Skip to main content
Applied AI for Management

AI Status Reports That Work One Folder, One Project, One Report

A project manager pointed AI at every project, SharePoint site and inbox he had, and got a status report nobody could trust. The fix is five steps in the right order: one folder, clear names, one report for one project, a skill, and an update after every use. Then the same big request works.

Untidy stacks of paper surround one neat open folder on an office desk, while a soft beam of light distils its pages into a single short report

"Find every record about my projects and build me a status report." A project manager gave his AI that single request, pointed it at every project, SharePoint site and email thread he had, and got back something with the shape of a status report and none of the reliability. The request was never the problem: it was the last step of a five-step method, asked first, and this article walks through all five in the order that works.

Why the everything-request comes back confident and wrong

Pattern-matching and following exact instructions are what AI does best. Working out what you meant but never said is where it is weak. Point it at everything and it has to guess all the time: which of three schedule files is current, which email thread belongs to which project, whether "status" means the sponsor's one-page summary or the team's task list.

Each guess is a spot where the report can go wrong and still read smoothly, which is exactly how his came to look so complete. A capable tool still has no access to what is in your head, and the accuracy of what comes out depends on how tightly you control what goes in.

Picture a sharp new hire on day one, handed the keys to every shared drive and told to have Friday's report ready. You would not get a good report. You would get a confident one.

Quality out starts with specific, structured input, and the first two steps exist to build it.

Steps 1 and 2: one folder, labelled for a new colleague

The first step is almost boring. Collect every document this one report needs into one folder, for one project. Leave out the rest of SharePoint and leave out your inbox; if the report does not need it, it does not go in.

The second step is organising and naming what you collected. Give the folders proper names, file each document where it belongs, and rename files so the name says what they are. Do it the way you would before a new colleague starts, so they could find their way around without asking you a single question.

A layout like this is enough for a typical status report. Adapt the names to your project:

  • 01 Plan: the current schedule baseline and the scope statement
  • 02 This period: progress notes, the risk and issue log, decisions taken
  • 03 Past reports: the last two status reports you sent
  • 04 Format: one report your sponsor liked, as the model to follow

File names carry as much weight as folder names. "notes.docx" tells a reader nothing, while "Steering committee minutes 2026-11-04.docx" says what it is and how recent it is. Labelling feels like admin, and it has a large effect on how quickly you reach a report you trust.

One folder has a second benefit. You know exactly what the AI has seen, so you can check that every document in it is one you are allowed to share with the AI tool you use.

Step 3: one report for one project, and expect a mess

Choose a single project and request the single report you need from it. Treat the first draft as the start of a conversation: point out what is wrong, argue with it, and stay in that same chat until the report is one you would send. An opening request can be this plain:

Using only the documents in this folder, write this period's status report for [project name]. Follow the format of the report in 04 Format. Where the folder is missing something the report needs, say so instead of filling the gap.

The corrections are where the real work happens. Yours might read like this:

  • "The schedule section uses the old baseline. The current one is the re-plan in 01 Plan."
  • "Risk 4 was closed at the last steering meeting. Check the minutes in 02 This period."
  • "Move decisions needed to the top. The sponsor reads the first three lines."

It will be messy, and that is the process rather than a failure. We took apart a summary that reads well and is still wrong, line by line, in a real AI sprint summary and the four things it got wrong.

Somebody will tell you to learn prompt engineering and context engineering first. Both are useful, and both shorten the road once you know them. Skip the rulebook for now: a long list of rules is a good way to stall before you start, and the cost of not knowing them is only a few more rounds of conversation.

Step 4: turn the conversation into a skill before you close it

When the report finally looks right, tell the AI what worked and ask it to keep it:

Here is what I liked. Look at our conversation, turn it into a skill, and save it as a markdown document.

A skill is a written set of instructions the AI reads before it starts a task. In Claude, skills are folders of instructions that Claude loads when a task calls for them, and anyone can write one in Markdown, as Anthropic's help article on skills explains. You do not write this one by hand: the AI drafts it from the conversation you just had, and next time you give it the skill before you start.

Read the file once before you save it. It should name the folder it reads, the sections of the report and every correction you made along the way. If one of your corrections is missing, ask for it to be added, because a correction left out of the skill is one you will be making again next week.

