Raise your ceiling; raise your floor
Raising the ceiling for expert AI users matters, but an AI-native organization also raises the median through shared systems and habits.
Last week we highlighted a report on AI maturity from Notion. We didn't spend much ink discussing the maturity levels they defined to create their groupings, but their 4-level framework is deceptively simple and one of the best we've seen. This week Addy Osmani makes a compelling case for a 6-level framework for individual maturity that we find rather convincing.
These maturity frameworks are useful not only because they help orient you to where you are today, but also give you a roadmap of areas you can improve in to continue up the maturity curve. But we think the most insightful moment from Osmani is actually the recognition that, contrary to popular narratives, you rarely inhabit only a single rung of the ladder:
A good day doing engineering includes touching several rungs, sometimes more: it's normal to switch between the eras a few times in the course of a task.
We think this holds just as well for organizations as it does for individuals. Your entire org won't inhabit the same rung of the ladder, and nor should it. Different workflows, different functions, different tasks require different approaches and they don't all need full-throated AI autonomy (at least not yet). So take note of your high-water mark (what's the highest level you're capable of) as well as your median (what level does most of your work happen at) and work to move both of those upwards over time. But don't obsess over fitting cleanly on a single rung.
Here's what else we're reading this week.
Agentic Autonomy Levels
Osmani does some deep and convincing analysis, showing his work to get from Steve Yegge's widely referenced maturity levels from early this year to the crisper set he proposes here. It's a lengthy read but a worthwhile one, for all the reasons we laid out above.
How Coinbase keeps AI spend flat while usage grows
Brian Armstrong from Coinbase on how they've flattened out the slope of AI spend while usage continues to grow. He gives concrete tactics for controlling spend; some are more advanced, but others are easy for anyone to employ. Our favorite part is this line making the case for better visibility:
Our engineers can use as many tokens as they want, from whatever model they want, but we’ve made usage visible – and the more you spend on AI, the more impact we expect.
Visibility is the most important gap in AI adoption for most companies we talk to, and this is one excellent reason why. You can't influence what you can't see. We feel compelled to point out, however, that at most companies broad adoption is a more urgent problem than controlling spend (so don't get too far ahead of yourselves).
Atlassian and Dropbox on AI transformation
Atlassian's Avani Prabhakar and Dropbox's Allison Vendt compared notes on stage at Atlassian's recent "Team '26" conference. Some interesting tidbits throughout, but this is the one that really caught our attention because we haven't seen anyone else sharing a concrete formula for measuring and defining "super-users":
We’re looking at whether someone is using advanced AI features (across 1P and 3P tools) at least 40 times a week within their role-specific workflows. About 30% of our people are there today.
Defining Taste
Mitchell Hashimoto, founder of Hashicorp and creator of popular terminal emulator Ghostty, likes to drop "cooler heads" takes that cut against some of the online hype. He's certainly earned the right to be listened to, and this dispatch on what taste is and why it matters more than ever is worth your time.
One thing to try this week
If you're reading the articles above and not sure where to get started, we think there's one clear answer: improve your visibility. So what's one thing you can do on Monday to move closer to being able to answer questions like:
- Who on your team is using tokens inefficiently (and how would they know)?
- How mature is your median AI usage?
- Who are your "super-users"?
All of these require measuring AI usage. If you have no visibility, stand up a simple dashboard to track basic usage. If you can see basic usage, make sure your teams can see it broken down in ways they can act on. If you have that, work on tracking more granular usage. Don't try to do too much, just take the next step.
Have a great weekend!
— Claire + Zach