The Real Cost of Skipping Accessibility Until Later
"We'll add accessibility later" is a promise almost nobody keeps, and the bill comes due with interest.
Every project has a moment where someone says "let's ship first, we'll make it accessible later." It sounds reasonable. It is almost never true. Later becomes a backlog item that never wins a sprint planning fight against a new feature, a client deadline, or a bug that's actually on fire.
The deeper problem is architectural, not procedural. Accessibility is not a layer you paint on top of a finished interface. It's woven into how you structure markup, manage focus, name components, and handle state. Retrofit it after the fact and you're not adding a feature, you're renovating a house around load-bearing walls you didn't know were load-bearing.
Consider a modal dialog built without accessibility in mind. It probably uses a div with a click handler, no focus trap, no ARIA role, no escape key handling, no return-focus behavior. Fixing that later doesn't mean adding a few attributes. It means rebuilding the component's interaction model from the ground up, then re-testing every place it's used, because by then it's been copy-pasted into fourteen other screens.
The cost compounds because accessibility bugs don't announce themselves the way visual bugs do. A misaligned button gets caught in a five-second glance at a screen. A form that traps keyboard focus, or a live region that never announces an error, can ship for months before anyone with a screen reader hits it, and by then it's touched by a support ticket, a legal complaint, or a lost customer, not a code review comment.
There's also a team-morale cost that's easy to underestimate. Engineers who are asked to retrofit accessibility into a mature codebase experience it as punishment work: tedious, unrewarding, and disconnected from anything they designed. Engineers who build with accessibility from the start experience it as just... building the thing correctly. Same outcome, wildly different experience, and the difference in quality shows.
None of this means accessibility has to slow you down early on. Semantic HTML, sensible focus order, and real labels cost almost nothing when they're the default instead of the exception. The complexity isn't in doing it right the first time. It's in believing you can cheaply undo not doing it.