How We Use AI at Kodenique: A Look Inside Our Development Process

We've written about how AI is changing software development in general. This post is more specific: how we actually work with AI day to day, what we refuse to automate, and what that means if you hire us. Transparency here matters — if you're evaluating a development partner, "do you use AI?" should get a straighter answer than a shrug or a sales pitch.

Where AI sits in our workflow

AI-assisted coding, human-owned code. Our developers use AI coding assistants throughout implementation — scaffolding features, writing routine logic, exploring unfamiliar APIs, drafting migrations. But every line that reaches a client codebase goes through the same gate it always has: a human developer reviews, tests, and takes responsibility for the change. Nothing merges because "the AI said so." In practice, AI writes a first draft; our engineers decide what survives.

AI-assisted testing. We use AI to generate unit tests and surface edge cases we might not have listed by hand — empty inputs, timezone quirks, concurrent updates. This has genuinely improved coverage on our projects, because test-writing time used to be the first thing squeezed by deadlines. The tests themselves get reviewed like any other code; a wrong test that passes is worse than no test.

Drafting documentation. AI drafts API docs, setup guides, and handover documentation from the code, and a developer who knows the system corrects and completes them. The result is that our projects ship with better documentation than the industry norm — not because AI writes great docs, but because it removes the blank-page excuse for not writing them at all.

Code review support. Before a human reviews a pull request, an automated pass flags obvious issues — missed error handling, inconsistencies, potential bugs. Human reviewers then spend their attention on the things machines are bad at: whether the change fits the architecture, whether it's solving the right problem.

What we refuse to automate

Just as important is where we deliberately keep humans in charge:

• Requirements and discovery. Understanding your business happens in conversations with you, not in a prompt. This is where projects are won or lost.
• Architecture decisions. Data models, system boundaries, technology choices — these carry multi-year consequences, and someone accountable makes them.
• Security-sensitive code. Authentication, payments, permissions, and anything touching personal data gets written and reviewed with extra human scrutiny, not less.
• The final judgment on shipping. Release decisions rest with a person whose name is on the project.

We're also careful about data. Client code and business information only go through AI tools whose terms we've vetted, and when a client has stricter requirements — regulated industries, sensitive IP — we adapt our tooling to match, and we're happy to put that in writing.

What this means for clients

The honest version of the benefit, without inflated multipliers:

• Faster delivery on routine work. Implementation phases are meaningfully shorter than they were a few years ago. Discovery and integration — the genuinely hard parts — compress less, which is why we don't promise miracle timelines.
• More budget where it counts. When boilerplate costs less, more of your budget goes to design, testing, and the edge cases that determine whether software survives contact with real users.
• Better coverage and docs. Test suites and documentation used to be casualties of deadline pressure. They aren't anymore.
• Small senior teams. AI assistance lets a compact team of experienced engineers deliver what previously took a larger group — and small teams communicate better, which quietly prevents a lot of project failure.

We've applied this workflow across client projects and our own products — Rukrok, Komerce, and Chargly — which gives us a useful feedback loop: we feel the consequences of our own practices in systems we maintain ourselves.

Honesty about the limits

AI-assisted development is not magic, and we'd rather under-promise:

• AI writes confident code, and confident is not correct. Without disciplined review, it ships subtle bugs faster than any human could.
• Gains vary widely by task. Greenfield CRUD work speeds up a lot; untangling a legacy integration with undocumented behavior speeds up much less.
• The tools change monthly. Part of our job is continuously re-evaluating what belongs in the workflow and what's hype — our process today isn't our process from a year ago, and won't be our process next year.

If you're comparing vendors, we'd encourage you to ask every candidate the questions in our guide to choosing a development company — including exactly how AI-generated code gets reviewed before it ships. Any team using these tools seriously will have a specific answer.

Want to see how this workflow would apply to your project — or just want straight answers about AI-assisted development before you commit to anyone? Get in touch. We'll walk you through our process and where it does (and doesn't) help your case.