// pinned
•
July 1, 2026
•
6 min read
Don't leave a trash ocean.
Everybody's being told to go adopt AI right now. Faster, cheaper, more of it, this quarter. And a lot of teams are doing exactly that - and quietly leaving behind an ocean of garbage in their wake.
I mean it literally. Half-checked docs that were never read past the first paragraph. Hallucinated "facts" copy-pasted into a knowledge base because the output sounded authoritative. Automations that someone spun up in an afternoon, felt like a genius for, and never touched again - still firing off emails to customers who churned eighteen months ago. Tech debt with a shiny new coat of paint. And the person who inherits all of it? They didn't build it, they can't read it, and they have no idea which parts to trust.
AI adoption without cleanup and ownership isn't progress. It's just faster mess-making.
Here's the part nobody puts on the conference slide: generating something is now nearly free, but maintaining it never got cheaper. A model can draft you forty pages of process documentation in ninety seconds. It cannot tell you when that documentation goes stale, notice when the underlying system changes, or feel embarrassed when a new hire follows it off a cliff. That's still a human job. It has always been a human job. AI just lowered the cost of creating the thing while doing absolutely nothing about the cost of living with it.
So the volume goes up and the ownership stays flat, and you get a trash ocean: technically "documented," technically "automated," functionally untrustworthy. Everyone's afraid to delete anything because nobody's sure what's load-bearing. That's not an AI problem. That's an accountability problem wearing an AI costume.
What responsible adoption actually looks like
Responsible adoption is boring, and I mean that as the highest compliment I can pay it. It looks like this:
- Someone owns every output. A name, not a team. If AI wrote it, a human is on the hook for whether it's right and whether it stays right.
- It gets verified before it counts. "The model said so" is not a source. Checked, corrected, or thrown out - those are the only three exits.
- It gets maintained. Every AI-generated artifact has a shelf life. Put a review date on it the day you ship it, the same way you'd date a carton of milk.
- It gets decommissioned when it's stale. Killing a dead automation is a feature, not a failure. Delete with confidence.
The whole ethic fits on a sticky note: leave the codebase and the knowledge base better than you found it - not buried under it. If your AI rollout added more things nobody trusts, you didn't adopt AI. You just increased your surface area for future embarrassment.
Why this matters double in a financial institution
If you run ops at a fintech or a credit union, this isn't a tidiness preference - it's exposure. An examiner does not care that your policy doc was "AI-assisted." They care whether it's accurate, current, and owned. Model risk management assumes someone can explain and stand behind the output; "the AI generated it and we shipped it" is the sentence that turns a routine exam into a very long afternoon. And member trust - the entire product, when you get down to it - does not survive being told something confident and wrong by a system your team never bothered to check.
Regulated environments are just the place where the trash ocean finally sends you the bill. Everywhere else, it's a slow leak. In financial services, it's a finding.
So adopt AI. Please. Adopt it aggressively - it's the best lever most teams have right now. But adopt it like an adult: own the output, verify it, maintain it, and clean up after yourself. Speed without stewardship isn't a competitive advantage. It's a spill.
- Elliott // I make hard things fun. Cleaning up is one of them.
// craft
•
July 16, 2026
•
2 min read
Whimsy is an adoption strategy.
I spend my days making AI boring on purpose: auditable, documented, accountable, the kind of boring an examiner loves. Then, for fun, I built an astrologer. She taught me more about adoption than most rollouts I've watched.
Her name is Stella. She reads a birth chart against live planetary data and returns a one-word verdict, and she's banned, by design, from ever saying "trust the process." The surprising part wasn't building her. It's that I built her with the exact same rigor I bring to the serious work: real ephemeris data instead of vibes, live transits pulled fresh every time instead of from memory, a strict output format so she can't ramble into horoscope filler. I gave astrology a governance framework. Nerdiest thing I've ever done, or the most on-brand.
And I actually use her. Constantly. Far more than the self-serious version I could have built instead. That's the whole lesson, and it isn't about astrology.
Whimsy isn't the opposite of rigor. It's what gets the rigor used.
