AI has made us faster at software development

More output in the same time

Not long ago, a senior developer at our studio would spend a full day setting up a typical CRUD module: data model, validation, API endpoints, tests, front-end form, documentation. Today the same task is done in a fraction of the time.

Not because the developer has become smarter. Because the tools have.

Claude Code, Cursor, and context-aware assistants that understand the entire codebase, not just the file you have open, have changed how much an experienced person can accomplish in a day. For mechanical work, and there is more of it in software development than people admit, the gain is dramatic.

Where the effect is genuine

Across multiple projects, these are the areas where AI has delivered the greatest time savings:

Data modelling and migrations. Writing a Postgres table, setting indices, creating migrations, and generating TypeScript types is now almost pure translation from a business description to code. AI gets the bulk of it right; we fine-tune the rest.

API endpoints and validation. Standard CRUD patterns, authentication, authorisation, input validation, error handling. Everything follows conventions that AI has seen thousands of examples of. Previously we wrote it. Now we review it.

Tests. Unit tests, integration tests, edge cases. AI is good at generating a broad test suite quickly. We ensure the tests are testing the right things, but the volume work is gone.

Refactoring. Does a data model need to change? A field split out? A service extracted? AI can carry the change consistently through the entire codebase, faster than we did manually and with fewer mistakes.

Documentation. The factor that was traditionally deprioritised because it was tedious can now be generated from the code and updated automatically. Onboarding new developers goes faster.

In summary: routine work that used to take a full day is now done in far less time.

Where the effect is overstated

But AI has not changed everything. These areas take equally long, or longer:

Complex architecture. How the system should be structured, which services should be separate, how data should flow between them, how domain objects are modelled. This requires business understanding and trade-offs that AI cannot make without the context we give it.

Integrations against legacy systems. When we talk to line-of-business systems from 2007 with poorly documented SOAP APIs, or a public agency with proprietary XML exchange, AI helps minimally. It is debugging, reading documentation, and direct communication with the vendor.

Security analysis. AI can flag obvious vulnerabilities, but threat modelling, assessing what is sensitive, and thinking like an attacker requires human experience with concrete environments.

User interviews and process mapping. Understanding what a case handler does all day, which exceptions are most common, and how many different scenarios must be handled requires sitting alongside the people who will use the system. AI helps synthesise notes, not conduct the interview.

Assessing release readiness. Whether something is ready for production, whether edge cases are covered, whether it will withstand real use, those are judgements based on experience, not code patterns.

What this means for the price you pay

We have made a deliberate choice to pass the gain on.

An MVP of a line-of-business system costs noticeably less today than it used to for the same scope and quality, and the same applies to larger case-handling systems. The final price is set as a fixed price after the pre-project.

Not because we earn less. Because we accomplish more per day, and projects take fewer calendar days. We have the same number of developers on the same project, but deliver in less time.

For you as a client, this means:

• Lower fixed price for the same deliverable
• Faster time to market
• More iteration within the same budget
• Less risk, because less capital is tied up for longer

For us it means:

• We can afford to do more of the project ourselves, fewer subcontractors
• We can take on more projects in the same period
• We can invest more in pre-project work, security and documentation, because we have the time

The risks we have seen

It is not all straightforward. We have seen concrete pitfalls, both at our studio and elsewhere.

AI-generated code that looks good but has not been tested in full. It is easy to let AI write 200 lines of code that compiles and works on the happy path, but fails on edge cases. Discipline around testing is more important than before, not less.

Security vulnerabilities that look harmless. AI recognises obvious anti-patterns, but not context-dependent problems. A SQL injection in an internal admin route is still a SQL injection. We do a manual security review on everything running in production.

Consistency problems. If five AI-generated modules use five different patterns for the same problem, the codebase becomes difficult to maintain. We set explicit conventions and hold to them.

False productivity. Producing more code faster is not always a gain. If the code solves the wrong problem, or solves a problem that did not need solving, AI has only made it faster to build the wrong thing. We spend as much time on clarification as before.

The new normal

The expectation of how long custom software development takes is shifting. Deliverables that previously took many months can often be done faster today, with quality that is equally good or better.

If you are evaluating a development partner, ask directly:

• Which AI tools do you use daily in your development work?
• Which specific projects have you launched in the past year with significant AI assistance, and what has that meant for delivery time?
• How do you handle quality assurance of AI-generated code?
• Does your pricing reflect that AI tools have changed productivity, or are you still billing on the old model?

A partner who has not integrated AI into their workflow is billing for work that no longer requires as many hours. A partner who has integrated AI but has not tightened quality control is delivering code with hidden defects. Both are red flags.

Want to see how we work in practice? We are happy to walk through a previous project and explain how AI tools were involved from day one. Send us a short note about the project you are considering.

You can also read more about how we work as a development partner, or about what we offer in app, system and AI development.