The internet cannot stop making IKEA furniture into something else. Apartment Therapy is rounding up ten places to shop for your next hack. Man of Many has a list of sixteen. The Spruce found eighteen different ways to turn one cabinet into a greenhouse for houseplants. At the same time, Livingetc published a piece titled “I Don’t DIY my IKEA Furniture — These Brands Do It For Me.” So the same trend that produced a thriving hack culture also produced a market for people who want the hack without doing the hacking.
I review AI toolkits for a living. That contradiction is the most familiar thing I have read all month.
Every toolkit market ends up here
The pattern repeats. A platform ships something modular and cheap. A community forms around bending it into shapes the original designers never planned. Then a second layer of businesses appears, selling the finished result to everyone who liked the idea but not the labor. Agent frameworks did this. So did no-code builders, prompt libraries, and browser automation tools. The hackers arrive first, the productizers arrive second, and both are correct about what they want.
What IKEA got right, and what most AI toolkits get wrong, is the part before either group shows up. The parts are legible. You can look at a shelf pin and understand what it does. The joints come apart. Nothing is glued shut. That is the whole reason a hack culture exists in the first place: the product was designed in units small enough to recombine.
Contrast that with a toolkit that ships as a single opinionated pipeline. You can use it exactly as documented or you can fight it. There is no third option, because the pieces were never separable to begin with. The Spruce did not find eighteen greenhouse cabinet hacks because the community is unusually clever. They found eighteen because a glass-front cabinet is a small number of well-understood parts.
The two honest customers
Dengarden ran a piece on four ways to make IKEA furniture look more expensive. Livingetc ran a piece about outsourcing the whole thing. Both audiences are real, and toolkit vendors keep pretending only one of them exists.
The DIY audience wants:
- Parts that come apart without breaking
- Documentation of how the parts actually connect, not just the happy path
- Permission to void the warranty and still get support
- A community whose modifications do not get patched out on the next release
The outsourcing audience wants:
- Something that works on day one, configured by someone who has done it before
- Clear pricing on the finished thing rather than the raw materials
- Someone else’s judgment about which of the sixteen approaches is the good one
Most AI toolkits try to serve both with one product and end up serving neither. The framework is too opinionated for the tinkerers and too raw for the people who just want the result. You get a codebase full of abstraction layers that exist to make things configurable, plus a marketing site promising it works out of the box. Both claims are half true, which is the worst amount of true.
What I actually check for now
The IKEA hack trend gave me a decent test. When I look at a new toolkit, I ask whether a third party could build a business on top of it without asking permission. Not whether they should, whether they could. If the answer is no, the parts are not really parts. They are a monolith with a config file.
The second test is whether the community modifications survive upgrades. A hack culture only compounds if the underlying product stays stable enough to hack against. If every minor version breaks the interesting extensions, you do not have a platform. You have a moving target, and the people doing the most creative work with your tool are the ones you punish hardest.
The third test is simpler. Read the docs and count how many times they tell you what not to do versus how something works. Assembly instructions describe mechanisms. Marketing describes outcomes. You can tell within a page which one you are holding.
The unsexy conclusion
Nobody wrote a listicle about hacking furniture that arrives welded together. The hack economy grew out of a specific design decision about modularity, and the resale economy grew out of the hack economy. That order matters. You cannot skip to the part where third parties build value on your platform if you never shipped separable parts.
For toolkit builders, the lesson is to pick a layer and be honest about it. Ship the flat pack or ship the assembled piece, and say which one you are selling. For the rest of us evaluating these things, the useful question is not how good the demo looks. It is whether anyone outside the company has managed to build something the company did not plan for. That number tells you more than any benchmark.
🕒 Published: