AI-DRIVEN DEVELOPMENT

Built fast with AI.
Checked by real people
before we hand it over.

With AI, software takes shape remarkably fast. But something built quickly is not automatically something your team can use. Before we deliver, we have people who actually do that job sit down with the screens, and we fix what tripped them up.

This “field test” is part of every standard plan. There is no extra charge for it.

Talk to us (online, 30 min, free) See pricing and process

Our founder handles your project personally, from first call to handover. We explain everything without jargon.

AI Builds fast, builds a lot
HUMAN The people who will use it try it

Whatever they struggled with, we fix before delivery

ISSUES

That system you had built —
is anyone still using it?

You rolled it out, and everyone quietly went back to managing things in spreadsheets.

“The old way was faster.” It never took hold.

You hear AI makes it cheap and fast, but you are not sure about the quality.

Nobody in-house is technical. There is no one to ask in the first place.

Software nobody uses is not a new problem — it long predates AI.
Most of the time the cause is the same: it was built without the people who would actually use it.

BACKGROUND

AI produces mistakes
quickly and at scale, too.

AI builds exactly what it is told, very fast. But if the instruction is vague, it will confidently build a great deal on top of that misunderstanding. It looks like it works. “Working” and “usable in daily operations” are different things. Approval rules, exceptions, who is allowed to see what, how it meshes with the way your team already works — none of that is decided by code alone.

8 What AI can build quickly The look of the screens, the common features, the basic path of entering and saving. This part comes together astonishingly fast.
2 What only people can close Exceptions, differing permissions, the fit with how work is actually done. This is where we spend our time.

  • Industry testing specialists note that quality is ultimately guaranteed by people, and that human review and testing matter more, not less, in AI-driven development.

    SHIFT ASIA
    View source

  • In a global developer survey, 66% said AI answers are “almost right, but not quite.” The professionals building with it feel this themselves.

    Stack Overflow
    Developer Survey 2025
    View source

  • AI raises productivity, but without a way to verify the output, rework and defects tend to increase.

    2025 DORA Report
    （Google Cloud）
    View source

AI output is “almost right.”
In daily operations, that “almost” is what kills adoption.

OUR ANSWER

We use AI for speed.
People do the checking.

PROMISE 01

We shorten the time it takes to build

We use AI for design and implementation, so the early stage that used to take months is much shorter. We get to something real quickly, and show you quickly.

PROMISE 02

Before handover, real people actually use it

People who do that job for a living sit down with the unfinished screens. Every place they hesitated, every place they stopped, gets recorded.

PROMISE 03

Whatever they struggled with, we fix

We do not hand you a report and call it done. We hand you the fixed version. You will not need to go to another firm to have it sorted out.

Because AI finishes the build early, we can afford the time to have people try it before delivery.
Speed is what pays for the checking.

FIELD TEST

The field test —
real users try it before you receive it.

The people who built it never get lost. They know how it was made. Only someone touching it for the first time can find what is hard to use.

Who tries it

From our partner database of test collaborators, we select people close to your customers or your staff. Reception staff, accounting staff, people working on site — we can gather by occupation.

This is not a database for sourcing developers. It exists to gather people on the using side.

Partner database of test collaboratorsRegistered by occupation, age group and comfort with technology
Narrowed to match your actual usersWe specify the industry, the role and the situation of use
Final participants confirmedNumbers are set per project, and increased when needed

What we look for

Not “does it work,” but “can they finish without hesitating.” We use the same set of checks every time and add project-specific ones.

  • Can they reach the end without being told how?
  • How many clicks does the everyday task take?
  • Do the words on screen make sense to someone in that role?
  • When they enter something wrong, can they undo it?
  • Is it readable for people who are less comfortable with technology, or older?
  • Can the same task be finished on a phone?

How many people

It is generally held that testing with a small number of people surfaces most usability problems. We set the number per project on that basis, and increase it when the work is complex or when there are several distinct kinds of user.

Source for the approach: Nielsen Norman Group Why You Only Need to Test with 5 Users (the proportion found depends on conditions and varies)

What it costs you in time

We recruit the participants. Your staff are not pulled away from their work, and you do not pay their compensation.

