Claude 101: Three Useful Workflows

You have a meeting tomorrow. The notes are scattered across a slide deck, a few emails, and the part of your brain that keeps remembering one awkward conversation. Opening a blank document feels like work. Asking Claude to “help me prepare” feels like it should help. Usually it produces something cheerful and generic that still leaves you with a blank document.

Claude gets useful when it has a job, the material for that job, and a result you can judge. The first useful workflows are ordinary pieces of work already sitting on your desk, where you can tell whether the answer earned its place.

Clever questions can leave Claude looking impressive but hard to apply. A real task forces the question that matters: what would make this output useful enough to use?

Turn Source Material Into a Decision Memo

The first good use case is a document that has too much information and needs a clear conclusion. A planning brief, a vendor comparison, a meeting summary, a policy draft, an incident timeline. Claude is good at turning a defined pile of material into a defined artifact.

The files, reader, and decision give Claude an editor’s job. A vague request such as “summarize these” asks it to guess what matters.

Read the attached documents. I need a one-page memo for the leadership team
to decide whether to renew this vendor contract.

Include the recommendation, the three facts that support it, the cost of each
option, and the open questions. Use only the attached material. Mark a fact
as unknown when the documents do not answer it.

That request has a source boundary, an audience, a format, and a failure mode. The last line is important. It makes an unsupported answer visibly wrong instead of quietly plausible.

The first response is a working draft, judged by the person who owns the decision. A number needs a source. A missing commercial risk needs to be named. A premature recommendation should return the missing evidence. One revision against a real standard turns a vague chat into a real process.

Review a Plan Before It Becomes Work

Claude can inspect an approach before the first draft exists, especially when the cost of a missed assumption is high.

Suppose a team plans to replace a manual weekly report with an internal tool. The usual first prompt is “design the tool.” That moves too fast. The better first job is to find the assumptions hiding inside the request:

We spend four hours each Monday compiling this report. Read the current report
and the notes below. Before proposing a solution, list the decisions we have
not made, the data sources that need an owner, and the ways the result could
be wrong without anyone noticing.

Rank each item by the damage it could cause. Then propose the smallest first
version that would let us test the workflow.

This is a review. Claude can point at missing definitions, dependencies, and edge cases fast. You still decide which objections matter. The model supplies a broad first pass; a person with responsibility decides what to do with it.

The pattern fits code, policy, hiring plans, research plans, and operating procedures. The useful prompt names the proposed action, asks for the conditions that would make it fail, and leaves the decision with the person accountable for the result.

Keep the Assignment Small Enough to Judge

Beginning users often give Claude a whole department’s worth of work. Research the market, write the strategy, make the slides, and draft the email. The result may be long and polished. It is hard to know which part deserves trust, because the job never had a boundary.

One useful assignment has a finish line. A decision memo can be compared with the source material. A plan review can be checked against the actual constraints. A status report can be read by the person who needs the status. When the result misses, the correction has a home: the facts, the requested format, or the standard used to judge it.

The next prompt gets easier. The work becomes a small process with a visible output. Claude gets a clearer brief, and you get a repeatable way to tell whether it helped.

Turn Repeated Work Into a Project

One-off work belongs in a normal chat. Repeated work needs persistent context.

Claude Projects are for that second category. A project can hold files and instructions that Claude uses in chats inside the project. In Claude’s new memory experience, each Project has its own memory space and project summary. Claude can also search past chats within that Project. The experience is available to Free, Pro, and Max users; Team and Enterprise remain on the legacy experience during the rollout. Put facts and rules that must persist into project knowledge or project instructions either way.¹

A good first project is small. A weekly status-report project can include the reporting template, a strong past report, a short description of the audience, and a few rules that actually change the output. Its project instruction might say:

Write for the operations leadership team. Lead with decisions and risks.
Use the reporting template. Separate confirmed facts from estimates. Keep the
report under 600 words unless a material incident needs more explanation.

Each week, new notes are the only material needed for the report. Claude has the format and the standing context. You supply the information that changed.

Projects become less useful when they turn into a junk drawer. A folder of old chats, stray PDFs, and every policy anyone has ever written gives the model more material to sort through and less clarity about the job. The useful files define the job. The rest can leave when they stop helping.

A Useful First Session

An existing meeting is a clean subject for a first session.

The relevant source material goes into a chat. A decision memo gets named sections. The result gets checked for factual errors and missing stakes, then receives one specific revision. The finished memo can enter the meeting, or its failure can name the missing part of the assignment.

That sequence produces a result, a correction, and a standard for the next run. If the same meeting happens every week, it has earned a Project. If it happens once, the chat can end with the work.

The quality of the result will mostly follow the quality of the assignment. Claude’s own guidance makes the same point: clear instructions, relevant context, and a specified output shape improve consistency.² I wrote more about that in The English Trap. Language models will fill gaps with something plausible. Good work starts by deciding which gaps they are not allowed to fill.

The first session also includes a few minutes of attention after the response arrives. The output gets compared with its source material and the standard it was meant to meet. Its correction is part of the workflow. It gives the next request a better brief than a collection of generic prompt tips ever could.

Claude becomes useful through one real assignment with an answer you can inspect. That is enough to find out whether it can take work off your desk.

Footnotes

1. Anthropic, “Use Claude’s chat search and memory to build on previous context”. ↩

2. Anthropic, “Prompting best practices”. ↩