The many levels of differentiation

There’s a mental model founders learn early: differentiate or die. Pick your niche, own your category, be the best [thing] for [person]. In "Do one thing well", I wrote,

When you do one thing well, you are solving a problem in a way where one aspect of your solution is 10x better.

It’s a good principle. But it’s an incomplete principle — and I think I have a better mental model for differentiation recently. Let's dive in.

The problem is that founders, myself included, often treat differentiation as a single decision. A positioning choice. Something you make once in a Google Doc with your co-founder at 2am and then hand to your marketing team.

"We’re the AI coding assistant for enterprise teams. Done. Ship the landing page."

But the companies actually winning right now aren’t differentiated at the headline level. They’re differentiated at every level of the stack — from the tagline all the way down to individual micro-workflows buried three clicks deep that most of their users haven’t even found yet. Differentiation isn’t a decision. It’s a hierarchy.

The Headline Is Just the Entry Point

When Cursor launched, the headline was simple: an AI-first code editor. That was the category claim. But what actually made Cursor compound wasn’t the tagline — it was the depth of differentiation inside the product.

Tab completion that predicts multi-line edits based on what you just did. An agent that can hold an entire codebase in context and make coordinated changes across files. A chat interface that understands your repo, not a generic model’s best guess. Inline diffs you can accept or reject line by line. Each of these is its own layer of differentiation. Each one serves a specific workflow. Together they create something that doesn’t feel like GitHub Copilot with better branding — it feels like a genuinely different category of tool.

The headline got people in the door. The hierarchy of differentiation kept them there and made them evangelize.

Why Most Startups Stop at Layer One

Founders are incentivized to think about top-level positioning because that’s what gets them funded, talked about, and covered. VCs pattern-match to category claims. “The Notion for [X]” or “Cursor for [Y]” is a fundable sentence. So founders optimize for the fundable sentence.

The internal product work gets framed as “building features” rather than “extending differentiation downward.” And when a competitor enters the market with a similar headline — and they will — the company that only differentiated at the top has no moat below the surface.

This is exactly what happened to the first generation of “AI writing tools.” The category headline was differentiated in 2021. By 2023, there were 400 companies with the same tagline. The ones that survived either found a specific vertical (legal, marketing, dev docs) or built deep workflow differentiation within that vertical. The ones that didn’t are dead or acqui-hired.

Differentiation All the Way Down

Think of your product as having five layers where differentiation can live:

Category claim: Who you are and who you’re for. The “best X for Y.” This gets you in the conversation, and sometimes make good landing page headlines. Like Harvey.ai, who is the "AI copilot for lawyers"; or Lovable, which enable you to "build apps from plain English". But its clear that alone, this is fragile. There are already dozens of “AI lawyer” and "Vibe Coder" clones.
Core interaction model: How the product feels to use at a fundamental level. Cursor’s tab completion unlocked “I describe changes” as a new way to code. Granola changed call recordings into passive capture + structured recall, wrapped in a local first, invisible UX.
Key workflow features: The specific moments in your user’s job-to-be-done where you’ve built something no one else has. These should map to the 5-8 most important things your user does in your product every week. Linear looks simple on the surface, but it has issue triage w/ auto-priority, cycle planning, Git-linked status updates, project → milestone rollup, Slack/issue sync, each solving a highly repeated moment of friction in eng team's week.
Power user depth: The features that only 20% of users find but that turn those users into your most vocal advocates. Think Figma's components, design systems, plugins, dev mode. I would think that micro-interactions and other bits of delight also fall into this bucket.
Durable differentation and moats: The things that get harder for competitors to replicate over time because they require your specific data, your specific network, or your specific feedback loop.

Most companies differentiate at layer 1. Good companies get to layer 3. The compounders — Cursor, Notion, Linear, Figma — differentiate at all five.

Every click hits different

When you’re doing product planning, stop asking “what features should we build next?” and start asking “at which layer of our differentiation hierarchy are we weakest?”

If you have a strong category claim but your core interaction model is the same as your competitors, that’s where to invest. If your key workflows are covered but you have no power user depth, you’ll retain but not evangelize. If you have depth but no data advantage, you’re one well-funded competitor away from being copied.

ChatGPT actually gave me a great litmus test while I was reseraching this article. Ask, "Is my differentiation still true after 5 clicks?". I think that's a wonderful way of thinking about it.

And when you’re pitching — to customers, to investors, to recruits — don’t just sell the headline. Walk people through the hierarchy. Show them the category claim, then the interaction model, then the workflow-level features, then the depth below that. The best investors in 2026 have seen enough “AI-first X for Y” pitches to know that the headline is table stakes. What they want to see is that you’ve thought about differentiation as a system, not a sentence.

Differentiation isn’t a decision you make once. It’s architecture you build continuously, all the way down. The headline gets you noticed. The hierarchy is the moat.