Antony Evans

grounded-practitioner Updated Jul 9, 2026

AI-redpilled: the test that separates talk from what's built

AI-redpilled is the claim that someone has moved past AI skepticism into daily use. It’s a test, not a title, and you can’t verify it by listening, since the loudest claims are usually the thinnest. The honest test is a screen-share for a person, and an accountability check for a company: what got built, and who owns it.

What does it mean to be AI-redpilled?

Being AI-redpilled means you’ve moved past skepticism into daily use: you run agents on real work, you’ve shipped something with them, and you’d rather show your screen than describe your opinion. It’s a test of behavior, not belief, and it applies to one employee or a whole company. The word borrows from crypto and meme culture: “pilled” means convinced past hedging, claimed constantly and checked rarely.

How does Laurel’s AI-maturity ladder expose the gap between talk and build?

JZ, Chief Product Officer at Laurel, built a four-level ladder to sort talk from build. She runs it in hiring interviews across every function, and the same ladder works one level up: judging a whole company’s AI claims rather than one hire (source video). The four levels, as JZ describes them:

Laurel’s broader operating system spans functions too, including customer success, design, finance and legal.

Her finding is consistent: again and again, the people who lead with the label turn out to be level one once she watches them work. Level one is not nothing; it is just not what the claim implies.

Why does a screen-share beat a claim?

JZ’s version of this test skips the survey question entirely. She asks candidates to open their laptop and show, live, the AI workflow they use for real work, not the version they’d describe in an interview. A resume or a LinkedIn post cannot fake a screen-share; conversation can, until someone opens the actual tool in front of you.

I score myself against this same ladder rather than asking anyone to take my word for it. We built a multi-agent system: one model drafts, two reviewers check, and a ground-truth pass runs against sources before anything ships, and this article went through it. On JZ’s scale that puts me at level four, since I’ve built something other people use, beyond my own tools, after months of false starts on the checking layer.

How does the individual test scale to a whole company?

A business cannot open its laptop for you, so the company version of the screen-share is not a workflow demonstration. It is an accountability question: who owns each initiative, and what shipped under their name. Laurel names two mechanisms instead of a mission-statement claim about “AI-native,” as JZ describes them:

Who is accountable when a company calls itself AI-native?

The honest answer names a person and a mechanism, not a value on a slide. Laurel’s clearest version is a dedicated function that someone owns full-time, rather than an effort squeezed into spare time between meetings.

“a lot of times when you say it’s everyone’s responsibility, it’s no one’s responsibility.” JZ, Chief Product Officer and product-team manager at Laurel, says that is exactly why the company runs a named AI Ops function instead of leaving AI adoption to enthusiasm (source video). One builder proved the value inside a single function; it then scaled to one AI-ops person per function. The same shift shows up in her own team: five product managers and four designers today, down from the hundreds of people she once managed.

Why does the delivery layer matter more than the tool?

A skill library nobody opens is the same as no skill library. JZ’s fix is a daily briefing pushed into Slack and email, delivered where people already work, not a new interface.

She has direct experience of the failure mode too: she over-built personal automations, hit information overload, and fixed it by consolidating everything into one daily surface. I’ve made a milder version of the same mistake: my own automations multiplied until I stopped reading half of them, and the fix was cutting the count, not building a smarter dashboard. Meeting people where they work decides whether a skill reaches everyone or just the few who’d have found it anyway.

Where does this test fail?

This test only tells you whether AI use is real, not whether it is good. That limitation matters: plenty of level-four builders ship things that are shallow or actively unhelpful. It also fails over a resume, a phone call, or a values statement, since it needs a live screen and a checkable artifact. Treat it as a filter for talk, not a substitute for judging the work itself.

FAQ

What does AI-redpilled mean? Someone who has moved past AI skepticism into daily, hands-on use, not someone who only talks about it.

How do I check if I am AI-redpilled or just talk about it? Open your laptop and walk someone through the AI workflow you use for real work. JZ’s screen-share test works the same way for a candidate or a self-check.

What is the AI-maturity ladder JZ uses at Laurel? A four-level scale: chat mode (level one), one automated workflow (level two), a personal tool (level three), and a shared tool other people use (level four).

What is the captain model and why does it matter for AI governance? It assigns one accountable owner per initiative, picked by whichever skill is most load-bearing, so testing and shipping sit with the same named person.

Does a company need a dedicated AI Ops team to be AI-redpilled? Not strictly, but Laurel found that AI adoption stays no one’s job until it is someone’s full-time job, hence the named AI Ops function instead of scattered enthusiasm.

Last updated: 2026-07-09