Expect the first skill to be rough, and the report it produces to be rough too. Keep going. Each round leaves the skill a little better. To see what a finished status report skill holds, we opened up a public one in what a status report skill actually contains.

Step 5: update the skill after every use, then ask for every project

Now do the same report for project number two. Load the skill first, and give that project its own folder laid out like the first one. Before you close the chat, ask:

Summarise the lessons learned from this conversation and update the skill with them.

Keep that loop going. When one or two projects come out right, go back to the request that started all this, now phrased as "Produce this report for all my projects." The ask has not changed since it failed. What changed is that the AI now has a worked example of a good report and the instructions that produced it.

What you hold at this point is small and easy to keep up: the skill, the folders it reads, and the habit of feeding each lesson back in. The following week, pick another report and build its skill the same way. A system is that, repeated: pieces that each work before the next one starts.

Why big AI projects stall, and the fix you already use

We see this pattern at work and among friends. A big ambition goes to the AI in one go, the results disappoint, and the whole idea is shelved. The missing piece is a first goal small enough to be realistic and clear enough to measure: one status report, for one project, that you would send as it stands.

Project managers solve this problem for a living. No team gets a whole project dropped on it at once: agile teams get stories, and traditional projects get work packages. Decompose the AI work the same way.

Keep the finished system in view, the one this project manager tried to get from a single request, and move toward it one small, working step at a time. Whenever a session produces something you can rely on, ask the AI to capture it as a template or a skill.

Pick the first report with three questions

The second trap is impatience: every report, automated, this month. It is a natural thing to want and rarely a realistic one. Be deliberate about the first one, because it is the one your colleagues will judge the whole idea by.

QuestionWhat it tells you
Which report is the most repetitive chore, the one you would happily never write again?Where you get your time and patience back first
Which one matters most to your project, or to your career?Where the result counts beyond your own week
Which result could you put in front of your manager and your team?Whether other people will see the value

List the reports you produce in a normal month and mark each one against the three questions. Start with the one that answers the most of them.

Weight the third question more heavily than it looks like it deserves. A report people can watch working makes the case on its own. Colleagues who see it start offering to improve the skills, and before long you are sharing skills across the team instead of building alone.

The same request, in the right order

The AI was never stupid. It was asked to read a project manager's mind across every project at once, with nothing to show it what good looked like.

What changes is the unit you think in. Drop single tasks, single chats and clever prompts, and think in systems, which is how managers, project managers and Scrum Masters already organise people and work every day. That is the core skill of the job, and it carries over to AI unchanged.

This week, make one folder for one project and ask for one report.

For a guided route from there, our course AI Systems with Claude works through this kind of build against one running project, and two of the five systems you build in it handle the sprint report and the stakeholder update. Each system is yours to keep.

AI Systems with Claude, for Scrum Masters and Project Managers

Most AI pilots in project management do not fail on the model. They fail because somebody pointed the thing at a decision instead of at a task. This course is built around that distinction, and around the work that follows once you get it right.

Eight teaching sections, taught against one running project from the first lecture to the last. You watch a system get built, then you build the same one against your own sprint. Nothing here is a demo that works only on the example.

You finish with five working systems and you keep them. The sprint report machine turns your board export and standup notes into the report you currently write by hand. The retro intelligence system tells you what the team keeps saying, not just what it said this time. The stakeholder comms engine drafts the update in the register the audience expects. The backlog health monitor flags the quiet decay nobody has time to check for. The risk and dependency tracker follows the chains that actually bite.

Seventy-six working files ship with it, across four sections, so every system is built against real sprint data rather than material invented for a slide. Four sprints of retrospectives, board exports, standup notes, backlog and risk data.

Explore the Course


Build the Sprint Reporting System, Not Another Prompt

AI Systems with Claude is our own course for Scrum Masters and Agile Project Managers, taught against one running project across eight teaching sections. You build five working systems and keep them: the sprint report machine, the retro intelligence system, the stakeholder comms engine, the backlog health monitor, and the risk and dependency tracker. Direct from HK School of Management and Technology, not on Udemy.

See What Is Included