We show our working

Most tools give you an answer and ask you to trust it. Ours show you the working too. What we looked at, what we found, and the bits we couldn’t reach. Three tools, one for developers and two for the public service teams who have to get it right.

ULA DEV · StudioFrom source document to working prototype

Ula is the AI company that thinks AI sucks.

It sucks at design, it doesn’t build what you want, and it often doesn’t make things meaningfully better for end users.

We think the reason AI falls flat is because AI agents often don’t have the right context to know what you want or how to build it.

Give AI agents good enough scaffolding and they work really well, the “next token prediction” pushes outputs towards results you actually want. Scaffolding in terms of development means clear design docs, UI mock-up images and small working demos of features. Things that slap the slop outta the system.

Here at Ula we want to solve the problem of translating design ideas into exceptional scaffolding, to help AI build the real innovations that “make it perfect, no mistakes” somehow doesn’t achieve.

A bodybuilder hunched over a keyboard at a desk of monitors, captioned: “ok ChatGPT, make me an app that makes $1M /month. Don’t make any mistakes.”

Three features

The bits that make the difference.

Small working pieces, forms that bend to people, and an export your agent can actually read.

01 · ULAs

Small UI elements that make your systems pop.

Click the U at the bottom of the page to open the ULA maker. Describe any small UI feature and it builds one, right here. The idea is simple: make the small working feature, test it on a page, drag it, drop it, use it, and then it becomes part of a larger spec.

  • want a countdown timer, easy
  • want a tip calculator, now we’re talking
  • want a tip calculator that charges whoever ordered the lobster extra, good luck getting Claude to figure that out alone

You can test and improve the designs easily, and save the ULAs you make to reuse across pages and projects.

Made in the maker

Lobster surcharge splitter

Fields, a live total, and a note about the rounding. Drag it anywhere on the page.

02 · Forms

Next gen forms that give people what they want.

We think that in the age of AI, annoying forms have no right to exist. Filling in fifteen questions back to back is nobody’s idea of a good time. So we’ve created a standard where you decide what data you need, and Studio works out how to ask for it.

That could be a traditional static form. It could be a conversation that talks to people like people and picks up on what they’ve already said. It could be uploading a document instead of retyping your address, name and nationality. It could be breaking one horrible question into three that actually get answered.

The big idea: if your system asks users for a lot, it should be able to accommodate a lot of preferences for how they give it.

And whichever way someone answers, it’s the same form underneath. Same questions, same rules, same branching. We publish the tests that prove it, so you don’t have to take our word for it.

03 · Design packs

Design packs that would make Steve Jobs blush.

Getting your designs out of a design tool and into Claude isn’t easy. Direct Figma connectors don’t guarantee clean code, and plenty of developers are frustrated that AI comes back with slop.

You can try to prompt your way out of it, but without the right scaffolding it is very hard to explain the exact features and styles you want. Studio exports your drag-and-drop designs as a design pack you can hand to any agent. The spec, the images and the working UI demos, all in one go. Bumpers to keep the bowling ball out of the gutter.

Will it always be perfect? No. But it will always be better to prompt with supporting material than with text alone, and Studio lets you put that together quickly and painlessly.

Specbooking.contract.jsonv1.2
Images6 screensannotated
Demos3 ULAsworking
ULA GOV · CheckWatch a public service journey reveal itself
ULA GOV / CHECK

See the whole journey. Know what to fix first.

Give Check the address of a public form and it walks through it the way an applicant would, then stops before submitting anything. On purpose. What comes back is a ranked list of what gets in the way, each one showing the rule it breaks and a picture of where it happens.

Exploration complete

Garden waste permit

StartEligibilityAddress
7 prioritised findings

3 deterministic · 3 accessibility · 1 AI-assisted

Stopped safely before submission.

Bounded explorationIt works inside limits you can see, stops before it submits, and tells you which paths it never reached.
Structure, not trafficCheck sees your form the way a visitor does. It can’t see your traffic, and it won’t pretend to.
A working improvementA rebuilt version you can try, with every change listed, explained and sourced. Nothing gets published until a person signs it off.

Inspection you can challenge

Not another unexplained score.

Check keeps what it saw separate from what it worked out, tells you what it couldn’t reach, and every recommendation traces back to the thing that caused it.

01 · Explore

Map real paths without completing a transaction.

It records the questions, the branches, the validation and why it stopped where it did. Then it shows you the parts it never got to, instead of quietly leaving them out.

Steps mapped14
Branches3
Submitted0
Blocked1
02 · Evidence

Distinguish rules, accessibility tests and assisted judgement.

A rule that always applies, an automated accessibility test, and something the model noticed all look different on the page. Nothing we inferred is dressed up as something we measured.

Moderate barrier

Preparation is revealed too late

Observed on step 6 · Service Standard point 5

03 · Improve

Move from a report to something people can try.

Open the rebuilt journey and see every change beside the original, with the reason and the source. Disagree with one? Set it aside, and the record shows you did.

Before you start added

Documents and acceptable alternatives now appear before step 1.

Human review gate

6 accepted · 1 suppressed with rationale.

ULA Check demo

Review the evidence, then act.

Switch between the finding, its source and the improved version. This uses the same card, evidence, status and action patterns as the UI Lab.

Finding 4 of 7

The start page hides preparation work.

People only find out which documents they need once they’ve already started.

1
Observed

Document request appears on step 6.

2
Recommended

Add a “Before you start” list.

Continue the ULA GOV journey

Turn findings into a specification.

Check tells you what your service does today. Spec is about what it should do next. It turns findings, policy and what your team already knows into requirements precise enough to test, each one still pointing back at why it’s there.

ULA GOV · SpecDefine the service before delivery begins
ULA SPECIn design, not yet built

A definition of done that survives the handover.

One document that policy, service and delivery people can argue over before anyone writes any code. Every requirement keeps a link to the finding, document or interview behind it, so a reviewer can push back on the reasoning and not just the conclusion.

Specification ready

Garden waste service

Needs12approved
Rules18testable
Sources9linked

Structured and traceable

From policy intent to a verifiable contract.

Requirements stay joined to their reasons, the people who agreed them and the tests that prove them. Nothing in it says anything about your organisation that it can’t point at a source for.

01 · Define

Make requirements precise enough to test.

Turn policy intent and Check findings into service rules, user needs, data definitions and acceptance criteria that someone can actually test.

User needprepareBeforeStartapproved
RuleshowDocumentListrequired
EvidenceCHECK-004linked
02 · Align

Keep policy, service and technical decisions connected.

Reviewers can see where a requirement came from, argue with it on the spot, and see which tests change if it moves.

Ready for delivery

18 requirements approved · 0 unresolved conflicts.

Traceability

Policy → need → rule → test

Every delivery test retains the reason it exists.

03 · Assure

Give suppliers a contract everyone can verify.

A versioned spec means fewer arguments about what was meant, and you can see whether delivery matches it while it’s being built rather than at the final gate.

Requirements18
Testable18
Sources12
Unknowns0