How to Collaborate With AI on Real Work

Collaborating with AI means more than asking it questions and accepting the answers. Useful collaboration has a structure: you define what you are trying to accomplish, the AI contributes a draft or analysis, you review and correct it, and the result is recorded somewhere both of you can use next time. Tools like fenbs give people and their AI assistants one shared record—projects, tasks, Decisions and rules, and Where we left off—so each session continues from where the last one stopped rather than starting over from a blank prompt.

This guide is for anyone using AI on a real project: a designer working with Claude, a developer using Cursor, a consultant using ChatGPT, or a team using multiple AI tools across the same body of work. The principles are the same. You can read more about using AI at work for related workflows.

Divide responsibilities clearly before you start

Before the first prompt, decide what you are asking the AI to do and what you will do yourself. Good AI tasks are ones where the AI can produce a first draft that you review: a written section, a list of options, a summary of a document, or a first pass at code. Bad AI tasks are ones where the AI cannot tell what right looks like—aesthetic judgments, ethical calls, or decisions that require knowledge of internal context the AI does not have.

Write down the division of labor at the start of the project, not in your head. "AI drafts, person reviews and approves" is a workable rule. "AI decides" is not. Recording that rule means it is available to every session—and every team member—rather than existing as an implicit assumption that changes when someone new joins.

How to give an AI the right context

An AI assistant with no context about your project will produce generic output. Giving it the goal, the constraints, the decisions already made, and the current state of the work produces something useful. You should not have to re-explain the whole project at the start of every session.

Keep the important context—project goals, standing rules, current tasks, and Where we left off—in a place the assistant can read automatically. That is the practical value of a tool like fenbs: at the start of a session, the assistant reads the project context including what has been decided and what is still in progress, rather than receiving only what you remember to type.

How to review AI output well

Reviewing AI output is a skill. A quick read that ends in "looks good" is not a review; it is an approval based on nothing. A useful review asks: does this answer the request? Is every factual claim checkable? Does it follow the rules the project has already established? Are there any gaps or assumptions that need flagging before this moves forward?

Budget time for the review in your schedule. If you are treating AI output as a final draft from the moment it arrives, you are not collaborating—you are outsourcing without oversight. The AI is a first-draft contributor; you are the editor, the decision-maker, and the person responsible for the result.

How to record decisions from AI-assisted work

Every time you accept a direction, change an approach, or establish a constraint, that is a decision. If it only exists in a chat window, it will need to be re-explained the next time someone works on the same project. Recording decisions outside the chat is what makes collaborative AI work compound over time rather than reset with every session.

A decision record should say what was chosen, why it was chosen, and who made it. The "who" matters because people decide; AI assistants can record and propose, but the approval stays with a person. That distinction is important when something goes wrong and you need to trace where a direction came from.

What good AI collaboration looks like in practice

Say a three-person content team uses AI to draft articles. At the start of each session, the AI reads the editorial guidelines, the current article brief, and the lessons from previous issues. It drafts a structure. An editor reviews and adjusts the headings. The AI drafts the sections. The editor reviews each one, corrects errors, and adds judgments the AI missed. The final article is recorded as done, and the review notes become a lesson for the next issue.

That loop—context in, draft out, review, record—works for writing, research, analysis, and code. The key ingredient is the shared context that makes each session faster than the last. You can find the AI employees guide useful for thinking about what AI assistants are well-suited to contribute.

How to do this with fenbs

1. Create a free account and add a project for the work you are doing with AI.
2. Record the division of responsibilities as Decisions and rules. Include what the AI drafts, what a person must review, and what requires human sign-off.
3. Add the current tasks to the board. Give each task a plan that includes what the AI will contribute and how you will check it.
4. Connect your AI app through Integrations. At the start of each session, the assistant reads the rules, current tasks, and Where we left off.
5. After each session, save progress: what was completed, what was checked, what remains, and what the next session should start from.

fenbs works alongside whatever AI app you use—Claude, ChatGPT, Cursor, Codex CLI, or others. It does not replace the AI app or automate the work. It holds the shared record that makes each session with any AI app more productive than the last.