Context Engineering - The Most Important PM Skill Nobody Is Teaching
You're not getting bad AI outputs because the AI is bad. You're getting bad outputs because you start from zero every single time.
Dear readers,
Thank you for being part of our growing community. Here’s what’s new this today,
Context Engineering - The Most Important PM Skill Nobody Is Teaching
Note: This post is for our Paid Subscribers, If you haven’t subscribed yet,
Every morning, product managers across the world open a new Claude or ChatGPT tab and begin typing. They explain their product. They describe their users. They outline their constraints. They upload a doc and hope the AI “gets it.” Then they get a mediocre output, tweak their prompt, and try again.
This is the wrong approach, and it’s costing PM teams hours every week and millions in bad product decisions.
The skill that separates the top 5% of AI-powered PMs from everyone else isn’t prompting. It isn’t vibe coding. It isn’t even knowing how RAG systems work. It’s context engineering, and almost nobody is teaching it.
The Gap Nobody Talks About
In February 2026, Harvard Business Review published a piece by Stanford researchers arguing that the real unlock for AI adoption isn’t prompt engineering or even generative AI literacy, it’s applying product management thinking to how AI is used inside organisations. Specifically: defining valuable problems, running structured experiments, and building reliable workflows around AI outputs.
The researchers were onto something important, but they undersold the tactical implication: the PM skill most urgently needed right now is knowing how to engineer the context your AI works within, not just the prompts you give it.
Productside’s 2026 AI PM workflow guide puts it bluntly: “Without persistent context, AI is guessing. And guessing is how hallucinated TAM numbers, bad assumptions, and incorrect requirements slip through unnoticed.”
Most PMs are unknowingly asking AI to guess. Every. Single. Day.
What Is Context Engineering? (And Why It’s Different From Prompting)
Prompt engineering is about how you ask. Context engineering is about what the AI knows before you ask.
Here’s the clearest way to think about it:
Imagine hiring a brilliant new PM onto your team. On day one, you ask them to write a PRD for your next feature. They produce something generic because they don’t yet know your product, your users, your company’s voice, or your competitive positioning. Three months later, their PRDs are exceptional because now they carry all that context in their head.
Prompt engineering tries to cram all of that context into a single message. Context engineering builds the equivalent of that 3-month onboarding, permanently, systematically, and reusably.
Technically, as Sombra’s January 2026 engineering guide defines it: “Context engineering is the systematic design and management of the information an AI model encounters before generating an answer.”
It includes:
System-level instructions, who the AI is, what it knows, and what it should never do
Persistent product knowledge, your PRD templates, user personas, competitive landscape, brand voice
Structured memory, decisions already made, constraints already established, past outputs already validated
Tool access definitions: what the AI can and cannot reach when operating autonomously
When you build this infrastructure once, every AI interaction that follows becomes dramatically more accurate, more consistent, and faster, because the AI is no longer starting from zero.
Why PMs Are Uniquely Positioned to Master This
Here’s the irony: context engineering is fundamentally a product management problem, not a technical one.
It requires:
Defining what information matters (a prioritisation problem)
Structuring that information cleanly (a systems thinking problem)
Updating it as the product evolves (a backlog management problem)
Knowing when outputs are trustworthy (a QA and evaluation problem)
These are exactly the muscles PMs have been building for years. The engineers building AI tools are often worse at this than PMs are, because they think about context as a data pipeline problem, not a knowledge architecture problem.
This means context engineering is the rare AI skill where PMs have a genuine, structural advantage over their technical colleagues, if they choose to develop it.
The 3 Layers of a Well-Engineered Context Workspace
The best AI-powered PM teams in 2026 operate with what you can think of as a 3-Layer Context Stack:
Layer 1 : The Foundation (Static Context) This is what never changes. It lives at the top of every AI workspace and includes:
Your product’s one-paragraph description
The 3 core user personas with their jobs-to-be-done
Your company’s strategic priorities for the year
Your product principles and non-negotiables
Your preferred output formats (PRD structure, ticket format, etc.)
