I Ranked Every Claude Feature. Here's the Tier List.

<!-- IMAGE: hero - the full tier board. A clean S/A/B/C/D tier-list grid infographic with all 60-plus Claude features sorted into rows, S tier glowing at the top, frankyao.com brand colors, the kind of board you would screenshot and argue with. -->

TL;DR

• I ranked every Claude feature by one thing only: how much it changes my actual day doing knowledge work and automation. Not benchmarks. Not building apps.
• The whole S tier shares a pattern. Every feature up there removes me from the loop. They let the work keep going while I am gone.
• The crown goes to /goal: set a finish line and Claude keeps working until it is met.
• A, B, C, and D tiers are full of genuinely useful tools. They just do not change how I work the way the walk-away features do.
• You will disagree with some of these. Good. That is the point of a tier list.

Most "best Claude features" posts are written by people who opened the app twice. This one is not. I run Claude every day for real automation and knowledge work, and I have opinions that come from actually living in the thing, hitting its limits, and rebuilding my workflow around the parts that earned it.

So here is the lens, and it matters: I ranked every feature by how much it changes my actual day. Not how clever it is. Not how it scores on a benchmark. Not whether it can build a slick app. Just one question, asked feature by feature: does this change what my Tuesday looks like, or doesn't it? That lens puts some famous features lower than you would expect, and some quiet ones near the top.

You will disagree with some of these. Good. A tier list nobody argues with is a tier list nobody read. Let's count down.

What is the one pattern that runs through the whole S tier?

Before the rankings, the through-line, because it is the actual point of this piece.

When I lined up the twelve features that genuinely changed how I work, I noticed they all do the same thing. They take me out of the loop. /goal, Routines, /loop, Subagents, Agent Teams, Remote Control, Auto Memory. None of them are about getting a smarter answer to a single question. They are about letting the work continue without me sitting there feeding it the next prompt.

That is the real shift of the last year, and most people are still missing it. The change was not a better chatbot. It was the move from AI that answers to AI that runs while you are gone. The features that changed my work are not the smart-answer features. They are the walk-away features.

Keep that in your head as we go down the list. D tier is the stuff that makes a single session usable. S tier is the stuff that makes the session unnecessary. Now the countdown.

S Tier: the features that actually changed how I work

<!-- IMAGE: S-tier countdown graphic. A vertical countdown strip, 12 down to 1, each rung labeled with the feature name and a one-line tag, /goal sitting at the bottom as the crowned number one with a small finish-line flag. -->

These are the twelve. I am counting them down from twelve to one, and number one is the crown. Every one of these earns its spot by doing work I used to do by hand, which mostly means babysitting a prompt.

12. Ultraplan. This is where the S tier starts, and it starts with planning. Ultraplan offloads the planning step to the cloud, then lets me review and comment in the browser. Planning was the part I never wanted to delegate, and once I did, a whole category of work stopped needing me at the keyboard. I open the browser, read what it proposed, leave a comment, and move on.

11. /insights. A local report on my last 30 days. It surfaces my own patterns and friction points, the things I do over and over and the places I keep getting stuck. I expected a vanity dashboard. Instead it is the feature that tells me what to automate next. You cannot remove yourself from a loop you have not noticed you are in, and /insights makes the loops visible.

10. Auto Memory. It remembers my preferences and patterns per repo, on its own, without me writing anything down. The "on its own" part is the whole thing. I am not maintaining a notes file, and I am not re-explaining how I like things done every time I come back. The tool quietly learns the shape of how I work in each project.

9. Agent Teams. A named team of agents that run in parallel and message each other. This is the first feature on the list that genuinely feels like delegating to a group instead of a single helper. They run at the same time, they talk to each other, and I check the result. When the job is big enough to need more than one worker, this is the difference between watching one thing happen and watching several happen without me in the middle.