The tools people open are the ones with a little delight built in. A compliance dashboard nobody opens protects nothing. A checklist that makes someone smile gets run. We keep shipping AI that's technically correct and emotionally dead, then act surprised when it sits untouched. Make the thing good. Then make it a little bit fun.
- Elliott // I automated my own horoscope. The rigor is the joke.
// adoption
•
July 8, 2026
•
3 min read
Nobody's opening the tool you bought
Your AI rollout isn't stalling because the tech is bad. It's stalling because everyone's quietly terrified - and no license fixes that.
Companies keep buying AI tools and then wonder why nobody opens them - or why a handful of devs are off having a field day while everyone else pretends the login screen doesn't exist. The tool was never the hard part.
I've watched six-figure rollouts die in total silence. Not because the software was bad - because on day one, someone quietly did the math nobody says out loud: if this does my job faster, what exactly am I for? At the same moment, their leadership was screaming "adopt AI right now, or else." Fear on the floor, pressure from the top, and a very expensive dashboard nobody wants to be the first to open.
Here's what most leaders miss: people don't resist AI. They resist feeling replaceable, exposed, or quietly obsolete. No demo fixes that. No dashboard fixes that. You fix it by answering the question they're too polite to ask - out loud, on purpose, before they have to.
So I started opening with a version of this: "This eats the worst four hours of your week. You keep the part that actually needs a human." Adoption stops being a fight and turns into relief. Same tool, same cost, completely different reception - because the threat got named and defused before it had a chance to fester.
Most "AI adoption problems" are trust problems your org hasn't earned yet. The tech works fine. It's the safety that doesn't feel real.
The fix isn't another platform or a sterner mandate. It's fewer licenses and more honest conversations - wildly radical, I know. The org that says out loud "we're automating the tedious part so you can do the part only you can do," and then actually protects that promise, gets adoption. The one that buys seats and sends a hype email gets a very quiet graveyard of tools nobody ever opened.
- Elliott // I get called in when the rollout's gone quiet and nobody will say why.
// adoption
•
June 16, 2026
•
4 min read
From AI-curious to AI-fluent
Most AI rollouts don't fail because the tool is bad. They fail because the plan was an all-hands email, a pile of licenses, and a hope. I've sat on the other side of this: founding member - and only woman - on an AI Adoption Council, my team's designated AI Champion, the person who actually built the Claude-powered automations and then trained everyone to use them. Here's what makes adoption stick instead of fizzle.
Start with one painful, repetitive task. Not a moonshot. Nobody's life is changed by a demo of AI "reimagining the enterprise." People are changed when the thing they hate doing every Tuesday takes four minutes instead of forty. Win the Tuesday. Momentum is built on small, undeniable, personal relief - not on vision decks.
Name a real champion and give them time, not just a title. "AI Champion" printed under someone's name with zero hours attached is not a role, it's a decoration. A champion needs protected time to answer dumb questions patiently, build the first automations, and be genuinely reachable. I know because I was one. The title was the easy part; the calendar was where it actually lived.
Write the dos and don'ts down where people can find them. Tribal knowledge doesn't scale and it doesn't survive turnover. If the only place your AI guidance exists is in the head of one enthusiastic person, you have a single point of failure, not a program. Put it somewhere searchable. Update it when reality changes.
Measure adoption and outcomes, not vibes. "People seem excited" is not a metric. Who's actually using it, on what, and what did it save? Track the real numbers, because the excitement fades and the data is what lets you defend the program when someone inevitably asks what all this cost.
The loudest skeptic in the room converts the hardest - if you let them see it work instead of arguing them into it.
Let a skeptic become a believer in front of everyone. This is the one that actually moves a room. Don't debate the holdout. Hand them the painful task, let them try the tool on their own terms, and get out of the way. When the person who swore this was hype turns around and says "okay, that's genuinely useful" - unprompted, in front of their peers - you've done more for adoption than any mandate ever could. Skeptics make the best evangelists precisely because everyone knows they weren't easy to win.
Curious to fluent isn't a training budget. It's a sequence: one real win, one real champion, written guidance, honest measurement, and a converted skeptic doing your marketing for free. Do those in order and adoption stops being something you push and starts being something people pull.
- Elliott // still the person who gets handed the "make this stick" problem.