Remember when assembling flat-pack furniture was considered a relationship stress test? There was a stretch of years where the joke wrote itself: two people, one Allen key, an instruction sheet with no words on it, and a growing suspicion that one of the wooden dowels had been swallowed by the carpet. IKEA assembly was shorthand for a task that was technically simple and emotionally brutal.
That joke has quietly expired. Look at what’s trending in home design now and you’ll find something different. Apartment Therapy has rounded up ten places to shop for your next IKEA hack. Man of Many published a list of sixteen of the best ones. The Spruce collected eighteen ways to turn an IKEA cabinet into a greenhouse for houseplants. Dengarden covered a DIYer sharing four hacks to make IKEA furniture look more expensive. And Livingetc ran a piece with a headline I keep thinking about: “I Don’t DIY my IKEA Furniture — These Brands Do It For Me.”
People stopped complaining about the instructions and started rewriting them. I write about AI agents for a living, and this is the clearest analogy I’ve found for what’s happening in that space right now.
Assembly versus authorship
A flat-pack cabinet arrives as a set of parts and a fixed intended outcome. Follow the steps, get the cabinet. That’s assembly. It’s the mode most people are still in with AI tools: you open a chat box, type a request, get an answer, and the interaction ends. Parts in, expected object out.
Hacking is different. A hack starts with the same parts but a different destination. You wanted a plant greenhouse, so you took a glass-front cabinet and treated it as raw material. The manufacturer never planned for that. The parts didn’t change; your intent did.
AI agents live on the hacking side of that line. An agent isn’t a single answer to a single question. It’s a system you point at a goal, that then figures out its own sequence of steps, checks whether the steps worked, and adjusts. You supply the destination. It works out the middle.
Three things the hack trend gets right about agents
The parallels are more useful than they first appear, so let me be specific.
- Standard parts make custom work possible. The reason IKEA hacks became a genre is that millions of people own the same cabinet. Shared components mean shared knowledge. The same is true of AI agents. They’re being built on a small number of widely available models and tools, which is exactly why so many people can share what worked.
- The instructions are a suggestion, not a limit. Nobody at a furniture company wrote a manual for the greenhouse conversion. Similarly, nobody publishes an official guide for the specific workflow in your specific job. Agents are most interesting when applied to problems the builders never anticipated.
- Someone will do it for you. That Livingetc headline is the most telling item on the list. An entire market grew up around people who wanted the customized result without doing the customizing. That’s the shape of the agent market too. Some people will build their own. Most will buy something pre-hacked.
Why this matters if you’re not technical
If you’ve been reading about AI agents and feeling like you need to become an engineer to participate, the furniture trend is a reasonable counterargument. The people producing the best hacks aren’t cabinetmakers. They’re people who looked hard at an object, understood what it was actually made of, and had a clear idea of what they wanted instead.
That’s a design skill, not an engineering one. Knowing what you want is the hard part. Knowing what a tool can be pushed to do is the second hard part. The Allen key is not the hard part.
The practical version of this, for someone getting started with agents, looks like: pick a task you do repeatedly and resent. Describe the outcome you want, not the steps you’d take. Then see how far a tool gets on its own before you intervene. You’re not writing code. You’re doing the thing the greenhouse-cabinet person did, which is refusing to accept the default use case.
The part nobody puts in the manual
Hacks fail. Search any of those roundups and you’ll find caveats about weight limits, ventilation, finishes that won’t take paint. Modified furniture is less predictable than furniture used as designed, and that’s the trade you accept for getting something closer to what you wanted.
Agents carry the same trade. Give a system room to choose its own steps and it will sometimes choose badly. The correct response isn’t to abandon the approach, it’s to check the work, especially early, especially on anything that matters.
Which brings me back to the Allen key. It was never the interesting object in the box. It was just the thing that let you decide what the box became.
🕒 Published: