I was deep in a spreadsheet, manually mapping perception gaps between a leadership team and their staff across eleven different organisational behaviours.
Charts. By hand. Me. At a computer. In 2024.
(I would like to point out that at no stage did I think this was a problem. That's the best part. I just thought it was consulting.)
Same analytical structure as the engagement before. Same gap analysis, same report sections, same blank template with a fresh client name at the top. I'd been doing this for years — like, years — and every single time I opened a new file and started from scratch like I'd never done it before in my life.
At some point, staring at the screen, I had one of those quietly awful realisations. The kind where you see your own habits from outside your body and think: oh shit. The work was great — clients found it valuable, the insights landed, the engagements went well. That wasn't the issue. The issue was that I'd been solving the same problem over and over rather than building something that improved with each iteration. And the more I sat with it, the more I recognised it wasn't just a quirk of how I work.
It's how most knowledge work operates.
The name for what most of us are doing
There's a phrase that's been rattling around my head since I came across it in a piece about AI and creative work: artifact thinking.
Artifact thinking is how most knowledge work gets done. You produce an output, the output is the point, then you start from scratch next time. Every deliverable is its own island. The process disappears the moment you send the file. Nothing compounds. You just do it again.
You do the thing. You deliver the thing. You get paid for the thing. Tomorrow you do another thing.
Sound familiar? (Of course it does. You're doing it right now.)
It's how almost every business I've walked into operates. Not because people are lazy or disorganised — most of them are working incredibly hard. It's just how the work has always been structured. The deliverable is the product. How they arrived at it is less important. Few get paid to have a better process. They get paid to deliver the thing.
AI didn't fix this. It sped it up.
A lot of businesses I work with have spent the last year "doing AI." And I use that phrase in the loosest possible sense.
They've bought tools. Run workshops. Had the all-hands where someone from IT presented a deck with a glowing robot on the cover and the word "innovation" seven times in fourteen slides. (I've seen this deck. I might have even created this deck. The robot is always vaguely threatening. It's terrible.) Their people use ChatGPT for emails, Copilot for meeting summaries, and some brave soul in marketing is generating social content with an image tool and hoping no one looks too closely at the hands.
What they've actually done is get faster at artifact thinking.
The proposal that used to take four hours now takes forty minutes. The job ad that needed two rounds of drafts comes out decent on the first pass. The weekly report that three people dreaded writing assembles itself while the coffee brews. That's genuinely useful. Speed matters! But the proposal still starts from scratch. The job ad still starts from scratch. The weekly report still starts from scratch.
AI gave them a faster typewriter. They're still typing individual letters.
There's a particular conversation I keep having. Someone shows me their "AI workflow" — strangely, it's often with the kind of pride you reserve for a new car — and what they've built is a prompt they copy-paste into ChatGPT every Monday morning. That's it. A sticky note, digitised. I've started nodding along because the alternative is explaining that a prompt you copy-paste each week isn't a workflow, it's a ritual to produce an artifact.
I wrote a piece a while back and mentioned what I called "AI theatre" — the pattern where AI initiatives create the impression of transformation without the reality of it. The endless proofs-of-concept. The quick wins that never compound. The innovation workshops where everyone nods along and goes back to doing exactly what they were doing before.
Artifact thinking is why AI theatre happens. Not because the tools don't work — they work fine. It's that you've handed them to a business that was never built to compound its own knowledge in the first place.
The AI implementation didn't fail. There was just nothing underneath it for the AI to amplify.
Two types of businesses
Some businesses are doing something different.
Not just faster. Different.
I watched this play out with a couple of clients in the same industry. Both MSPs (managed service providers: companies that run IT infrastructure for other businesses; basically the people you call when your computer's on fire and you need someone calm to show up), similar size, similar client base, same region. One had spent years building proper playbooks. The other ran mostly on institutional memory and the tacit knowledge of whoever had been around longest. They were good at what they did. They just hadn't written any of it down.