8. /rewind. Roll back code and conversation to any earlier checkpoint. This one is here for a quieter reason than the rest. The walk-away features only work if walking away is safe, and /rewind is the safety net. If an agent runs off in a bad direction while I am not watching, I roll the whole thing, code and chat both, back to a checkpoint and continue. It is the feature that lets me trust all the others.

7. Subagents. Spin off background agents, each in its own context window. The "own context window" detail is what makes this work instead of a mess. Each subagent gets its own clean room to think in, so a big background job does not pollute the main thread I am working in. I fire one off, keep going on my main task, and collect the result later.

6. /loop. Repeat a prompt on an interval until a condition is met. Plain on paper, big in practice. So much of real work is "keep doing this until X is true," and before /loop that meant me being the timer and the checker. Now I set the interval, set the condition, and the tool watches for the finish instead of me.

5. Remote Control. Drive my local Claude Code session from my phone or the web. This is the one that broke the desk-chair tether. My local session runs on my machine, and I can steer it from my phone or a browser anywhere. The work does not stop because I stood up. Being away stopped meaning being blocked.

4. Routines. Cloud agents that run on a schedule with my laptop closed. Read that last part again. Laptop closed. Up to now everything on this list still assumed my machine was on. Routines do not. They live in the cloud and fire on a schedule whether I am working, asleep, or on a plane. A scheduled agent that emails me a morning digest is the kind of thing that just happens now, and I did nothing that morning to make it happen.

3. Status Line. A scriptable live status bar showing context, effort, and anything else I script into it. This is the practitioner's pick, and I will defend it. It looks small. It is a bar. But it is scriptable, so it shows me exactly what I want to know about a running session at a glance. When you are running things that work without you, knowing their state at a glance is most of the battle.

2. Skills. Reusable /skills folders that teach Claude my own workflows. This is the one I would hand to someone else first, because it compounds. A Skill is a folder that teaches the tool how I do a specific thing, once, and then it knows. I have skills for the work I do constantly: ui-ux-pro-max for interface work, one that drives my whole SEO and content pipeline, one for spinning up a new client site from scratch. Each is a workflow I taught the tool a single time and never have to re-explain or re-supervise again. If /goal is the crown, Skills are the reason the rest of the kingdom keeps running.

1. /goal. The crown. Set a finish line, and Claude keeps working until it is met.

This is number one and it is not close, because it is the purest expression of the entire pattern. Every other feature on this list helps me step out of the loop in some specific way. /goal is the loop, gone. I do not set a prompt. I do not set an interval. I set an outcome, a finish line, and then I leave. The tool keeps working until the thing I asked for is actually true. Nothing changed my actual day more than that.

That is the S tier. Notice that not one of those twelve is "it gives smarter answers." Every single one is about the work continuing without me. That is not an accident. It is the thesis of this whole post.

<!-- IMAGE: the "walk-away" concept. A simple before/after infographic. Left side: a person glued to a desk feeding prompts one by one. Right side: the same person walking out the door while agents, loops, routines, and a goal keep running on their own. Caption: AI that answers vs AI that runs while you are gone. -->

What is in the A tier, so close to S?

These are the honorable mentions. Every one is genuinely good, and a couple I use daily. They sit in A instead of S because they make my work better without quite removing me from the loop the way the crown features do. If the S tier did not exist, half of these would be my favorite feature.