All we ask of you is to look at the results and tell us which things to fix and which are fine as they are.

What you receive

  • Short clips of just the moments where people got stuck
  • A list of the usability problems (where, how many people, what confused them)
  • A record of what we changed and how

We do not produce a thick report. Rather than adding to your reading, we hand you the fixed product. Because we also build, we can respond with repairs instead of findings.

Cost

The field test is included in every standard plan. There is no “testing fee” line in our pricing.

Commissioning usability research separately normally means a separate order, a separate quote and a separate schedule. We do it inside the same contract as the build.

PROCESS

What AI handles,
and what people handle.

AI-driven development is not magic. At each stage we let AI do what it is good at, so people can concentrate on judgement and verification. That is all it is.

Decide
Design
Build
Verify
Hand over
AI handles
Turns the conversation into a written record on the spot
Drafts the screens and the shape of the data
Writes the code, and applies changes immediately
Runs the automated checks across the whole thing
Drafts the documentation
People handle
Ask what is actually hard, and agree what we will not build
Settle the exceptions and permission differences. Only people can decide these
Show you something working every week and keep the direction aligned
Field testReal users try it, and we fix what they found
Finish the operating guide and handover documents

Scroll sideways to see the full table. The highlighted cell is what sets us apart.

When you get to see
what, exactly.

You will never be left with nothing to look at until delivery day. Timelines depend on scope, so we give you a dated schedule before we start.

  1. First call

    A 30-minute online conversation

    No preparation needed. Just tell us where your time is going right now. There is no charge.

  2. Week 1

    Scope, schedule and quote

    We also write down what we will not build. We start only once you agree to it.

  3. Early on

    You see a first version of the screens

    Not a written specification — screens you can actually click.

  4. Every week

    You see whatever is working at that point

    “Actually, could we change this?” is welcome here. Only a large change of scope means a revised quote.

  5. Before handover

    Field test day — real users try it

    Your staff are not pulled away. Whatever they struggled with, we fix before you receive it.

  6. Handover

    You receive the fixed version

    The full source, the handover documents, and the record of the field test. From that day, defect fixes are free for three months.

WHAT WE BUILD

This is the kind of thing we build.

If you are not sure which one fits your situation, just ask. If it is not something we should take on, we will say so plainly.

Replacing spreadsheet work

For example: several people editing the same file and overwriting each other, or hours a day spent copying figures from one place to another — brought into a single screen.

Internal business systems

For example: requests and approvals, stock or project management, daily reports and attendance. Who can see what is designed in from the start.

Booking and ordering

For example: bookings and orders currently taken by phone and email, turned into something customers enter themselves. We can connect to the calendar you already use.

Automating enquiry handling

For example: an AI chatbot on your website or LINE. It can also answer from your own manuals and past quotations.

Connecting the tools you already have

For example: linking your sales, accounting or chat tools by API so the copying and totalling disappears.

Connecting to your website

For example: enquiries from the site flowing straight into an internal dashboard, or letting your team update pages themselves.

TOOLING

We will show you the tools we use.

Names alone do not tell you much, so there is a plain-language line for each. We use all of these daily.

  • Claude / Claude CodeAI that helps write software. We use it from design discussions through to implementation.
  • ChatGPT / GeminiAI chosen per task — generating text, interpreting images, and so on.
  • GitHubWhere the code is stored, with a history of who changed what and when.
  • Python / TypeScriptThe languages the system is actually written in. We choose to fit the job.
  • Cloudflare / AWSWhere the finished system runs. We choose to fit the scale and the budget.
  • PostgreSQL / SQLiteWhere the data lives. Built so it can be exported and moved elsewhere.

AFTER DELIVERY

Delivery is not
where we stop.

What you receive, and what happens after

You will not be locked in to us. We hand it over in a state another firm could pick up.

  • What we build belongs to you. Whether or not AI was used to write it, the rights to the delivered work are yours.
  • You get the source itself. Nothing is delivered as a black box. You can hand it to another firm.
  • Handover documents are included. Both an operating guide and a record of how it is built.
  • We record why it was built that way. So that later on, it does not become something nobody can safely change.
  • Three months of defect fixes after delivery. For three months after handover we fix defects at no charge. The same period is stated in the contract.

