The Ladder Most PMs Don't Know They're On

What if your AI already knew your project?

Project managers have been using (or trying to use) GenAI for a few years now.

Status reports, meeting prep, risk summaries, funder updates.
It saves time. The outputs are decent. We keep using it.

But...is that beginning to bother you?

Do you find yourself saying:

"The outputs are fine. They're just never quite right."

Here's the thing: most folks are stuck on the first rung of a three-rung ladder.
And most don't know the ladder existed.

The Ladder

There are three distinct layers of GenAI value for project managers.

Most PMs discover Layer 1 and stay there. Not because it isn't useful, but because
Layers 2 and 3 are invisible until you know to look for them.

Each layer answers a different question. Each has a ceiling. And each ceiling (if
you're paying attention) is a signal pointing to the next layer up.

Layer 1: Prompts

This is where everyone starts.

You open a chat, type a request, get a response. You paste in meeting notes and ask
for a summary. You describe a risk and ask for mitigation options. You draft an email and ask AI to sharpen it.

This works. It saves real time. And the quality improves as you get better at asking.

A well-structured prompt (one that includes who you're writing for, what the context is, what format you need) produces noticeably better output than a vague one. Most PMs who invest in learning to prompt see immediate improvement.

So what's the ceiling?

Every prompt is a one-shot interaction. Without a project context document, the AI knows nothing about your project, your stakeholders, or what good looks like in your specific context. You are rebuilding its understanding of your world every single time you start a new chat.

You end up re-explaining the same funder sensitivities in every
prompt. Describing the same project phase. Setting the same tone requirements. Over and over.

The output was always "fine" because you are always starting from scratch.

The signal that you've hit the Layer 1 ceiling:

You start saving prompts — copying them into a doc to reuse them. You find yourself saying "the output is close but I always have to fix the tone." And at some point, you ask the question that matters most: Can it remember?

That question is the invitation to Layer 2.

Layer 2: Context Engineering

The shift from Layer 1 to Layer 2 is a single realization: the AI isn't failing
because your prompts aren't good enough. It's failing because it doesn't know
your project.

Context engineering is the work of fixing that: once.

Instead of explaining your project in every prompt, you build a persistent brief.
A living document that captures everything the AI needs to be useful:
the goal and purpose, the stakeholder landscape, the funder sensitivities, the tone
this project uses, the assumptions currently in play.

Here's what the difference looks like in practice.

A PM on a health project needs to draft a quarterly update for her funder lead.
On Layer 1, she pastes in the request and explains: "This is for James, who values
data over narrative and is currently concerned about indicator lag. Keep it brief
and lead with numbers."

She types that same explanation, some version of it, every time she communicates
about James.

On Layer 2, her context document has a stakeholder index. James is already there:
Values data over narrative, prefers brevity, currently concerned about indicator lag.

She loads that block once. Every interaction with James-related communications starts informed.

She stopped explaining. She started directing.

The compounding effect is real. A project at month eighteen with a well-maintained context document has accumulated institutional knowledge that makes every AI interaction faster and more accurate. The PM who has been prompting ad hoc for eighteen months is still starting from zero.

A few things belong in every PM context document:

• A project identity block: name, funder, phase, primary goal, current constraint
• A stakeholder index: who matters, what they care about, their communication
  preferences, current relationship temperature
• Live assumptions and risks: what's true today that might not be tomorrow
• A tone guide: three to five sentences on how this project communicates

The discipline is keeping it current. Not a monthly review — a simple trigger:
after any significant meeting or decision, update the one block that changed.
Two minutes, not twenty.

The signal that you've hit the Layer 2 ceiling:

You have good context and your AI interactions are consistently better. But you're
still doing the same five-step sequence manually every reporting cycle. The PM does step one, then step two, then step three — AI assists at each step, but the chain doesn't connect.

That's the ceiling. And it points to Layer 3.

Layer 3: Agents and Workflows

This is where the mental model changes most significantly.

Layers 1 and 2 are still about the PM doing the work with better tools. The PM
prompts, the AI responds, the PM judges and acts.

Layer 3 is different. The PM defines the work. The system does the sequencing.
The PM intervenes at judgment moments; not at every step.

A concrete example: a PM receives a stakeholder interview transcript. Today, they
read it, extract key themes, identify assumptions, map risks, and draft follow-up
questions. Four steps, done manually, in sequence.

A workflow version of that looks like this:

Transcript goes in. The system extracts key themes. Maps them to project assumptions. Flags assumptions at risk. Drafts clarifying questions for the next meeting. Output arrives for PM review.

The PM's job becomes reviewing the output, adjusting what's wrong, and deciding what to do next.

The cognitive work still happens. The AI does the mechanical sequencing. The PM
applies judgment where judgment matters.

PM work is judgment work: reading ambiguity, navigating relationships, making calls under uncertainty. No workflow replaces that.

What workflows replace is the repetitive mechanical sequencing that surrounds judgment.
Which frees the PM to operate at the level their judgment deserves.

The Through-Line

Look at the three questions each layer answers:

• Layer 1: What do I want right now?
• Layer 2: Who am I, what's my project, and what does good look like here?
• Layer 3: How does this work flow, and where does my judgment belong?

Most PMs are spending all of their energy on the first question. A few have moved to the second. Almost none are asking the third.

The ladder doesn't require jumping. It builds naturally: each layer creates the
appetite for the next. The frustration at Layer 1 is the signal.
The repetitive friction at Layer 2 is the invitation.

Ask the question that matters: What if it already knew?

Where are you on the ladder right now?

Get Intentional,
Paul

P.S. If you've been prompting for a while and the outputs still feel generic, it
almost certainly isn't your prompts. Contact me with what you're working on.