Executives want more speed, less bureaucracy. Project artifacts are an easy target. But is “do fewer of them” really the right solution? AI is fast changing the shape of this debate in a way I’m not sure PMO leaders are thinking about yet. (And to be fair…this is a recent “ah ha” for me too.)
The backlash against artifacts
“Fewer artifacts. More action.” That’s more or less the battle cry I’m hearing across industries right now. And to be honest, it’s overdue.
Like many of you, I’ve worked with companies for whom approving a Charter was a 6-7 week affair…just to become a 13-page document that no one ever read again. It was brutal.
Organizations are absolutely right to push back, although to be honest, I don’t know why it took so long.
But I’ll tell you something else: I think the conversation about how many artifacts to have, how complex they should be, which ones matter, and which ones don’t is quickly becoming an irrelevant discussion.
The first reason is that document-building is only half the hard part. You also have the conversations leading to the document. The back-and-forth where a team defines what they’re actually doing, finds the contradictions, and gets specific about vision, requirements, dependencies, and constraints.
You may be able to cut the wordsmithing and approval cycles. But you can’t trim the conversation without consequences.
But the second reason the discussion is becoming irrelevant is this:
We’re not tied to static artifacts anymore. In fact, that mindset is about to become a liability.
What we need are AI-queryable knowledge bases—AI-legible, clean information troves that PMs can use to mine analysis, catch contradictions, and build hyper-tailored outputs that give each audience exactly what they need. A dashboard for one meeting. A presentation for another. A navigable FAQ for a third. All built in minutes from the same source. That’s what AI unlocks — but it means the primary reader of your project information won’t be a human. It’ll be AI.
And that changes the whole artifact conversation from “What should our templates be?” or “How many should we have?” to something else entirely:
How do we build an information ecosystem that’s useful to AI…so AI can be useful to humans?
The artifact is dead. Long live the knowledge base.
Let me explain what I mean by this, because I’m not saying “stop documenting things.” I’m saying the container is changing.
Think about a project charter the way most of us have experienced it. Somebody writes a Word doc. It gets routed for review. Four people redlined it. A fifth person had opinions about the font.
Eventually, it’s approved, uploaded to SharePoint, and never opened again. The charter did its job (theoretically), but the information inside it is now locked in a format that’s hard to search, impossible to cross-reference, and useless to any AI tool you point at it.
If you’re thinking, “but AI can read Word docs,” you’re correct. It can.
The issue lies in how it reads them. AI is, to be blunt, lazy.
Give it a 13-page charter, and it will skim rather than read thoroughly. It will look at file names and decide if something is relevant before checking the content. It will fill in gaps with assumptions you didn’t ask for, like which version of a document is the most current, or which requirements still apply after a scope change. It will confidently summarize something it barely grasped. The information might be there, but the format makes it nearly impossible for AI to extract reliably.
Now consider what you truly want from your project information. You want to ask questions like:
- “What requirements does this new change request affect?”
- “Which risks are linked to this vendor?”
- “How has our scope changed since the original assumptions?”
- “What did the team decide about X, and has anything happened that should cause us to revisit that decision?”
These are analytical questions. They need organized, clear, searchable information. A collection of Word documents and PowerPoints can’t provide that. A knowledge base can.
Here’s the shift I’m describing:
Instead of a finished document that gets approved once and filed, imagine two layers:
The back end: a structured, AI-legible knowledge base. Documents become markdown files with clear hierarchies. You’d use databases for anything you’d ever want to filter or cross-reference (requirements, risks, decisions, assumptions). You update it continuously, and you curate it carefully. You don’t throw every meeting transcript you’ve ever had and ask AI to manage that context window and make good decisions about what is important.
The front end: whatever output the moment requires. AI pulls from that same knowledge base and produces a variety of hyper-tailored outputs in minutes, provided a human provides the judgment to identify what should be built, for whom, in what format, and with what parameters. Envision a custom dashboard for a meeting on project performance and another for a meeting on vendor performance. A one-page summary for the sponsor, a navigable HTML webpage to walk business users through requirements.
Same pantry, different meals depending on who’s coming to dinner.
