what if this newsletter was also an AI prompt?
Hi Reader
Starting this week, I want to try something new with you.
Today’s newsletter is also an AI prompt.
Read it, and when you get to the end, follow the quick 3-min directions.
And share with me what you learn or discover.
The new bottleneck is language
When I was advising Facebook/Meta from 2016-2023, I saw firsthand how technical debt can quietly tax speed, quality, and scale.
In software engineering, technical debt is the future cost created when teams choose a faster, messier solution today instead of a cleaner one that will hold up over time. It shows up when quick fixes, workarounds, and patches help you ship now. Then quietly make every future change slower, riskier, and harder. What looks efficient in the short term becomes expensive in the system over time. Meta product teams were constantly trying to reduce technical debt, but it often felt like whack-a-mole.
Fast forward to 2026.
Today, there’s another form of debt most companies are carrying, and AI makes it impossible to ignore.
As I’ve been sharing over the past few months, language is the new code base. If you’re building with AI, especially agentic AI, more of your business now runs on words than you realize.
The real question becomes:
How clear, precise, and transferable is the language your business runs on?
What is language debt?
Language debt is the accumulation of vague terminology, undefined concepts, and inconsistent meaning across an organization.
It compounds as:
Misunderstanding: teams think they agree, but operate from different definitions
Decision paralysis: same data, different interpretations, no resolution
Coordination friction: teams debate what words mean instead of executing
Cascading confusion: unclear external messaging creates internal role confusion
Amplification risk: AI systems encode whatever implicit language is already in place
Most organizations don’t yet treat language as a system. They treat it as culture, communication style, or “just how we talk.” But language is not just an expression layer. It is a coordination layer. The words a company leaves vague are the exact places where decision-making slows, ownership blurs, and execution starts to drift.
Whatever remains implicit in shared language becomes explicit in your system’s behavior.
Vague language about outcomes produces vague impact measures
Undefined language about roles produces a confusing org chart
Unclear customer-facing language produces unclear AI agents
The good news: the same AI tools that make this urgent also make it solvable.
Any of you who are working in Claude Code, Claude CoWork, or GitHub have discovered the power of README files.
They are foundational documents that tell the system how to think, behave, and operate.
With agentic AI, you need to bring the same mindset and approach to your strategy, your product, and your go-to-market.
Canonical-grade language systems, with precise definitions, structured frameworks, and clear distinctions, become the reference layer that both humans and AI can reliably execute against.
Organizations that build canonical-grade language systems are resolving decades of accumulated drift in months.
And it means you can compete, differentiate, and outperform.
That’s because the team with the clearest language wins.
Three levels of language debt
1. Language that can’t travel. Positioning or strategy language that requires so much explanation it can’t survive without a narrator. If it doesn’t land instantly without context, it’s creating drag at every handoff.
2. Undefined terms. Core concepts that teams use daily but have never formally defined. Every conversation requires renegotiating what the words mean before any work gets done.
3. Same word, different meaning. Multiple functions using identical language but interpreting it through different incentive structures. The data is the same…the labels each team puts on it are not.
Language debt compounds at scale
At 5 people, shared context fills the gap. Everyone interprets roughly the same frame because they’re in the same room
At 25, the debt starts showing…different teams develop different meanings for the same terms. Meetings get longer
At 300, the debt is structural. Every undefined term generates coordination cost every time it crosses a team boundary
Now layer AI into this equation.
AI amplifies clarity. It also amplifies confusion.
Every model, agent, and automated system amplifies whatever implicit language is already in place.
Better tools do not clean up muddy language. They multiply it.
Pro-tip: The organizations getting this right are treating language as infrastructure…the same way they treat data architecture or code quality. Clear language compounds. Vague language fragments.
Where is language working on your behalf, and where is it costing you velocity, traction, or success?
Upgrade your language system (takes 5 min):
Open your LLM tool of choice (Claude, ChatGPT, Gemini, etc).
Copy/paste this newsletter into it
Pick one internal document: a strategy doc, a PRD, a QBR, an exec memo, etc. Something that actually drives decisions.
And attach that document for analysis.
Copy/paste this exact prompt at the top of your chat: “Based on the concept of “language debt” described below, do a language system analysis of the attached document. Think ontologically. Identify the language that can’t travel, the terms used but never defined, the concepts that could mean different things to different teams, and the single term that, if clarified, would reduce the most coordination friction, and increase the probability of our success. Provide your output in table format, with clear rationale, and actionable language recommendations.”
Read what it outputs, and determine what is most useful.
If you like this upgrade, tell your LLM to add the concept of language debt to your memory for future usage.
Reply to this email and tell me if this has surfaced any new insights or breakthroughs.
This is one of hundreds of “thinking upgrades” I could continue to share going forward.
If this was useful, let me know. Equally, if it’s disappointing, I want to know that too.
I’m curious to hear from you.
Hit reply, and give me your thoughts.
Michael
P.S. When you’re ready, here’s three ways I can help:
If you want a second set of eyes on your narrative, reply and tell me more
If you need narrative architecture, I help operators do just that. Apply for Q2
If you want a storytelling keynote, I still do them on a select basis. Let’s talk