• Google Workspace CLI runs Gmail, Drive, Docs, and Sheets straight from the terminal. Proof the tool can drive basically any CLI.
• Dispatch lets me start a task on my phone and have Claude run it on my desktop and send back the result.
• Claude Desktop App is the Mac and Windows home base for chat, Cowork, and connectors.
• Claude Design turns a description into polished designs, slides, and one-pagers.
• Plugins and Marketplace installs whole bundles of skills, agents, and hooks in one go. It is how a skill like ui-ux-pro-max gets onto my machine in the first place.
• Workspace Connectors wire up Gmail, Calendar, and Drive so Claude works with my Google data directly.
• File Creation and Editing builds Excel, Word, PowerPoint, and PDF files I can download.
• Ultracode is the top effort tier: max reasoning plus automatic multi-agent workflows.
• Effort Levels let me dial reasoning depth from low to max with /effort.
• Channels pipe Telegram, Discord, or iMessage into a live session, two-way.
• Hooks run my own commands automatically on session events.
• Auto Mode auto-approves the safe actions and blocks the risky ones, so I am not clicking yes all day.
• /background pushes a session to the background to reattach later.
• Agent View gives me one dashboard of every background agent and its status.
• Plan Mode has Claude research and propose a plan before touching anything.
• /btw lets me ask a quick side question without derailing the main task.
• /usage breaks down token and cost for the session.
• /context shows exactly what is filling the context window.
• Prompt Caching reuses repeated context so big sessions cost less and respond faster.

What lands in B tier?

Genuinely useful, just not my top thirty. B tier is where the powerful-but-specialized tools live, the ones that are excellent when you need them and quiet when you do not. They earn B because they help in specific moments rather than reshaping the whole day.

• Dynamic Workflows has Claude write a script that runs up to a thousand agents for huge jobs.
• Git Worktrees run agents in isolated copies of the repo so they never collide.
• Ultrareview is a fleet of cloud agents that verifies every bug before reporting it, which means fewer false alarms.
• /deep-research fans out web searches and returns one cited report.
• Claude in Chrome is a side panel that clicks, fills forms, and acts across my tabs.
• Artifact Apps build shareable apps powered by live Claude with no code.
• Computer Use points and clicks to run any app when there is no connector for it.
• Advanced Research chains searches across the web and my apps into a cited report.
• Projects keep chats, files, and instructions together in one workspace.
• Mobile Push pings my phone when a long task finishes or needs me.
• /recap gives a quick recap of where I left off.
• Interactive Charts draw charts and diagrams right in the chat.
• Artifacts render outputs in a side panel I can edit live.
• Cross-Device Convo lets me start on my phone and continue on desktop in one thread.
• Browser Recording records a browser task once and repeats it on demand.
• Memory carries relevant context from past chats into new ones.
• Connectors Directory lets me browse and add vetted connectors.
• Claude for M365 puts native Claude inside Word, Excel, PowerPoint, and Outlook.

Why are Cowork and voice in C tier?

This is the tier where my personal lens shows hardest, so I want to be honest about it. C tier is "fine tools, just not how I work." These are good features. They rank low for me because I live in the terminal and in automation, not in a no-terminal desktop or a spoken conversation. Your ranking here could be the opposite of mine, and you would not be wrong. If you work in a desktop app instead of a terminal, mentally promote this whole tier.

• Voice Mode is full spoken conversations, hands-free.
• Cowork is the no-terminal agent desktop for knowledge work.
• Interactive Connectors add and manage MCP connectors right inside the CLI.
• Citations ground answers in your documents with exact citations through the API.
• Chat Search searches across all your past conversations.
• Claude in Slack does the task when you mention it and posts back.
• Local File Access connects a folder so Claude reads and organizes your files.
• Cowork Plugins are bundles of skills and connectors for Cowork.
• Cowork Projects group Cowork tasks into a persistent workspace.
• Incognito Chats are chats that are not saved, remembered, or trained on.
• /teleport pulls a Claude web session down into the terminal.

What is the D tier, and why is it not an insult?

D tier is table-stakes and plumbing. Low here does not mean bad. These are real and useful, they are just nobody's favorite, because they are the floor that everything else stands on. Half of them you would only notice if they vanished.

