Skip to content
All posts
Custom Software

Custom Software Development in the Age of AI: What Changed

AI made writing code fast. Deciding what to build and proving it works did not speed up. What that means for the price, the timeline and who you hire.

For as long as custom software development has existed, the expensive part was typing it. A quote was mostly a count of developer hours, and the hours went into writing code. AI coding tools have changed that. A working screen that once took a developer a day can now appear in minutes.

So business owners reasonably ask: if the code is nearly free, why is custom software not nearly free, and why does it not arrive the same afternoon?

Short answer: AI removed the typing from custom software development and left the two jobs around it. Design is deciding exactly what the software should do in every situation. Verification is proving that it does. Those are now the bottleneck, they are where your money goes, and they are what to judge a developer on.

What did AI actually make faster?

Writing code that matches a clear description. Give an AI tool a precise instruction ("a form with these six fields that saves to this table and emails this person") and it produces that form quickly and usually correctly.

That is a real change. Work that was slow because of volume is no longer slow: standard screens, forms, reports, connections between two well-documented systems, converting data from one shape to another.

Two things did not get faster, because they were never typing.

Why design is now half the job

Software does exactly what it is told, including in the situations nobody thought about. Take a booking system for a business with several locations. Before any code exists, somebody has to decide:

  • Can a customer cancel an hour before the session, and do they get their money back?
  • What happens when two people book the last slot at the same moment?
  • Who is allowed to see another location's bookings?
  • When a staff member leaves, what happens to the bookings in their name?
  • If the payment goes through but the confirmation email fails, what does the customer see?

None of these are coding questions. They are questions about how your business runs, and the only person who can answer most of them is you. A developer's job is to know which questions exist, ask them before the build starts, and write the answers down.

AI does not remove this step. It makes skipping it more dangerous. Ask an AI tool for "a booking system" and it will answer every one of those questions for you, silently, with a plausible guess. You get software that looks finished and runs on rules nobody chose.

When code was slow to write, a vague brief showed up early, because the developer got stuck and came back with questions. Now the tool does not get stuck. It fills the gap and keeps going. The cost of a missing decision moved from the start of the project, where it was a question, to after launch, where it is a refund dispute.

Why verification is the other half

The second job is finding out whether the software does what was decided. This used to be partly covered by the pace of the work: a developer who wrote every line by hand had at least read every line.

With AI, code arrives faster than anyone can read it carefully. It also arrives looking right. The screen loads, the button works, the demo goes well. The problems that matter are the ones a demo does not show:

  • A logged-in user who changes a number in the address bar and sees someone else's records.
  • A calculation that is correct for the example you tried and wrong when the quantity is zero.
  • A change made on Tuesday that quietly breaks something built the week before.
  • A backup that has been "running" for months and has never been restored.

Finding these takes deliberate work: tests that run automatically on every change, a second person reading the code, a copy of the system where real scenarios are tried before customers see them, and monitoring that tells someone when it fails at night. None of that is new. What is new is the proportion. When writing took most of the time, checking was a slice at the end. Now checking is a large share of the whole job, and a build that skips it is the kind of app we see on a vibe code cleanup.

So has AI made custom software cheaper?

Somewhat, and unevenly. Here is where the effort in a build sits now, in general terms.

Part of the build Before AI tools Now
Deciding what it should do Often rushed, to get to the coding The main event. Mistakes here are copied into everything quickly.
Writing the code Most of the hours A small share of the hours
Checking it works A phase near the end, often squeezed Continuous, and a large share of the work
Running it after launch Hosting, fixes, updates Unchanged

The practical effects for a buyer:

  • Standard work costs less than it used to. If your project is mostly ordinary screens and well-understood connections, there is less labour in it.
  • More fits in a version one. The same budget covers more features than it once did, as long as each is designed and checked properly.
  • The floor did not drop to zero. A project still needs someone to work out the rules with you and prove the result. That time is governed by how complicated your business is, not by how fast code can be typed.
  • Cheap is easier to fake. Anyone can now produce something that demos well for very little. The difference between that and software you can run a business on does not show on the screen.

Be wary of any quote that seems to price only the code.

What you are paying a developer for now

If the typing is cheap, the value of a developer is in four things:

  1. Asking the questions you have not thought of, before the build, and writing down your answers.
  2. Choosing a structure that survives change: how the data is organised, who is allowed to do what, where the system connects to your other tools.
  3. Proving it works, with tests, review and a staged copy you can try yourself.
  4. Being accountable afterwards. An AI tool does not answer the phone when the invoices are wrong.

That gives you better questions for any developer, including us:

  • Before you write anything, what will I get in writing about what the system does?
  • How do you check code an AI tool wrote before it reaches my customers?
  • Can I try the software on a private address while it is being built?
  • If it does not work the way we agreed after launch, who pays for the fix?
  • Who owns the code and the accounts it runs on?

A developer who talks only about how fast they can build is describing the part that stopped being hard.

How Rootech works with this

Rootech is a Melbourne software studio working Australia-wide, with delivery outside Melbourne done remotely. The way we run projects lines up with where the work now sits:

  • Scope and price are agreed in writing before you sign. That written scope is the design work: the decisions are made at the start, where they are cheap to change.
  • You test it as it is built. The build runs on a live address you can open, so checking starts in week one and not at handover.
  • Bug fixes on anything we built are included after launch. If it does not work the way we agreed, fixing it is our cost. That only works for us if we check properly first.
  • You own the code, and the accounts stay in your name. Launch comes with a handover of the code, documentation and credentials.

Most builds ship in two to four weeks. Bigger ones get a fixed timeline up front.

If you are weighing up a build, or you already have an AI-built prototype and want to know what stands between it and real users, tell us about it at /#contact. For the practical side of this, read AI Wrote the Code. How Do You Know the Software Works?.

Frequently asked questions

Can I just build my business software myself with an AI tool?

For a prototype, or a tool only you use, often yes, and it is a good way to work out what you want. The risk starts when other people's data, money or bookings depend on it, because the tool has made design decisions you have not seen and nobody has checked the result. See From Prototype to Production for what changes at that point.

Why does a custom build still take weeks if AI writes code in minutes?

Because the weeks were never only typing. They go to agreeing what the software should do in each situation, connecting it to the systems you already use, trying it against real scenarios and fixing what that turns up. AI shortened one of those.

Is AI-written code worse than code a person wrote?

Not inherently. Unchecked code is worse than checked code, whoever wrote it. AI makes it possible to produce a great deal of unchecked code very quickly, so the checking has to be deliberate.

What should I prepare before talking to a developer?

A plain description of the job the software does, who uses it, and the awkward cases you already know about: the exceptions, the customers who do it differently, the thing that goes wrong every month. The awkward cases are the most valuable part of a brief.

Does a fixed price still make sense when the tools keep changing?

Yes. A fixed price is a price for an agreed outcome, not for hours. Faster tools change how the developer gets there. They do not change what was promised.

Free AI workflow audit

Wondering what this looks like for your business?

We’ll map where your hours and leads are leaking, and hand you a prioritised plan with dollar figures attached. The audit fee comes off your first project.

Get in touch