Anthropic · Tool guide · Reviewed 17 October 2026

Claude

Claude is especially good at careful writing, reading long documents, research and coding.

A good choice when you want to improve a difficult document, work closely through a problem or keep details straight across a long piece of work.

Strong at: Rewriting without flattening your voice, reading long documents, working through hard problems and coding
Takes getting used to: It has several separately named tools; start with Chat and add Research or Cowork when a task needs sources or files
You still own: Checking sources, adding the context only you know and making the final call

Tap the job in front of you

What do you need to do?

Use Claude Duet
Anthropic has opened a waitlisted US Pro and Max research preview for Claude Duet; The Human Bit has not tested it. Anthropic says Duet reads, drafts, and prepares work on its own. It shows the proposed action and waits for approval before sending messages, making payments, signing documents or deleting anything. Trading securities or moving money between accounts remain off-limits even with approval.
Real example
Ask for a short brief from approved connected accounts, then compare it with the original threads.
Try this
Prepare a short brief with links to original threads. Separate confirmed deadlines from inferred urgency. Draft replies only; do not send messages, make payments, sign documents or delete anything.
You still do
You choose the connected accounts and explicitly approve any send, payment, signature or deletion. The Human Bit recommends verifying urgency against original threads and checking recipients and figures; these are not vendor-enforced guarantees.
Waitlisted US Pro and Max research preview, not Team or Enterprise yet. Shares Claude usage limits, including background checks. Vendor claims only; not tested by The Human Bit. Checked 22 October 2026; review by 29 October 2026.
Ask Bit to do it with me

Use Claude Duet

Anthropic has opened a waitlisted US Pro and Max research preview for Claude Duet; The Human Bit has not tested it. Anthropic says Duet reads, drafts, and prepares work on its own. It shows the proposed action and waits for approval before sending messages, making payments, signing documents or deleting anything. Trading securities or moving money between accounts remain off-limits even with approval.

Real example: Ask for a short brief from approved connected accounts, then compare it with the original threads.
Try this: Prepare a short brief with links to original threads. Separate confirmed deadlines from inferred urgency. Draft replies only; do not send messages, make payments, sign documents or delete anything.
You still do: You choose the connected accounts and explicitly approve any send, payment, signature or deletion. The Human Bit recommends verifying urgency against original threads and checking recipients and figures; these are not vendor-enforced guarantees.

Waitlisted US Pro and Max research preview, not Team or Enterprise yet. Shares Claude usage limits, including background checks. Vendor claims only; not tested by The Human Bit. Checked 22 October 2026; review by 29 October 2026.

Ask Bit to do it with me

Try it on real work

Useful ways to start with Claude.

Turn rough notes into a clear update

You have scattered bullets, half-finished thoughts and a deadline. You know what happened, but the person reading the update was not there and should not have to decode your notes.

Finished looks likeA concise update that tells the reader what changed, what is blocked and what happens next, while keeping uncertain details visibly marked for you to confirm.

Open the complete task workflow

Compare difficult options

You have several believable options, priorities that pull against each other, and no single choice that is best for everyone.

Finished looks likeA side-by-side comparison, the strongest case against your front-runner, and sharper questions to answer before you commit.

Open the complete task workflow

Create a presentation from source material

The thinking is all there, spread across notes and files, but the storyline and the slide order are still a mess.

Finished looks likeAn editable draft deck with a storyline that holds together and source material you can trace back.

Open the complete task workflow

Features in this guide

10 Claude features explained.

Claude DuetAnthropic has opened a waitlisted US Pro and Max research preview for Claude Duet; The Human Bit has not tested it. Anthropic says Duet reads, drafts, and prepares work on its own. It shows the proposed action and waits for approval before sending messages, making payments, signing documents or deleting anything. Trading securities or moving money between accounts remain off-limits even with approval.

Best for: A low-risk morning brief or draft reply, once admitted to the preview.
Real example: Ask for a short brief from approved connected accounts, then compare it with the original threads.
Start with: Prepare a short brief with links to original threads. Separate confirmed deadlines from inferred urgency. Draft replies only; do not send messages, make payments, sign documents or delete anything.
Where you make it better: You choose the connected accounts and explicitly approve any send, payment, signature or deletion. The Human Bit recommends verifying urgency against original threads and checking recipients and figures; these are not vendor-enforced guarantees.

ChatThe conversational workspace for writing, analysis and thinking together.

Best for: Developing an idea through several thoughtful passes.
Real example: Give Claude a messy strategy memo and ask it to diagnose the argument before editing.
Start with: Ask me three questions that would materially improve your answer before drafting.
Where you make it better: Make sure its questions reflect your real audience and constraints.

Extended thinkingSupported Claude models can spend more time reasoning through a difficult problem. Available controls vary by model.

Best for: Complex planning, technical reasoning and careful comparisons.
Real example: Stress-test a proposed operating model against five failure scenarios.
Start with: Use extended thinking. Show the assumptions that drive the conclusion and test the opposite case.
Where you make it better: More thinking does not fix missing or false inputs.

Web searchSearches current web information and provides sources to inspect.

Best for: A recent fact, current source or discussion of a specific page.
Real example: Compare the current official pricing pages of three products.
Start with: Search the current official sources for [question]. Link every source and separate direct facts from your interpretation.
Where you make it better: Open the links that your decision depends on.

ResearchInvestigates a question across several sources and returns a longer report with links.

Best for: A decision where a quick web search would miss important evidence.
Real example: Compare the customer demand, costs and local rules for entering a new market.
Start with: Before researching, show me the questions you will answer and the kinds of source you will use.
Where you make it better: Open the sources behind the recommendation and check whether local perspectives were missed.

Available on paid Claude plans; limits vary.

