“AI-driven” has become a label that every agency puts on its homepage, so it’s worth being specific about what it means for us and what it doesn’t.
What changed
A few years ago, a feature moved through a predictable pipeline: someone wrote a spec, someone designed it, a developer implemented it over days, someone else reviewed it, and tests were written if there was time.
Today, at Loam, the slowest part of that pipeline is no longer typing code. AI agents draft implementations, write tests, migrate code between APIs, produce localisations and generate the first version of documentation. A change that used to take a day of focused implementation can be drafted in an hour.
That shifts where the time goes:
- Shaping the problem. A precise spec is now the most valuable artefact in the project. An agent will happily build the wrong thing very quickly.
- Reviewing. Every change an agent produces is read by a senior engineer before it ships. Review is where architecture, security and taste are enforced.
- Verifying. Agents write tests, but people decide what needs testing. We run the app, click through it and check the edge cases ourselves.
What didn’t change
The decisions that make software good or bad are still human decisions. Which feature to build first. What to leave out. How a screen should feel. Whether a dependency is worth its weight. When “done” is actually done.
These decisions come from experience: from having shipped things that worked and things that didn’t. That’s why we think AI makes senior engineers more valuable, not less. The leverage of a good decision is multiplied by the speed of execution behind it, and so is the cost of a bad one.
What this means for clients
- Smaller teams, less coordination. A senior engineer with agents covers ground that used to need several people, and the communication overhead that came with them.
- Faster first versions. Prototypes on your real data in days, not weeks, so decisions get made on something you can touch.
- Code you can keep. We hold agent output to the same standard as hand-written code: readable, tested and consistent with the rest of the codebase.
What it means for our own products
We use the same workflow on our own apps. journeybot, Tama, Dami, Timo, Hotspot Meter and Kahudo are built and maintained by a very small team, including the marketing sites, App Store listings and release automation. Running real products is how we keep the workflow honest: if a practice slows us down or lets bugs through, we feel it first.
If you’d like to see what this looks like on your project, tell us about it.