Both adopted AI tools at roughly the same time.
Here's what the first business actually had. Their client onboarding wasn't a folder of documents someone had to remember to send — it was a documented sequence with defined steps, owners, and triggers. When they added AI, it ran inside that structure: pre-populating client context from discovery notes, flagging risks based on patterns from previous onboardings, drafting the first-week communication sequence for the account manager to review and send. Their sales process had a qualification framework that AI could research against before a meeting — so their reps showed up with prepared questions rather than generic ones. Their quarterly client reviews ran on a playbook, which meant AI could pull the relevant data, structure the narrative, and surface anything worth flagging. The account manager's job shifted from assembling the thing to thinking about what it meant.
The system got smarter with every client. The team got better at their jobs without having to work harder at them.
The second business?
Same AI tools. Better-looking proposals. Slightly less painful report writing.
The only real difference was what those tools had to work with. AI multiplies what already exists. Hand it a real system — documented, repeatable, refined — and it amplifies it. Hand it artifact thinking and you get a very efficient way to produce one-off outputs.
That gap widens every month as the AI labs push out smarter models. And most businesses on the wrong side of it can't tell, because the AI tools feel like they're working. Proposals are faster. Emails are better. Demos look sharper. Nobody's measuring whether any of it compounds.
This is a leadership problem dressed up as a tech problem
The reason businesses stay stuck in artifact thinking isn't time, or tools, or budget.
It's identity.
A lot of founders and senior leaders have built their entire career on being the person who knows how to do the thing. That knowledge is the value proposition. It's what they sell, what they're proud of, what gets them in the room. And systematising it — writing it down, building it into a process anyone can run — can feel like giving it away. Like admitting the thing that makes you valuable isn't quite as special as you believed.
And, hey, I've felt this too. It took me longer than I'd like to admit to recognise that the logic is exactly backwards. The person who can systematise their expertise is harder to replace, not easier. You're not giving it away. You're encoding it. The skill becomes infrastructure. Infrastructure compounds. Individual expertise, delivered one client at a time from memory, doesn't.
But that's the uncomfortable decision nobody makes explicitly: that the way we do this thing is worth documenting, refining, and building into something that improves over time. There's always a more urgent version of the deliverable that needs to be done today. The system can wait. After the pitch, after the quarter, after things settle down.
The system always waits.
Most leaders are too busy producing artifacts to build the machine that produces them. That's not a technology problem. It's a culture problem — specifically, a failure to treat the process as the product.
AI doesn't fix that. It just makes it more expensive to ignore.
So, yeah. Back to that spreadsheet.
I eventually stopped doing it the same stupid way.
I spent the better part of a year building a proper system around the diagnostic work — survey logic, perception gap analysis, reporting, all of it systematised into an actual tool. Before Faultline, my diagnostic quality depended partly on how tired I was the night before delivery. Now every engagement runs the same rigorous process regardless. The survey is consistent. The gap analysis more scientific. I spend my time on what the data means, not building the charts that display it. That shift — from constructing the output to interpreting it — changed what the work feels like from the inside. It's the difference between operating the camera and directing the film.
The businesses getting real returns from AI aren't the ones with the biggest budgets or the most sophisticated tools. They're the ones who decided — usually before AI was even part of the conversation — that their processes were worth systemising. That's the actual advantage. Not the model. And definitely not the prompt.
Don't ask "how do we use AI better?" The question worth sitting with right now is something simpler but more uncomfortable: what's one process in your business that should be documented before you touch AI again? Not your most complex one. Not the one that requires the deepest expertise. The most repeated one. The thing your team rebuilds from scratch most often. Start there.
That's it. That's the move.
Most people reading this will recognise the problem. They'll feel it. And then they'll open a blank deck and start from scratch, because the deliverable is due tomorrow and the system can wait.
It always waits.
Until the businesses that didn't wait are running laps around the ones that did.