ProjectsKeeps one job’s chats, instructions and files together.

Best for: A client, course, product or personal goal you return to regularly.
Real example: Keep current brand guidance, approved examples and campaign work in one place.
Start with: Use the files in this Project. If two files disagree, show me the conflict instead of choosing silently.
Where you make it better: Remove old files and keep separate clients or private subjects in separate Projects.

ArtifactsMakes a document, piece of code or small interactive tool beside the conversation.

Best for: A calculator, worksheet or training tool that someone can use more than once.
Real example: Build a quote calculator from approved pricing rules.
Start with: Build the smallest working version first. Let me test it before adding visual polish.
Where you make it better: Test the rules, empty answers and unusual inputs before sharing it.

CoworkWorks through a longer task and produces editable files.

Best for: Turning a set of source files into a report, spreadsheet or presentation.
Real example: Turn interview notes into a first-draft report and action tracker.
Start with: Show me the file you will produce and the checkpoints first. Ask before changing another app or file.
Where you make it better: Review the plan and every final file before you use them.

Cowork is becoming part of ordinary Claude on Pro and Max. New Pro and Max tasks run in the cloud; Team web/mobile access remains beta and Enterprise requires administrator enablement. Connected local folders still require the desktop app to stay open.

Claude CodeAnthropic's coding agent for working directly inside software projects.

Best for: Understanding, changing and testing a real codebase.
Real example: Trace a checkout bug, implement the fix and run the existing test suite.
Start with: First map the relevant code path and reproduce the issue. Propose a minimal fix and acceptance tests before editing.
Where you make it better: Review diffs, security implications and deployment decisions.

Claude DesignA conversational tool for designs, prototypes and presentations.

Best for: Exploring an interface or visual direction as an editable prototype.
Real example: Create three mobile onboarding directions from a product brief.
Start with: Start with low-detail structure and interaction. Do not choose final visual styling until I approve the flow.
Where you make it better: Bring brand taste, user knowledge and accessibility review.

When you need the detail

Models, plans and sources.

One job, start to finish: One idea, three rooms

An illustrative workflow to try across Claude tools; it is not a measured customer result.

Think it through in chat
You are stuck on a launch plan. You talk it out with Claude on its deeper reasoning setting, argue with its suggestions, and end with a one page brief in a markdown file. The thinking is yours; Claude just kept up and wrote it down cleanly.
Make it in Claude Design
You hand that brief to Claude Design and get back a first brand system and a set of social layouts. You throw away half, keep the two that feel like you, and refine those. The taste is the part a tool cannot supply.
Build it in Claude Code
You give the brief and the design to Claude Code, connected to your GitHub. It plans the work, opens a pull request, and you review the approach before anything merges. It ships the typing; you own what goes live.

Models and modes: start with the default

ChooseWhat it isUse it forRole

Sonnet 5.5Everyday writing and coding

A model to try for writing, analysis and coding. Check your current plan and model menu for access.

Start here

Opus 5.5For difficult multi-step work

Hard coding, careful analysis and lengthy tasks where the extra cost is justified. Paid plans.

Complex work

Fable 5.1For long-running work

Tasks that run through many steps on their own. Fable uses usage credits on Pro. Max includes Fable within 50% of weekly limits; Team requires premium seats for that allowance. Enterprise access follows its plan.

Long tasks

Haiku 5.5Short, repeatable tasks

Try it on a small example before choosing it for repeated work. Check your plan and compare the time and corrections needed.

Test on your task

Start with: Chat for writing and questions; Research when you need sources; Cowork when you need finished files.

Model names and the access notes in this table were rechecked 17 October 2026. Examples are suggestions to try, not measured performance claims.

See the official sources for the separate 17 October 2026 check. Prices, account access and older advice still need the review dates shown on this page.

Plans and prices (5)

FreeUS$0
Chat, web search, file creation, Artifacts, Design, Slides and Docs with Sonnet and Haiku; Free also includes up to five Projects.
ProUS$20/mo
More usage, Opus, Claude Code and scheduled tasks. Fable uses usage credits. US$20 monthly or US$200 billed annually (shown as about US$17 a month).
MaxFrom US$100/mo
People who work with Claude all day: five or twenty times Pro's usage and early access to new features.
TeamUS$25/seat/mo
Teams of 2 to 150. Standard seats: US$25 monthly or US$20 with annual billing. Premium: US$125 monthly or US$100 annually, with five times Standard usage.
EnterpriseUS$20/seat/mo + usage
Billed annually with a 20-seat minimum, plus model and task usage. Includes spend limits, roles and audit logs; sales-assisted arrangements may vary.

Plan prices and access rechecked 17 October 2026; Prices exclude applicable tax; confirm the current plan before paying.

From the field: how people actually work with it

Attributed notes from people using it on real work, kept apart from the official material.

The shift people describe is from doing the work to specifying the outcome and validating the result. What stays yours is the higher order work: deciding what to build, reviewing the approach, and checking it is actually right.
The Human Bit editorial suggestion
A common, boring, reliable habit: plan first on anything non trivial, keep a lean instructions file, and treat a verification pass as non negotiable rather than optional.
The Human Bit editorial suggestion
Teams get more from review when a person runs an AI assisted pass before opening a pull request, so human reviewers spend their attention on architecture and long term maintainability, not obvious issues.
The Human Bit editorial suggestion

Official sources (14), checked 17 October 2026

Last checked against official Claude material on 17 October 2026. Plan access and limits change; confirm the plan page before you pay.

Your first try

Use one real task, not ten demo prompts.

Start with the workflow, then use the quieter links only if you need a comparison or a narrower Bit handoff.

Open the first task

Compare it with another assistant Ask Bit which jobs suit Claude