AI Workflow With Claude: From Brief to Reviewed Result

A Claude AI workflow is a defined sequence that starts with a written brief, moves through a set of prompts, and ends with a person reviewing the result before it is used. Claude holds context well within a single conversation, but it does not carry memory into the next one, so your brief, constraints, and approved decisions need to be ready at the start of each session. Tools like fenbs keep that context outside any one conversation, so your Claude work is repeatable across sessions and accessible to other AI apps when you switch.

What to prepare before you brief Claude

Claude produces more useful output when your brief is explicit about what the task is, who it is for, what format the result should take, and what constraints apply. A brief that takes two minutes to write usually avoids two rounds of revisions.

• Source material or information Claude will work from.
• The task and the desired output format, including length if it matters.
• Constraints: tone, audience, things to avoid, and required structure.
• Review criteria so you can check the result against a standard, not an impression.

Claude Code reads AGENTS.md and CLAUDE.md files in a repository automatically at the start of a session. For development work, those files are the natural place for your brief and standing rules. For writing, analysis, or research work in Claude.ai, a short context block pasted at the top of each conversation serves the same purpose.

How to run a Claude workflow

The steps below apply to writing, analysis, research, and most single-task workflows. Adjust the review criteria and source material for your situation.

1. Open a new conversation and paste your context: the task, constraints, and output format.
2. Provide the source material in the same message or the next. Claude performs better with an explicit source than when it fills gaps from its training data.
3. Request the output in the format you specified. If the format is complex, give a short example of what the structure should look like.
4. Review the result against your criteria. Mark what to keep, what to revise, and what to ask again.
5. Give specific revision feedback. "The second paragraph assumes background the reader does not have" is more useful than "make it clearer."
6. When the output meets your criteria, save the prompt that worked, the source format it expected, and any constraint that made the difference.

What Claude reads at the start of a session

Claude does not carry memory between separate conversations. Each new session starts from zero. If your brief, the decisions, and the constraints live only in a previous chat, the next session either rediscovers them or ignores them. That inconsistency is the most common reason a Claude workflow that worked once does not produce the same quality the next time.

Paste your brief at the start of each new conversation, or keep it in a document you copy from. For a team, a shared location prevents different people from running the same task with different briefs. Claude Code Projects added shared memory in late September 2026, which keeps context available across conversations within a project; check the current feature set at the Anthropic help pages.

The claude-code-task-tracking-workflow guide covers how to connect Claude Code to a task system so that progress is recorded outside the conversation. The what is Claude AI page explains the difference between Claude, Claude Code, and Claude Projects for readers who are choosing between them.

Where a Claude workflow breaks down

The most common failure is a well-written brief that still omits a key constraint. Claude will make a reasonable assumption to fill the gap, and the output will look correct until someone checks it against what the original instructions forgot to say.

• Omitted constraints: any constraint not in your brief is a gap Claude fills with a reasonable assumption.
• Skipping review: a workflow without a review step means errors reach the next stage undetected.
• Not saving the working prompt: a prompt that produced the right result this time is lost if it stays in the chat.
• Mixing Claude Code and Claude.ai context: they read from different places; keep your context consistent between them.

Saving prompts for a repeatable Claude routine

A reusable Claude prompt is specific enough to produce consistent output and general enough to work with different inputs of the same type. Write it with placeholders for the variable parts: the document, the audience, the specific constraint. Test it on three examples before treating it as stable.

Save each tested prompt with a note about the input format it expects and the model version it was tested with. The ai-agent-approval-workflow guide covers how to build the review step that checks the output before it moves forward. For any task that draws on sources, always include a check against the primary source, not only against the output.

How to do this with fenbs

Set up a project for the Claude workflow you want to make repeatable. Record your brief and constraints as context notes. Add tested prompts to Skills and prompts. Connect Claude Code from the Integrations page and it reads the project context natively at the start of each session.

1. Create a free account. Add a project for the workflow and write your brief as the project description.
2. Record agreed decisions and constraints as Decisions and rules. Connected assistants read rules first.
3. Add tested prompts to Skills and prompts. Note the input format and the model version they were tested against.
4. Connect Claude Code from Integrations. Ask it to read the project context at the start of each session.
5. At the end of each session, save progress: what was completed, what remains, and the next step.