The Smartest AI Workflow I Have Ever Seen Ran on Three Pages of Prompt
Project managers are quietly building their own AI chief of staff. The duct tape is the interesting part.
A few weeks ago I was talking with a project manager who runs large industrial projects. Real ones, with safety officers and subcontractors and go-live dates that cost serious money when they slip. Somewhere in the conversation he mentioned, almost apologetically, a side project of his.
Every week, he feeds an AI model his project charter, the project plan, the risk register, the action tracker, and the last six weeks of status reports. Then he adds the current week's meeting notes and any relevant emails. On top of all that sits a prompt he has iterated on for months. It covers three A4 pages in font size 10.
Out the other end comes a list of specific open topics he needs to chase down before writing his end-of-week status report. He has a second prompt that helps him prepare sharp questions for the weekly team meeting. A third one, about 200 lines, assembles everything and drafts the status report itself. He even runs scenario checks: the safety officer found discrepancies during vehicle inspections, the subcontractor says compliance takes two extra weeks, does this move the critical path and the go-live date?
He called it manual and clunky. I think it is one of the most sophisticated AI workflows I have ever seen a working professional build, in any field. And I have been building software for a long time.
But he was right about the clunky part. And the reason it is clunky tells you almost everything about where AI in project work is actually stuck.
The analysis was never the hard part
Here is the thing he said that stuck with me, close to verbatim:
The AI is good at analysing lots of text sources. The challenge is to obtain all the information, and the effort to write it down comprehensively.
Read that again. The intelligence is not the bottleneck. The bottleneck is assembly.
Every single week, this PM performs the same ritual: hunt down the current version of the plan, the register, the tracker, the reports. Export them, copy them, paste them into a chat window. Rebuild the entire context of his project from scratch, by hand, so a model can spend thirty seconds being brilliant about it.
Then the chat window forgets everything, and next week he does it all again.
This is not a niche problem. Once you know the shape of it, you see it everywhere. People pasting spreadsheet exports into chat windows. People maintaining personal prompt libraries that are really just descriptions of where their data lives and what it means. People whose "AI workflow" is 10 percent asking questions and 90 percent schlepping context.
The three-page prompt is not a prompt. It is a hand-written database schema, refreshed weekly, by a human, for free.
The data was structured before you flattened it
What makes this genuinely absurd is that the information was structured to begin with.
A project plan is rows with dependencies, dates, and constraints. A risk register is rows with owners, severities, and statuses. An action tracker is rows with assignees and due dates. This is about as structured as data gets. Then, to make it legible to an AI, we flatten it into prose, paste it into a text box, and ask the model to mentally reconstruct the table we just destroyed.
We are converting databases into essays so a machine can convert them back.
The fix is not a better prompt. The fix is to stop moving the data to the AI and start running the AI where the data already lives.
What that looks like in practice
This is the bet I am making with Wisegrid, the work management platform I am building. Grids, projects, dependencies, dashboards, the tools PMs already know. The difference is that AI is not a panel bolted onto the side. It is inside the formula engine, the same place your SUMs and lookups live.
That means a formula cell can think. A few real examples:
=CLASSIFY([Risk Description]@row, "high, medium, low")
Drag it down a column and every risk in the register gets triaged. New row arrives, it gets classified too. No export, no paste, no three pages of context, because the AI is reading the same cells you are.
=EXTRACT([Meeting Notes]@row, "committed follow-up actions")
Your meeting notes column quietly becomes an action tracker.
=SUMMARIZE([Weekly Update]@row)
Six weeks of status history, condensed in place, always current.
The part I care about most is that these results are just values. They compose. You can wrap one in a condition:
=IF(CLASSIFY([Feedback]@row, "urgent, normal") = "urgent", "Escalate", "")
You can COUNTIF over a classified column. You can feed an extracted value into a lookup, a rollup, a dashboard. The AI output becomes data that the rest of your sheet builds on, instead of a paragraph trapped in a chat transcript that nobody will read twice.
And because the context problem is solved structurally, the weekly ritual disappears. "What changed this week, what has gone stale, what needs chasing before Friday's report" stops being an assembly job. The register knows its own stale rows.
Where this goes next
The PM's scenario question is the north star: subcontractor slips two weeks, does go-live move? Today he answers that by pasting his plan into a chat window and hoping the model reconstructs his dependency chain correctly. The right answer comes from evaluating the actual chain, the real "start no earlier than" constraints, the real critical path, and that is only possible when the AI and the plan live in the same system. That is the direction everything above is pointed.
Two things I decided early that I think matter:
AI usage is metered in plain dollars, with a hard ceiling. You see exactly what each call costs, in currency, not in mystery tokens. Most cells cost a fraction of a cent. When you hit your ceiling, things stop cleanly instead of surprising you on an invoice. I built this after watching too many products hide AI economics behind vibes.
Nothing writes without you. Formulas resolve to values you can see and audit. The built-in co-pilot that assembles sheets and automations for you does it while you watch. An AI that silently edits a project plan is how you get a confident, wrong go-live date.
The takeaway, even if you never use my product
If you are doing the three-page-prompt ritual today, you have already done the hard part. You have specified, precisely, what an AI chief of staff for your projects should do. You are just paying the assembly tax weekly to run it.
Stop optimizing the prompt. Move the data. Whatever tool you choose, the AI that wins for project work will not be the one with the best chat window. It will be the one that never needs the context pasted in, because it was standing next to the data all along.
If you want to see what that feels like, Wisegrid is at wisegrid.co. You can try it free for 7 days, import your existing sheets, and point a CLASSIFY at your real risk register. The first time a column triages itself, the three pages of prompt start to feel like a fax machine.
I am the founder and I answer my own email: ryan@wisegrid.co. If you have built one of these duct-tape AI workflows, I genuinely want to hear about it. The best product ideas I have are coming from people like the PM with the three-page prompt.
Top comments (0)