The design decisions behind Makeform.ai's form generator

The category is crowded. "AI form generator" is now a phrase you can put on a landing page without explaining further. So the interesting question about Makeform.ai is not what it does — type a description, get a form — but how the team decided what not to ship.

Form builders are deceptively hard. The schema is simple. The behaviors are not.

What the team cut

Talking to the people building Makeform, the recurring theme is removal.

• They cut conditional logic from the v1. Not because it is hard, but because most AI-generated forms with conditional logic were unusable.
• They cut "AI completes the user's answer." It is a great demo. It killed conversion in tests.
• They cut multi-page wizards as the default. Even when the user prompt implied them. Long forms have lower completion. The AI did not know that. The team did.

Cutting features in v1 is the kind of decision that does not appear on a roadmap doc. It only shows up in retention.

The interesting decision: what counts as a "form"?

A form, on the surface, is a list of fields. The Makeform team treats it as something stranger: a small program that runs across a human interaction. That reframing matters because it forces them to think about validation, defaults, accessibility, and abandonment as first-class concerns.

You can see this in small places. The default field types are conservative. The labels are short. The required-field rules are sane. None of this is AI magic. It is taste, calcified into defaults.

The AI is, charmingly, the cheapest part of the product. Swap GPT-4-class for Claude-class or Gemini-class and the output barely changes. Swap the defaults and you get a different product.

A category bet worth watching

There are roughly three plausible futures for AI form builders.

1. The category collapses into general-purpose AI app builders. "Form" becomes a sub-mode of "build me a tool."
2. The category collapses into the incumbents. Typeform, Tally, and Google Forms quietly ship "generate a form" as a button.
3. The category remains its own niche, because forms have annoying, specific requirements (compliance, embed targets, payment integrations) that general tools are bad at.

I would not bet on option 1 yet. The general-purpose tools are good at code; they are not good at the small editorial decisions that make a form pleasant to fill out. Option 2 is real but slow — incumbents tend to ship AI features that are technically correct and emotionally dead. Option 3 is where Makeform seems to be playing, and it is a more honest game than the marketing makes it sound.

What good looks like

The test I keep applying to AI form tools is whether the output is something I would actually deploy. Not "would this be useful as a starting point" but "would I, today, put this in front of a paying customer."

Most fail. The AI invents fields that the prompt did not ask for. The styling is generic. The validation is wrong in subtle ways.

Makeform mostly passes, and the places it fails are interesting. It does not, for example, do an especially good job at multi-step forms that branch on previous answers. It is fine at translating "an event signup with food preferences" into something I could ship by lunch. That gap — between the well-trodden case and the long tail — is where the team's next year of work probably lives.

That is a more honest pitch than "AI form generator." It is also, I suspect, the one the founders would give you if you asked them off the record.