AI agents for speed, a senior to make it last.
AI agents write much of the code. I learn your business, make the architecture calls and check every release myself before it goes live. For you: shorter timelines, the same standards, and one point of contact from scoping to launch.
What it changes
Faster, without lowering the bar.
Koala Studio is an AI-native studio: built from day one to work with AI, under human control. For you, it changes four things.
Timelines
Shorter timelines
A repair platform live eight weeks after kick-off, for SOS CHR. A native Android app built in six weeks, for ResulCare. Complete projects, not prototypes.
Contact
One point of contact
The person you explain your business to is the person who decides and checks the work: me. No project manager in between, nothing lost in translation.
Standards
The same standards
A written spec, automated tests, my review before every release. AI speeds up writing code. It doesn't remove a single check.
Price
No bargain-basement pricing, no miracle claims
My prices are public. I don't promise that AI does everything: I promise software that works, delivered on the timeline we agreed.
AI writes a large part of the code, under my direction: that's why I deliver fast. The method is why it holds up.
How a project runs
From scoping to launch, nothing hidden.
- 01
Scoping
I learn your business, your constraints and what the project needs to deliver. I also tell you what not to build.
- 02
Spec
Together we write the spec, a detailed brief setting out what the software must do, for whom, and under which rules. It's a plain-language document that you review and sign off. It's the reference for the whole project, for the agents and for me.
- 03
Staged delivery
The project is broken into short stages, with a demo every week. You test on real cases and stay in control of priorities.
- 04
Tests
Every feature comes with automated tests that run on every change. They check that what worked yesterday still works today.
- 05
My review
Before every release, I check what's going out myself: how it behaves, its security, and whether it matches the spec. Nothing ships without that review.
- 06
Go-live and aftercare
Deployment, monitoring, and bugs fixed for 3 months after delivery.
Who does what
The agents write. I decide.
The AI agents
- Write the code from the spec
- Write the automated tests
- Propose fixes when a test fails
- Draft a first version of the documentation
- Handle the repetitive work
Me
- I learn your business and write the spec with you
- I decide the architecture, the tools and the priorities
- I check every release before it goes live
- I decide what ships, and when
- I talk to you, report to you, and answer for the result
This isn't vibe coding: an agent never decides the architecture, or what goes into production.
I work mainly with Claude Code and Codex. They're coding agents: they read the project, write code, run the tests and fix their own mistakes, under my direction.
These tools change fast. The method doesn't depend on any one of them: spec, tests, human review. If a better tool appears tomorrow, I change the tool, not the method.
My workshop
My collaborators at work.
Here is my team of agents in action. To work with them without losing sight of anything, I built my own control tool: Canopée, a small virtual office where I see what each agent is doing, which one is waiting for my decision, which one is resting.
The errors that occur in production on each project flow back there on their own, and the project's agent analyses them before I approve the fix. It's also what lets me answer you quickly: I know where your project stands without piecing it back together.
Canopée runs locally on my machine and is never exposed to the Internet.
My collaborators at work. Demo data: fictional projects.
What protects your project
What protects your project.
Checks before every release, and a project you can take over at any time.
Checks
Automated checks between the moving parts
When a project has both a mobile app and a server, an automated check confirms on every change that they still speak the same language. I put it in place on my own products, after seeing how easily a mismatch could slip through unnoticed. If a server update could break the app on users' phones, it's caught before release.
Instructions
Written rules for the agents
Every project has its own instructions: conventions, things never to do, areas to watch. They're versioned alongside the code. On an app taken over in production, this documentation lets the agents work safely on code they didn't write.
Rollback
Releases you can roll back
On every project, the database is backed up automatically before each deployment, so it can be rolled back if needed.
Confidentiality
Your confidentiality requirements, agreed during scoping
If your field has rules (health, legal, personal data), I set them with you: the data AI tools can access is agreed with you during scoping.
The code and the accounts are yours.
- The source code lives in a repository in your name, or is handed over to you.
- Hosting, domain names and app store accounts are in your name.
- The documentation and the agents' rules are part of the project. They're yours too.
- I use standard technologies. Another engineer, with or without AI agents, can take the project over.
You deal with one person: me, with 20 years in the trade. I'm backed by a network of trusted specialists I bring in to add capacity when a project needs it, or to take over if I'm unavailable.
The spec, the documentation and the agents' rules are written so the project never depends on one person's memory. Code, access, documentation and the agents' instructions are handed over to you with every delivery: another engineer can take over without starting from scratch. My background.
Frequently asked questions
What people ask me about AI.
Can you trust a supplier that codes with AI?
Yes, if an experienced engineer makes the decisions and checks the work, whatever the set-up: agency, freelancer or a studio like mine. Ask any supplier three questions: who reviews the code before it goes live? Are there automated tests? Who owns the code and the accounts? My answers: me, yes, and you.
Is AI-written code secure?
Not without review. Agents make mistakes, sometimes security mistakes. That's why every release is tested automatically, then checked by me before it goes live. Security is part of what I check.
Is it cheaper because it is built with AI?
It's faster, which lowers the cost of a project of a given scope. But the core work is still mine: understanding, deciding, checking. My prices are public, on the Offers and pricing page.
Who's responsible if a bug reaches production?
I am: I decide what ships, and I fix it. Bugs are fixed for 3 months after delivery.
Which AI tools do you use?
Mainly Claude Code and Codex. The tool may change; the method stays the same.
Which data do the AI tools see?
The data you decide: what AI tools can access is agreed with you during scoping.
Can we develop the software further without you?
Yes. The code uses standard technologies, it's documented, and it's yours. Another engineer can take it over, with or without AI agents.
Let's see if the method suits your project.
30 minutes to talk about your project and your constraints. Put your questions about the method to the person who applies it. No commitment.