We're often asked: "do you do the work, or does the AI?". Here's our answer, in plain words, and why the line matters.
For two years, AI has dominated the conversation. Every agency talks about it, every vendor sells it. We use it, but we refuse to put it in the window. Not posture, responsibility. At Yumea, AI helps us build. What we build, we decide, verify, and answer for. A first name is signed under each delivery. The sentence is short. Here's the detail.
What AI actually does for us
It speeds up build tasks that commit no one. Drafting code from a clear spec. Finding a bug faster in ten thousand lines. Proposing several architectures for one module so we can pick one. Formatting content, catching a typo, summarising an incident log. In every case, the output is reviewed by a named person before it reaches you. We never ship anything with AI as sole author.
What AI doesn't do
It doesn't choose the technology you'll live with for years: we do that with you, out loud. It doesn't choose a commercial or product strategy: it has no access to your context, your last five years of clients, what you told us in a half-whisper at the end of a meeting. It doesn't answer your clients for you: the emails your clients receive from us are written by humans, reviewed by a second human, and signed by a first name. It doesn't run security tests: a security test unplayed by a human is not a test, it's a hypothesis. That's in our internal doctrine.
Why the line matters
Because in production, a person fixes things, not a prompt. If a system goes sideways at three in the morning, AI won't call you back. We will. If a bug hits your till on a Friday night, AI won't open your client database to work out what happened. A named developer does. We prefer to say this publicly before an incident forces the question. AI speeds up making. It doesn't speed up responsibility. And responsibility is what you buy when you work with an agency.
What we've seen on the ground
A bug generated by a poorly-framed prompt, caught during human review before shipping. The error was subtle: the code did exactly what was asked, but what was asked was wrong. AI had no handle to catch that. A human did.
FAQ
Which AI do you use? Several, depending on the task. We pick them for what they do well, not for who sells them.
Where do they run? Tasks that touch your data run on self-hosted AI in Europe. Other tasks may pass through external services. We always tell you what runs where.
Precise question about how we work with AI? Ask it. We'd rather answer upfront than after an incident. Let's talk →