Hear the Three Tiers (and Name the Failure Mode)

Part 2 of the field-guide companion to Builder-Leader. Chapter 1 was about where you stand. Chapter 2 is about what you start to hear once you're paying attention. Three drills, no tools, your next meeting is the lab.

Drill 1: hear the three tiers

From outside, AI looks like one thing. From inside, it's three, and they move at completely different speeds.

• The model is the raw capability you rent: the frontier system itself.
• The app is the product wrapped around it: ChatGPT, Copilot, the chat box.
• The harness is the structure you build around the model: the saved instructions, memory, tools, skills, and workflows that turn a chat window into something you operate.

Here's the drill. In your next meeting or thread where AI comes up, every time someone says "AI," "it," or names a product, silently ask: which tier are they talking about?

Most of the time the answer is app, and the speaker doesn't know it. Occasionally it's model, and they're reading vendor marketing back to the room. Almost never is it harness, and when a harness conversation does surface, it usually comes from one person while everyone else glances around.side quest

Write down one paragraph afterward: what tier did the conversation live at, and what did the speakers assume the other tiers were doing? If your honest answer is "everyone was talking about a different tier and nobody noticed," that's the most common result, and it's the whole point. You can't reason about what's moving until you can hear which layer people mean.

Drill 2: name the failure mode

Executives confuse "AI doesn't work" with "we're stuck in a known failure mode with a known cause." The fix starts with the name. There are three worth knowing:

• Prompt doom loop: the same task re-prompted endlessly, each attempt slightly rephrased, never converging. Effort with no ratchet.
• Pilot purgatory: a demo that worked, got a green slide, and never reached daily use. The org has dozens.
• Silent failure: the agent produced something plausible and wrong, nobody caught it, and it shipped. The most expensive of the three because it doesn't announce itself.

Open a note. Three headers, one per mode. Under each, write one or two specific incidents from the last six months. Name the project, name the moment. Then circle the header with the most entries and underline the single incident that cost the most.

You now have a page you could hand a peer that says "this is what we're fighting, specifically," instead of "AI is hard." If the page is thin, you're either lucky or not yet looking.

Drill 3: find your near-Group-3

Somewhere in your orbit are two or three people paying for a frontier product and using it for real work. You can name them in under a minute. Pick the one whose output has most visibly changed in the last six months.

Two lines. Line one: what specifically changed about what they ship, or how fast. Line two: what specifically are they doing that you aren't.

That person is the shortest path to knowing what you'll look like in six months. Copy what they do. And if you genuinely can't name anyone who fits, that is the finding: your professional orbit is still on the near side, and changing that is the first move, not a later one.side quest

Why these three

The book calls the quiet operators "Group 3,"³3The book's term for the subset inside the paid, technical AI user base who don't just use these systems but build the structure around them, the harness. Most orgs have two or three; the drills above are how you find yours. the subset inside the paid, technical user base who don't just use the systems but build the structure around them. The three drills train the three channels you need to find them and to reason about the terrain: which tier a conversation is really about, which failure mode is actually biting, and who near you is already across.

None of it requires a terminal. It requires listening differently for a week.

The full Chapter 2 field guide is in the public repo: field-guide/02-what-you-see.md. The book is at builder-leader.com.

Next: why your organization can't cross this gap institutionally, with a number you can run on your own pilot list.