AI-Assisted Development: Three People, Twenty People's Work

"We have no IT department." "The budget is limited." "We don't even know where to start." The reasons small businesses hold off on building systems are always the same.

Since 2024, though, rapid progress in AI has changed the picture considerably. Serious system development, once the preserve of large companies, is now achievable by a team of one to three engineers.

This article explains what AI-assisted development actually involves, with real numbers and the names of the tools, based on the corporate sites, membership sites and booking systems motiv has built. Not an abstract claim that AI makes development faster, but which AI, used how, and what changes as a result.

AI is now lifting programmer productivity by a factor of five to ten. With AI assisting across design, implementation, testing, debugging and documentation, a single engineer can cover what used to take five.

■ Where AI is particularly strong

• Generating routine code: boilerplate, CRUD, forms, authentication flows

• Refactoring existing code: readability, performance, type safety

• Investigating bugs: time from error message to root cause cut to a fifth

• Writing tests: initial unit and end-to-end tests can be delegated

• Documentation: first drafts of READMEs and API docs

This overturns the assumption that a large team means high quality. A small team decides faster, loses nothing in communication, and generates no wasted process. Now that AI has raised productivity so sharply, a small team is not the cheap option — it is the rational one.

motiv uses several AI tools, each for a different purpose.

■ Claude (Anthropic)

Used for: design, complex logic, reviewing long stretches of code

Strength: it holds a great deal of context, which makes it good at changes spanning many files and at design discussions. The code quality is high and it is careful about safety. We use it for the main implementation work.

■ GitHub Copilot

Used for: real-time completion in the editor, generating short functions

Strength: integrated into the editor, predicting the rest of a line from the first few characters. It halves the typing and keeps you in the flow of thinking while writing.

■ Cursor / Claude Code

Used for: larger changes that require understanding the whole project, refactoring across many files

Strength: it proposes changes with the whole codebase in view, so the result stays consistent. It works from plain-language instructions like "add this feature".

■ ChatGPT (OpenAI)

Used for: technical research, comparing libraries, explaining concepts

Strength: a broad knowledge base, useful when investigating an unfamiliar stack or weighing up options.

Switching between these four according to the task is what maximises development speed.

Here is the comparison in concrete terms.

■ Conventional (large team)

• Timeline: six months to a year

• Team: 5–20 people (project manager, designer, front end, back end, QA, infrastructure)

• Cost: ¥5 million to several tens of millions, mostly labour

• Communication: 40% of the time goes on meetings, specifications, Slack and ticket management

• Maintenance: a separate contract with a separate team; every change of specification means a new quote

■ AI-assisted (small team)

• Timeline: three weeks to three months

• Team: 1–3 people (full-stack engineers plus AI)

• Cost: ¥1 million to ¥5 million

• Communication: talk directly, implement immediately, review a working demo

• Maintenance: the same team continues; the people who built it also run it

■ But is the quality really there?

This is the most common question, and the answer is that quality tends to improve. There are three reasons.

1. **A modern stack**: Next.js, React, TypeScript, Tailwind CSS — the same world-class tooling used by Google, Meta and Vercel. Against an ageing PHP-based system, the difference in speed, security and maintainability is stark.

2. **AI review**: code is reviewed by AI as well as by people, so far less is missed.

3. **Clear ownership**: with a small team, nobody has to ask who wrote this. Every feature has an obvious owner.

More and more agencies claim to use AI, and the reality varies. Check these five points before you commission anyone.

■ 1. Can they name the tools they use?

"We use AI" on its own is not an answer. Choose a company that can name Claude, Copilot, Cursor and explain what each is for.

■ 2. Will they show you what they have built?

A demo site, a public repository, a service in production — a company that will show you what it has actually built with AI is one you can trust. motiv's corporate site, membership site and booking system are all publicly visible.

■ 3. Are the timeline and cost realistic?

Even with AI, discovery, requirements, design and operational planning take time. Be wary of anyone claiming they can build anything in two weeks. A quote in the three weeks to three months range is the realistic one.

■ 4. Are they open about the stack?

Look for a modern setup — Next.js, TypeScript, PostgreSQL, Cloud Run. A company working on an ageing CMS base with heavy plugin dependence will not see much real benefit from AI.

■ 5. Will the same team handle maintenance?

Some suppliers stop at delivery and treat operation as a separate contract. In an AI-assisted model, the same small team ideally runs the system too, so changes can be made at the same pace.

■ A good fit

• Small businesses nervous about a first system project

• Owners who do not know where to begin

• Anyone who has had a quote from a large systems integrator and found it too expensive

• Projects that go beyond a website — booking, payments, membership management

• Anyone who wants the same partner for maintenance afterwards

• Anyone who wants a partner who takes the trouble to understand their industry

■ Not a good fit

• Modifying part of an existing large-scale core system (a major integrator suits this better)

• Sectors with extremely strict compliance requirements, such as finance or healthcare

• Projects where the only criterion is the lowest possible price (AI-assisted work still costs what it costs)

We start by listening to your situation and will tell you honestly whether this approach is the right one. Feel free to come to us at the "is this even possible?" stage.

AI has lowered the barrier to system development dramatically. A small team, AI, and a world-class stack together mean a small business can now have a system matching or exceeding what a large company would get — in three weeks to three months, for ¥1–5 million.

The point is not that AI lets anyone build cheaply. It is that you should work with engineers who are good at using it. Use the five checks above when choosing a supplier.

motiv builds and runs its own corporate site, membership site and booking system this way, and brings that experience directly to client projects. Please get in touch.