The $495 Screw
Simplicity has a smaller market.
When I was younger, I remember hearing a story about a factory whose machine had stopped working. They called in a specialist. He walked around the machine, tightened a single screw, and handed them an invoice for five hundred dollars.
The manager was furious. Five hundred dollars for turning one screw? He demanded an itemized bill. The specialist sent it back the next day:
Tightening one screw: $5
Knowing which screw to tighten: $495
The parable has been around forever, and it endures because it exposes something uncomfortable about knowledge work: the answer is often simple. A diagnosis, a judgment call, a single decision that only looks obvious in hindsight. But simplicity doesn’t feel like value. One turn of a screwdriver doesn’t feel like five hundred dollars. So the knowledge tends to be dressed up — wrapped in process, padded with justification — until it feels hefty enough to be worth what you paid.
This is everywhere. Apple Watch boxes are engineered to weigh more than the watch inside them, because a $400 purchase shouldn’t feel like picking up an envelope. McKinsey’s answer fits on a napkin, but it arrives inside a six-week engagement — the interviews, the workshops, the war room, the hundred-slide readout for the C-suite. The knowledge was there on day one. The other five weeks and six days are the packaging that makes the napkin feel earned.
We need the weight. It’s how value registers.
The Tape Aisle
Last week my HVAC system had an issue. I took a photo and sent it to Gemini. It told me I needed to reseal a joint with tape. So I drove to the hardware store, and found myself standing in an aisle staring at fifteen rolls of tape I couldn’t tell apart — foil tape, aluminum tape, HVAC tape, duct tape (which apparently can’t handle the pressure and heat of HVAC ducts), three different widths, two different brands of what looked like the same thing. I took photos of the shelf and sent them back. Gemini told me which one. I drove home, and it walked me through the fix.
An HVAC specialist would have had the right tape on the truck and known exactly what to do. But would that have justified the truck roll and the service call? The specialist’s real value — knowing which tape, knowing the technique — is the same $495 from the parable. It’s genuine expertise. But when I can get that expertise from a photo and a thirty-second exchange, the performance that used to surround it — the scheduling, the visit, the diagnosis you watch happen in person — just falls away.
AI is stripping the packaging off the knowledge.
26,000 Lines of Common Sense
But, that same packaging instinct is running at full speed in the AI tool ecosystem — just in the other direction.
There’s a popular open-source SEO skill for Claude Code that has nearly five thousand GitHub stars. It promises to replace $300 a month in SEO tools with a single terminal command. The repo contains 243 files, 18 subagents, and over 26,000 lines of markdown. Nine AI agents run in parallel. The scale feels serious.
I cloned it and started reading.
The technical SEO agent says: fetch the page, check stuff, provide recommendations. The content agent says the same thing. Buried in there are genuinely useful bits — a few reference thresholds, some crawler token names, schema templates. Maybe two hundred lines of data the model doesn’t have memorized. The rest is telling the model to do what it would already do if you typed “run an SEO audit on this URL.”
Twenty-six thousand lines. Two hundred that add something new. The other 25,800 are the $495.
And just like the specialist, the creator can’t ship two hundred lines. Nobody stars a repo that’s a single page of reference data. Nobody clones it, nobody writes a blog post about how it replaced their $300 tool stack. Two hundred lines feels like a Post-it note, not a product. So the answer gets dressed up — eighteen subagents instead of one prompt, 243 files instead of a page of notes — the same way McKinsey dresses up a napkin into a six-week engagement. The market rewards the weight.
On the other end, people collect these repos the way I used to collect articles in Evernote — clipping everything that looked interesting, building a graveyard of aspiration disguised as progress. Downloading 243 files of SEO instructions you haven’t read is not the same as understanding SEO. But it feels like it. The weight is convincing from both ends.
The Shrinking Window
There’s a window between “too dumb to do it without detailed instructions” and “smart enough to figure it out.” Every one of these skills lives inside that window — and the window is closing.
Two years ago it was wide. Models needed elaborate prompting to do coherent work. Today they already know what an SEO audit involves, the same way Gemini already knew which tape I needed. The 26,000 lines aren’t teaching the model; they’re reminding it. Each new generation absorbs a little more of what used to require explicit instruction, the way a junior employee gradually stops needing the checklist.
The things I actually keep in my own AI setup — the handful of instructions that genuinely change how the model works for me — fit in a few short files. Not because I’m disciplined, but because most of what I tried to codify turned out to be stuff the model already knew.
The screw was already tight. I just hadn’t checked.