That’s a different model than “write a charter, get it approved, move on.” It means the information is living, not static. It means formats are disposable (because you can regenerate them in minutes).
And it means standardization—the thing PMOs have spent decades chasing—might matter a lot less than it used to. If AI can produce ultra-tailored outputs from well-structured data, enforcing a single template for every project and every audience starts to look less like discipline and more like a constraint.
One important caveat: this only works in one direction. AI can take structured, explicit information and turn it into polished, human-ready output. It can’t reliably go backward.
Hand it a PowerPoint full of persuasive, abstract language, and it’ll skim. It’ll summarize. It’ll confidently tell you something that isn’t quite right. Organizations that rushed to deploy tools like Microsoft Copilot are learning this the hard way—decades of messy SharePoint sites (duplicate files, stale documents, inconsistent naming) make AI output wildly unreliable. The tool isn’t the problem. The information underneath it is.
What this changes for PMOs and Project Managers
The real shift isn’t just technological. If the knowledge base becomes the new center of gravity, the project manager’s job is no longer to produce documents. It’s to produce meaning.
That entails facilitating conversations that build clarity, curating the information AI needs to work with, and exercising judgment to determine which outputs are needed, for whom, and whether AI got it right.
That’s a different skill set than most PMs have been trained for.
Here’s what it looks like in practice:
- Say what you mean. Persuasive, political, or abstract language is a liability in an AI-legible knowledge base. “Commitment to best-in-class operations” tells AI nothing useful. “Reduce the time it takes for monthly close from 8 days to under 5” gives it something to work with. AI doesn’t care about your corporate template or your publicly consumable verbiage. But if you want better AI output, neither should you. This is a real shift for people who’ve spent their careers writing for executive approval. Writing for AI means being specific, being direct, and dropping the language that sounds good but says nothing.
- Think in databases, not documents. PMs already do this for some things (risk registers, issue logs). But consider how much project information is still trapped in a narrative format that would be far more useful as structured data. Requirements. Assumptions. Stakeholder commitments. Decision history. Every one of these is something you might want to filter, cross-reference, or query—and none of them work well as paragraphs in a Word doc or bullet points in a PowerPoint. The shift is from “write it up” to “structure it so it’s queryable.”
- Formats will become ultra-tailored. This goes back to the standardization point. When AI can produce any format from the same structured source, the old debate about whether your status report should be a one-page PowerPoint or a two-page Word doc becomes irrelevant. You build both. Or neither. Or something completely different — a live dashboard, an interactive FAQ, a brief tailored to the specific questions your sponsor asked last week. The format serves the audience and the moment, not the template library.
- Context management is the skill that matters most. This is the one I think most people are underestimating. AI doesn’t just need clean data. It needs someone who understands what’s important enough to tell AI about. Someone who checks whether AI drew on the right sources, caught the right dependencies, or made a logical leap that doesn’t hold up. Managing project knowledge may seem like a job that will get delegated to AI. And some companies are betting (short-sightedly) that it will. But until agents can operate (a lot) more cheaply and effectively, and until they can operate with the same sense of stakes and accountability that humans can, context management will be an incredibly important human responsibility.
And speaking of context management, here’s something else I don’t think companies have thought through. PMs also must hold context that doesn’t belong in a database.
Like, for instance, the conversations that won’t happen in a meeting anymore, because everyone knows AI will record it, transcribe it, parse it, and post it. PMs will have to understand the real weight of the unspoken dynamics and integrate that perspective into their recommendations. A well-stocked pantry doesn’t cook dinner by itself. Someone still has to know what to make, who it’s for, and what’s missing from the recipe. That’s the PM.
The job is no longer about document production and standardized reporting. It’s about building the context, clarity, and AI-legibility needed to push real work forward.
If You Only Do One Thing
Run a context audit on one active project. Take your existing project information—whatever you’ve got, whether it’s a charter, a status deck, a requirements doc, or a shared drive full of meeting notes—and ask an AI tool to answer specific questions about the project using only that information.
How accurate are the answers? How useful? How much correcting and filling in are you doing?
If the answer is “a lot,” you aren’t yet poised to take advantage of AI’s best capabilities. And that matters for PMOs who want to leverage AI’s frontier capabilities.