How we handle your information

Written so you can use it directly when explaining this internally. Anything unclear is settled in writing before we start.

  • Never used to train AI. We do not provide information you entrust to us as AI training data.
  • An NDA before we start. Your template or ours, whichever you prefer.
  • Personal data is replaced with dummy data. Development and testing never use real individuals' information.
  • Test participants are bound too. They confirm before taking part that what they see stays private.
  • Security is checked before delivery. Because code written by AI can be syntactically correct and still unsafe.

PRICING

We publish our prices up front.

A number on its own tells you nothing, so the duration and what is included sit right next to it. The field test is in all of them.

Try it first

For seeing what is possible

FromUS$650+

  • One working prototype, quickly
  • Field test included
  • A record of how people used it
  • Credited against full development
Ask about this plan
Most requested

Put it to work

For using it in the real workflow

FromUS$3,200+

  • Everything needed to run in production
  • Field test included (size to match the project)
  • Handover documents and operating guide
  • Three months of defect fixes after delivery
Ask about this plan

Keep improving it

For growing it as you use it

QuoteMonthly, on quotation

  • Monthly improvements and new features
  • A field test with every significant change
  • Operational support and monitoring
  • Cancel whenever you like
Ask about this plan
  • The field test is included in every plan. There is no “testing fee” line in our pricing, and we never add one later.
  • What you pay for “Try it first” is credited against full development if you decide to go ahead.
  • Payment is split in two: on starting and on delivery. We never bill additional hours on top.
  • Scope is fixed in writing before we start. Including what we will not build, so nothing is open to interpretation later.
  • Prices vary with scope and duration. The figures above are indicative; we quote properly once we have talked.

Judge us on how we work, not on how many logos we have.

We are a young company. On the sheer number of past projects, we cannot compete with the large firms — and we are not going to pretend otherwise.

What we do instead is show you how we work, what is included and what it costs — all of it, on this page, before you contact us. The process, the checks, the deliverables, the contract terms.

Our founder handles your project directly from first call to handover. The person you talk to is the person who builds it.

FAQ

Questions we are often asked.

Yes. We explain things without jargon. The only thing you need to decide is what is causing you trouble. We propose what to build and which technology to use.

It varies with scope. “Try it first” is short; something for real operational use takes longer. We give you a dated schedule before starting, and we work to it. We do not write dates we cannot meet.

“Try it first” starts at US$650 and “Put it to work” at US$3,200, as a guide. The field test is included in both. Ongoing work is quoted monthly.

That is fine. You see something working every week, and changes can be raised there and then. Only a large change of scope means we re-issue the quote.

It is a fair concern. That is exactly why we have real users try the screens before delivery and fix what they struggled with. This step is in every standard plan.

Projects have been abandoned for exactly that reason. We take the checking on ourselves — recruiting participants and covering their compensation — and all we ask of you is the final judgement call on the results.

You do. The rights to the delivered work are yours, and you get the source itself. We do not retain rights or restrict how you use it. This is written into the contract.

We do not provide information you entrust to us as AI training data. Where personal data is involved, we replace it with dummy data before working on it. An NDA is signed before we start.

We are, regardless of whether AI wrote it. For three months after handover we fix defects at no charge. That period and the scope it covers are stated in the contract.

Research indicates that code written by AI can be syntactically correct and still not safe (Veracode 2025 GenAI Code Security Report). We therefore run security checks before delivery, and for sensitive systems we can arrange an external assessment.

Preventing that is the whole point of testing before delivery. If something still proves hard to use, we fix it at no charge for three months after handover — and beyond that, the ongoing plan covers it.

Yes. You receive the full source and the handover documents. Nothing is built so that only we can touch it. If you like, we will join the handover call with your new provider.

RELATED

Our workWebsites, business systems and AI agents we have built. AI co-creation partnersWe build a prototype from your operational problem at no cost. We are looking for companies willing to test it. Taskal — our sister serviceInternal tools, enquiry handling and the everyday problems in between, taken off your hands.

Start by telling us
what is getting in your way.

Thirty minutes online, at no cost. No preparation needed. A human writes the reply.

Talk to us