• Memory Files (CLAUDE.md) is a file Claude reads every session to know your project.
• Context Lifecycle is /clear, /compact, and /resume to manage the conversation.
• Web Search has Claude search the web and cite its sources.
• Permissions control what Claude can do without asking.
• Fast Mode is Opus with faster output, toggled with /fast.
• File Uploads drop in images, PDFs, and docs.
• Output Styles change how Claude formats its answers.
• Headless Mode runs Claude as a one-shot command in scripts.
• Sandboxed Bash runs shell commands safely in a sandbox.
• Transcript Viewer scrolls back through the whole session with Ctrl+O.
• History Search searches past prompts with Ctrl+R.
• Shell Mode runs a raw shell command with the! prefix.
• Voice Dictation lets me talk instead of type with /voice.
• Custom Themes change the CLI color theme.
• IDE Extension puts Claude Code inside VS Code and JetBrains.
• Mobile Apps put Claude on iOS and Android.
• Code Review reviews my git diff with /code-review.

Here is the thing about D tier, and it brings the whole post back around. D tier is the stuff that makes a single session usable. S tier is the stuff that makes the session unnecessary. The features at the bottom keep you in your seat and make the seat comfortable. The features at the top get you out of the chair entirely. That gap, from "usable session" to "no session needed," is the entire story of what changed this year.

The walk-away test

If you take one thing from this list, take the test I now run on every feature before I get excited about it: does this keep me in the loop, or does it let me leave it? Smarter answers are nice, and I will take them. But the features that actually changed my work are the ones that let the work happen while I am somewhere else. Set a finish line with /goal. Teach it a workflow with Skills. Let Routines run with the laptop shut. That is the difference between a faster chatbot and a tool that does the job while you go do something better with your day.

Rank them however you want. Just rank them by your real day, not by the demo.

FAQ

What is the difference between Skills and Subagents? They solve different problems. Skills are reusable /skills folders that teach Claude your own workflows, so you codify how you do a task once and it knows it from then on. Subagents are background agents you spin off, each in its own context window, to handle work in parallel with your main thread. Skills teach. Subagents delegate. In my ranking Skills sits at number two and Subagents at number seven, both because they remove me from the loop in different ways.

Do I need to know how to code to use these features? Not for a lot of them. Plenty of the highest-impact features are commands and toggles, like /goal to set a finish line, /loop to repeat until a condition is met, and Routines to run cloud agents on a schedule. There is also Cowork, a no-terminal agent desktop, and Voice Mode for spoken conversations, both built for people who do not want a terminal. The deeper, scriptable features like the Status Line and Hooks reward technical comfort, but you can get most of the walk-away benefit without writing code.

What does /goal actually do, and why is it ranked number one? You set a finish line, and Claude keeps working until that finish line is met. I ranked it number one because it is the purest version of the pattern that runs through my whole top tier. Instead of giving a prompt or setting an interval, you describe what done looks like and step away, and the tool works toward that outcome until it is true. Nothing changed my actual day more than handing over the finish line instead of the next instruction.

Why are Cowork and Voice Mode ranked so low if they are good features? Because my ranking lens is "how much does it change my day," and I work in the terminal and in automation. Cowork is a no-terminal desktop and Voice Mode is hands-free spoken conversation, and both are genuinely good. They just are not how I personally work, so they land in C tier for me. If you work in a desktop app or prefer talking over typing, you should mentally promote that whole tier. The low ranking is about fit, not quality.

What does it mean that the S-tier features "remove you from the loop"? It means they let the work continue without you sitting there feeding the next prompt. Routines run on a schedule with your laptop closed. /loop repeats until a condition is met. Subagents and Agent Teams run in the background and in parallel. Remote Control lets you steer from your phone. Auto Memory learns your patterns on its own. The common thread is that the work keeps going while you are gone.

Will I disagree with this tier list? Probably, and that is fine. These are my opinions as someone who uses Claude every day for automation and knowledge work, ranked by my real day rather than by benchmarks or app-building. Your work is different, so your tiers will be different. The useful exercise is not to agree with me. It is to run the same test on your own work and see which features actually change your week.

---

I write about using AI for real work, not the hype version, here every week. If this is your kind of thing, read more on the blog, see what I actually do, and if you want help putting any of this to work in your own setup, get in touch. I answer.