your specific user ID so nobody else can control your session.
Discord follows the same pattern with its own plugin.
Current limitations: you need to keep a terminal session open (combine with tmux, screen, or a background process). Only Anthropic approved plugins during the research preview. Permission approvals still require terminal access. But the plugin architecture is built for expansion. Slack, WhatsApp, and iMessage have all been requested, and the docs for building custom channels are already public.
The Channels setup took me about 10 minutes. Telegram bot, pairing code, done. Now I message my Claude Code session from my phone while I'm out. Recently I told it to refactor an auth flow from my phone while grabbing coffee. Came back to my desk and the PR was ready for review. That's when this stopped feeling like a tool and started feeling like infrastructure.
Hooks
Shell commands that fire automatically at specific lifecycle events. Pre-commit, post-tool-call, when Claude tries to edit certain files. These aren't about AI intelligence. They're about deterministic control.
Auto-lint every file Claude writes. Block sensitive files from being committed. Send a Slack notification when Claude finishes. Run type-checking after every edit. Enforce compliance rules that must be followed 100% of the time.
Here's a starter hook that prevents Claude from ever committing secrets:
Add this to .claude/settings.json and it catches credential files before they ever reach your repo. Zero reliance on the model getting it right. Deterministic.
Recent additions include PostCompact hooks (fires after context compaction, useful for logging what got compressed) and ExitWorktree hooks.
The decision framework: hooks are deterministic guarantees. Use them for business rules that must work every time. Prompts are probabilistic guidance. Use them for preferences and soft rules. If the business would lose money or face legal risk from a single failure, use hooks.
MCP (Model Context Protocol)
The open standard for connecting Claude to external services. Databases, APIs, GitHub, Slack, Telegram, Google Drive, and anything else you can build a server for.
Think of MCP as USB-C for AI. Instead of building custom integrations for each data source, you build against one protocol. MCP servers expose data and capabilities. Claude connects to them. The entire Channels feature is built on MCP. The Telegram and Discord integrations are themselves MCP servers. The plugin architecture runs on it. If you're building anything that connects Claude to external data, you're building on MCP whether you know it or not.
MCP server configuration lives in .mcp.json at the project level (version controlled, shared with the team) or ~/.claude.json for personal servers. Environment variable expansion (${GITHUB_TOKEN}) keeps credentials out of version control.
Before building a custom MCP server, check if a community server already exists for your integration. Jira, GitHub, Slack, Notion, Linear, and dozens of other common tools already have community servers. Only build custom when community servers can't handle your team specific workflows.
Run /mcp periodically to check token costs per server. I've seen projects where a forgotten MCP connection was eating 15% of the context window every session. Disconnect servers you're not actively using.
Plugins
This is where your team's standards actually live. One person builds a plugin that captures your team's code review standards, deployment checklist, and architectural patterns. Everyone installs it. Consistent, on-brand, process compliant output across the entire organization because the standards live in the plugin, not in individual memory.
Plugins bundle skills, hooks, subagents, and MCP servers into a single installable unit. If you've built a workflow for code review that combines a skill file, a subagent, and a pre-commit hook, package it into a plugin and distribute it through the marketplace or your team's private marketplace.
Plugin skills are namespaced (/my-plugin:review) so you can run five plugins without them stepping on each other. The March 20 release added the ability to declare plugin entries inline in settings.json via a source: 'settings' marketplace source.
Start here: install one official plugin that matches your role. Use it for a week. Then build your own that encodes the specific standards your team uses. That second step is where the real leverage kicks in.
Headless Mode and CI/CD
Claude Code runs in non interactive mode with the -p flag. That's how you integrate it into automated pipelines. PR reviews, security scans, test generation, documentation updates. Without the -p flag, your CI job hangs waiting for interactive input.
Combine with --output-format json and --json-schema for machine parseable structured output. Automated systems can post findings as inline PR comments. A basic GitHub Actions setup: trigger on PR open, run claude -p "review this diff for bugs and missing tests" --output-format json, parse the output, post as review comments. Takes about 15 minutes to set up and catches issues before any human reviews.
Always use an independent Claude instance for code review, not the same session that wrote the code. The session that generated code retains reasoning context that makes it less likely to question its own decisions. A fresh instance catches more.
Claude Code Security
This one is a big deal for any team shipping production code. Claude Code can now review entire codebases to identify security vulnerabilities. Traditional scanners match code against known patterns and flag everything that looks suspicious, which means 30-60% false positive rates and a lot of wasted time triaging noise. Claude reasons about code semantically instead, tracing data flows across the codebase and looking for logic flaws that cross file boundaries.
