Our engineers use AI coding assistants every day. They write boilerplate, generate tests, explain unfamiliar code and draft whole features from a description. Our delivery is faster. But it is not ten times faster, and it is worth understanding why.
Where the time goes in a real project
| Activity | Share of a typical project | AI impact |
|---|---|---|
| Understanding requirements and decisions | 20–30% | Small |
| Design and architecture | 10–15% | Moderate |
| Writing code | 25–35% | Large |
| Testing and fixing | 15–20% | Large |
| Deployment, data migration, training | 10–15% | Moderate |
AI compresses the middle of the project dramatically. The edges, where people have to agree what to build and adopt it afterwards, still move at human speed.
What gets genuinely better
- Prototypes in days rather than weeks, so you can react to something real sooner.
- Much higher test coverage, because writing tests is no longer tedious.
- Faster upgrades of old code and frameworks.
- Better documentation, generated alongside the code.
The risks to manage
AI-written code can look right and be subtly wrong. It can introduce security issues, use outdated patterns or pull in unnecessary dependencies. Every change still needs review by an engineer who understands the system, plus automated tests and security scanning.
How we use AI in our delivery process
- Discovery: AI summarises workshop notes and drafts user stories, which the team then reviews with you.
- Prototyping: clickable prototypes and first working screens are generated quickly so you react to something real.
- Development: engineers use AI assistants for boilerplate, tests and refactoring, and own every line that ships.
- Review: every change gets a human code review plus automated tests, linting and security scanning.
- Documentation: API docs, handover notes and user guides are drafted with AI and checked by the team.
What this means for your budget
Expect faster prototypes and more features within the same budget, rather than a project that costs a fraction of before. The savings show up as better test coverage, quicker iterations and less time waiting for the first working version. Be wary of quotes that assume AI makes the whole project several times cheaper; the hard parts are still human.
Signs a vendor uses AI well
- They show you working software early and often.
- They can explain any part of the code they deliver.
- Automated tests and security scans run on every change.
- They are clear about which AI tools touch your code and data, and under what terms.
AI makes good engineers faster. It does not make unclear requirements clear.
