The Four Claude Features That Change How Much Work It Does for You

The pattern behind the four

Watch someone who gets ordinary results from Claude and the same thing happens every session. They paste background, copy data in from another app, describe how they want the work done, and come back tomorrow to do it again. None of that is thinking. It is setup, repeated.

The four features that a recent tutorial by Ali H. Salem focuses on each remove one of those repetitions. Projects remove re-explaining context. Connectors remove copying between apps. Skills remove re-describing a procedure. Scheduled tasks remove the need to show up and start the work. Looked at that way, they are not four separate tricks. They are a sequence, and the order matters.

1. Projects: stop re-explaining your situation

A project is a workspace with its own chat history, a knowledge base of uploaded files and text, and project instructions that shape every chat inside it. Claude's help center says Projects are open to all users, with free accounts limited to five, and that on paid plans Claude can switch project knowledge into a retrieval mode as it nears context limits, which it says can expand capacity by up to 10x. Team and Enterprise plans can share projects with view or edit access.

Use one project per ongoing body of work: a client, a product, a research topic. Put the stable material in the knowledge base and the standing rules in the instructions. Memory, which Anthropic extended to free users in March 2026 according to 9to5Mac and Bloomberg, complements this by carrying personal preferences across chats, but it is not a substitute for a curated knowledge base.

The failure mode is dumping everything in. A knowledge base full of outdated drafts gives confident answers from the wrong version. Curate it like documentation.

2. Connectors: let Claude reach the work

Connectors let Claude read from and act in other services, such as email, calendars, document stores and business systems. They are built on the Model Context Protocol. Anthropic maintains a directory of reviewed connectors, and custom connectors can point at any remote MCP server by URL. The help center lists custom connectors on Free, Pro, Max, Team and Enterprise, with free users limited to one; on Team and Enterprise, owners add them for the organization.

The value is obvious: Claude works on the actual inbox or document instead of whatever you remembered to paste. The risk is equally plain, and Anthropic states it: once connected, Claude can take actions in those services, and custom connectors may not be verified by Anthropic. Connect only what the work needs, prefer read access where it exists, and watch for instructions hidden in content Claude reads, such as an email asking it to forward something.

3. Skills: write the procedure down once

A skill is a folder of instructions, and optionally scripts and templates, for a recurring task: how your team formats a report, the checks a contract review must cover, the structure of a weekly update. Claude loads a skill when a task matches its description, so a skill costs nothing on unrelated work. The help center lists Skills for Pro, Max, Team and Enterprise, and says the feature requires code execution to be enabled.

The common mistake is writing the description as a summary instead of a condition. "Helps with reports" loads unpredictably. "Use when drafting the monthly board report from the finance export" loads when it should. Keep facts that change, such as prices, owners and current figures, out of the skill and in the project or a connected source, so the procedure does not go stale.

4. Scheduled tasks: work that starts without you

Scheduled tasks run a saved prompt on a cadence: hourly, daily, weekly, on weekdays, or on demand. Claude's help center lists them on paid plans, with Team and Enterprise in beta, created either by describing the task to Claude or by setting a name, prompt, approval mode, frequency and an optional model and folder manually. Cowork has since been folded into Claude, so expect the naming to shift. The current article says tasks run remotely, so they continue while your computer is asleep, but tasks that need local files or apps only run locally.

This is the feature that feels most like having an assistant, and the one that most punishes skipping the earlier steps. A daily briefing built on a stale project, a sloppy connector scope and no skill produces a confident, wrong summary every morning, unattended. Start with read-and-report tasks, review the first week of output, and use approval mode for anything that sends, edits or deletes.

A week to set it up

Day one, create a project for your main body of work and curate its knowledge. Day two, connect the one service you copy from most, read-only if possible. Day three, turn the instruction you type most often into a skill with a precise trigger. Run all three by hand for a few days. Only then schedule the task, starting with something you read rather than something Claude sends.

If your source material is large and keeps changing, consider keeping it in a retrieval index Claude reaches through a connector, rather than uploading copies that drift out of date.