Anthropic reports a false positive rate below 5%. In testing with Opus 4.6, their security team found over 500 vulnerabilities in well reviewed open source projects, including bugs that had gone undetected for years. Claude then runs its own findings through a red-teaming process to filter out false positives. Point it at your repo and see what it finds. You might be surprised.
Voice Mode
Claude Code supports voice input for hands free coding. Talk to it instead of typing. You're reviewing code on one monitor, dictating refactoring notes without switching windows. Or you're pacing around your office trying to figure out why the auth flow is broken and you just narrate the fix. Toggle with /voice. It had some rough edges early on (WebSocket drops, activation quirks on fresh installs) but Anthropic has been patching it steadily.
Code Review and Automated PR Workflows
Claude Code also handles automated code review on pull requests. In headless mode (-p), it reads a PR diff, evaluates code quality, flags bugs, checks for missing tests, and posts findings as inline PR comments. Combined with the --output-format json flag, this slots directly into GitHub Actions or any CI pipeline. Teams are also using Claude Code for automated preview generation and merge preparation, where Claude reviews changes, runs tests, generates a summary, and stages the merge if everything passes.
Beyond Chat, Cowork, and Code, there's more. Here's what else shipped.
Claude for Excel and PowerPoint
Claude works as add-ins inside both apps. The March 11 update lets them share full context, so work in Excel (data analysis, formulas, pivot tables, conditional formatting) flows directly into PowerPoint (presentations, visualizations) without losing information. Skills work inside the add-ins too, and organizations on Amazon Bedrock, Google Cloud Vertex AI, or Microsoft Foundry can connect via an LLM gateway.
The workflow that saves the most time: drop your raw data into Excel, tell Claude to analyze it, build the pivot tables, and highlight the key trends. Then open PowerPoint and tell Claude to build a presentation from those findings. The shared context means it already knows the data, the key takeaways, and the numbers. No re-explaining, no copy pasting between apps, no reformatting. One creator reported going from raw quarterly data to a board ready deck in 20 minutes.
Microsoft launched their own "Copilot Cowork" on March 9 powered by Claude's models, as part of a new $99/user E7 licensing tier. Anthropic's technology is becoming the engine inside other companies' products.
Custom Visuals in Chat
Shipped March 12 as a beta, available to all users including the free tier. Claude can now create interactive charts, diagrams, and visualizations inline in conversations. Built using HTML and SVG, they support hover and click interactions and evolve as the conversation continues.
These are different from Artifacts. Artifacts are persistent, shareable documents in a side panel. Inline visuals are ephemeral. They live in the conversation flow, change as you ask follow ups, and can disappear as the chat evolves. Think of them like Claude pulling out a whiteboard mid conversation. You can save them as SVG/HTML or convert to a full artifact if you want to keep them.
Use inline visuals when you're exploring data or explaining a concept and want to see it while you talk. Use artifacts when you need a finished deliverable to share or download. The sweet spot: "show me a flowchart of how this auth system works" in the middle of a debugging conversation. Claude draws it, you spot the broken step, you keep talking. No context switch to a separate tool.
The API
For developers building on Claude, the biggest practical changes:
Adaptive thinking with configurable effort levels replaced the old budget_tokens approach. Set effort to "medium" on Sonnet 4.6 for most tasks and you'll cut costs significantly without losing quality. The new "max" effort level on Opus 4.6 is there when you need peak capability but burns tokens fast. Thinking tokens are billed as output tokens at $25/M for Opus, so effort level is your primary cost control lever in automated pipelines.
Fine grained tool streaming is GA, no beta header needed. Structured outputs are GA. Data residency controls let you pin inference to the US at 1.1x pricing. The 1M context window works automatically for requests over 200K tokens, no code changes required.
Code execution is free when combined with web search or web fetch. Dynamic filtering on search results at no extra cost beyond standard tokens. Web search and web fetch are both GA now with no beta header required. This is the move most people miss.
API Skills are new and most developers haven't found them yet. Anthropic ships pre built skills for PowerPoint, Excel, Word, and PDF processing. You can also upload custom skills via the /v1/skills endpoints to package domain expertise and organizational workflows into reusable capabilities. Skills require code execution to be enabled. If you're building document processing pipelines on the API, this replaces a lot of custom tooling.
Context compaction automatically summarizes older context as conversations approach limits. When a session fills up, the API identifies portions that can be compressed while preserving key information. Fewer compaction events overall since the 1M window went GA.
