Ooh, that’s a cool modal.
And it was. Clean layout, smooth animation, thoughtful spacing. The kind of thing that gets a few reactions in Slack. The kind of thing that makes you pause before you say anything.
But then you look closer. The border radius is off — not wrong, just… not ours. The shadow is doing something we never decided on. The interaction pattern works, but it’s not the one we spent three months debating. It came from nowhere, and now it exists, and someone is proud of it.
That’s the moment I want to talk about.
It has a name in my head.
creative bypass
Not malicious. Not lazy. Actually the opposite — someone motivated enough to go build something, to not wait, to ship. The tool was right there, it was fast, and it worked. The problem isn’t the intention. The problem is what got skipped in the process.
The design system. The documentation. The decisions that already happened. The why behind every token, every component, every pattern that the team spent real time on.
AI tools make the bypass frictionless. That’s their pitch.
Frictionless in the wrong direction is just faster drift.
And I get it. I really do.
These tools are genuinely impressive. You describe something, and in seconds you have a working component, styled, interactive, ready to drop in. No Jira ticket. No waiting for the design system team to prioritize your use case. No digging through documentation trying to figure out which token does what.
For someone trying to move fast — and everyone is always trying to move fast — it feels like the obvious choice. It feels like empowerment.
That’s not nothing. The instinct to build, to unblock yourself, to not wait — that’s exactly the kind of energy a product team needs. The tool didn’t create the problem. It just made a pre-existing shortcut a lot more tempting.
Shortcuts have a tab open somewhere.
Every component that lives outside the system is a component someone has to understand, maintain, and eventually reconcile. Or worse — nobody does, and it just stays there, quietly diverging. The modal that looked neat becomes the reference for the next thing someone builds. And then the next. And suddenly you have a parallel system nobody designed, nobody documented, and nobody owns.
The cost isn’t always visible at first. That’s what makes it dangerous. It doesn’t break anything immediately. It just adds weight. A slightly different shadow here, an undocumented interaction there. Small decisions made in isolation, without the context of everything that came before.
And the hardest part? You can’t blame the tool. The tool did exactly what it was asked to do. It just wasn’t asked the right question. Nobody told it about the three months of debate behind that border radius. Nobody fed it the decision log, the accessibility rationale, the edge cases the team already hit and solved.
That’s not information a prompt can carry. That lives in the system. Or it should.
The answer isn’t to ban the tools. That’s not realistic, and it’s not the point.
The point is to make the system the path of least resistance. Faster than the bypass. More trustworthy than a prompt. The kind of resource someone actually reaches for because it genuinely unblocks them.
That means two things.
First, internal tooling that connects the dots. Not another third-party layer to maintain — something built for your ecosystem, your tokens, your patterns. Tooling that speaks the language of your system because it was built inside it. When that exists, the AI tool stops being the shortcut. Your system is the shortcut.
Second, documentation that explains the why. Not just what a component does, but why it exists, what problem it solved, what was considered and rejected. The decisions that happened in a room or a Figma file or a long Slack thread — written down, findable, human. That’s the context no AI tool has access to unless you give it. And when you do, suddenly your system becomes something people trust instead of something people route around.
The modal that started this wasn’t a failure of ambition. It was a failure of access.
Someone needed something, didn’t know the system had it or could support it, and went elsewhere.
That’s fixable. Not with a policy. With better foundations.
So next time someone shares a modal in Slack and the reactions start rolling in — pause before you say anything.
Not because the work isn’t good. Maybe it is. But ask where it came from, and whether the system could have gotten them there faster. If the answer is no, that’s not a people problem. That’s a systems problem.
And that one is yours to solve.