The Numbers
Anthropic closed a widely reported $30 billion Series G at a $380 billion valuation on February 12, 2026. Led by GIC and Coatue. The second largest venture funding deal of all time behind OpenAI's $40 billion raise. Microsoft and Nvidia were among the contributors.
Run rate revenue hit $14 billion, growing 10x annually for three straight years. Claude Code alone is at $2.5 billion in run rate revenue, more than doubling since January. Business subscriptions quadrupled in the same period. Eight of the Fortune 10 are Claude customers. Over 500 customers spend more than $1 million annually, up from roughly a dozen two years ago. Customers spending over $100K annually grew 7x in the past year. Enterprise accounts for 80% of the business, and any organization can now purchase Enterprise directly on the website with no sales conversation.
On the enterprise infrastructure side: HIPAA-ready Enterprise plans launched in January for healthcare organizations processing protected health information. The Enterprise Analytics API (February 13) gives programmatic access to usage and engagement data aggregated per organization per day. The kind of stuff that gets procurement teams to sign.
Anthropic also launched the Claude Partner Network with a $100 million investment in training, comarketing, and technical architecture support. The first professional credential, the Claude Certified Architect (Foundations), dropped March 12 as a proctored architecture level exam. It tests agentic design, MCP integration, Claude Code configuration, and production reliability patterns. Accenture is training roughly 30,000 professionals on it. Anthropic Academy provided 13 free courses on Skilljar as the official prep path on March 2. Two additional courses have been added since, expanding the selection to 15. Additional certifications for sellers, developers, and advanced architects are planned for later this year. If you're a consultant or agency, this credential stack is going to matter when enterprise clients start asking who on your team is certified.
The internal numbers: Anthropic's engineers use Claude for roughly 60% of their own work, up from 28% a year ago. They ship 60 to 100 internal releases per day. Cowork went from zero to launch in 10 days, built entirely with Claude Code. The tools are building the tools, and that feedback loop is why the shipping pace went from months to weeks to days.
What This All Means If You're Building with AI and Agents Right Now
The infrastructure layer is commoditizing fast. Features that required custom frameworks six months ago are now native platform features. The moat was never the infrastructure. It's always been taste, distribution, and what you build on top of the tools.
For builders shipping products on Claude, the extension system is where the leverage lives. Skills, subagents, Agent Teams, hooks, Channels, MCP, plugins. A well configured Claude Code setup with custom skills and scoped agents is a different tool entirely than someone typing prompts into a chat window. Learn the layers. Configure them for your workflow. That investment pays back on every session.
For knowledge workers, Cowork changes your daily workflow starting this week. Build your context folder. Set your global instructions. Install two plugins. Use AskUserQuestion on everything. Set up one scheduled task. The phone to desktop bridge with Dispatch means dead time becomes productive time. Today's Computer Use launch takes it even further.
For team leads, the plugin marketplace and enterprise features mean you can standardize how your entire organization uses Claude. Build plugins that encode your team's institutional knowledge and distribute them. That's how you go from "we use AI sometimes" to "AI is how we operate."
The pace isn't slowing down. It's accelerating because Anthropic is using their own tools to build the next generation of tools. Each new model makes it faster to build the next one. That compounds, and it changes the math on everything.
Learn the platform now. Not next quarter or next month. Literally now.
If you made it this far, you're already ahead of 99.9% of people who will bookmark this and likely never come back to the resource. They'll keep using Claude like a basic chatbot. You won't.
I'm not an engineer. I'm self taught. I'm not claiming to have all the answers or the best Claude setup for every single use case here either, and if someone claims they do, they're lying to you. Everything here comes from building with this stuff daily, breaking things constantly, and writing down what actually works so nobody else has to guess. You need to be having fun and breaking things. It's how you learn. If something in here is wrong, missing, or outdated, tell me for sure too. I'd rather fix it than let someone build on bad information. Thanks for reading everyone!
If you're not following me (@kloss_xyz) already, turn on those post notifications for all things agents, systems, and building with AI.
More articles are on the way.
Resources
- Claude Documentation (start here for any feature): docs.claude.com
- Claude Code Documentation (extensions, skills, agents, hooks): code.claude.com
- Claude Cowork (launch post + feature overview): claude.com/blog/cowork-research-preview
- Claude Blog (every feature announcement as it drops): claude.com/blog
- Claude Code Channels (Telegram/Discord setup guide): code.claude.com/docs/en/channels
- Claude Partner Network (free to join, credential access): anthropic.com/partners