«InsightsFrom practice
Does agentic AI coding make us 10x faster? An honest answer
Agentic AI coding is the industry's current promise, and the numbers being passed around are spectacular: 10x faster, half the cost. We build with it every day, including our own AI SaaS, Gardeo. So here is an honest take instead of a marketing number.
The promise meets the measurement
Agentic AI coding is the industry's current promise: 10x faster, half the cost. We build with it every day, including our own AI SaaS, Gardeo. That is exactly why we think an honest read matters more than a marketing number.
The cleanest measurement so far comes from the research organization METR. In a 2025 randomized controlled trial, 16 experienced open-source developers worked through 246 real tasks in their own mature projects, using modern tools like Cursor Pro and Claude 3.5 and 3.7 Sonnet.
The result is surprising: with AI they were on average about 19 percent slower. More striking still is the perception. Those same developers estimated afterwards that AI had made them about 20 percent faster. The study describes a specific setting and is explicitly a snapshot of early-2025 tooling, but the gap between feeling and measurement is the real message.
Where the speed is real anyway
That does not mean AI coding delivers nothing. At the start of a project the gain is real and tangible: a greenfield feature, a prototype, a migration that follows a clear pattern, boilerplate, tests, scaffolding. Here we get to something working noticeably faster.
The first clickable version appears in hours rather than days. You can show something sooner, gather feedback sooner, try more variants. This is exactly the part behind most of the tenfold screenshots doing the rounds.
The only thing that matters is being honest about which part that is: the beginning, the approximate, the not-yet-production. There the speed is real, and we use it deliberately and selectively.
Where it breaks: the bottleneck isn't typing
Inside an existing, complex codebase, exactly the setting of the METR study, it looks different. Once context, architectural decisions and integrations enter the picture, the effect shrinks, sometimes to nothing or below.
The reason is simple: the bottleneck in real projects is rarely typing code. It is specification, review, debugging, integration, testing and coordinating in a team. An agent does not make that work ten times faster, and the subjective sense of speed routinely overestimates the real gain.
Sometimes it even gets more expensive. When more code arrives faster, someone still has to understand, review and stand behind every line. Unreviewed output is not delivered software, it is just uncertainty piled up faster.
Why senior judgment matters more
When code can be produced almost arbitrarily fast, the value shifts. It no longer sits in the writing but in the judging: does the architecture hold? Does data protection stand up? Will it break when you scale? Is the quickly generated approach still maintainable a year from now?
This is exactly where experience decides, and where the study's finding makes sense: someone who truly knows a mature codebase is often faster by hand than directing an agent and vetting its suggestions. AI-assisted development does not make a senior redundant, quite the opposite: it makes their judgment the part that counts most.
The machine supplies options; the responsibility for the right one stays with a person. That is not a romantic stance, it is simply what the numbers suggest.
How we work, and what it means for price and project
We use agentic coding where it holds up: faster to a first prototype, more iterations in the same window, less dull routine work. But every line that ships to production is senior-reviewed and owned by us, not taken from the model unchecked. Gardeo, our own privacy-compliant AI SaaS, is built exactly this way.
Speed does not translate into a blanket discount. What you pay for is not the typing, which the machine increasingly handles, but the judgment that keeps the software sound. On a fixed price, building faster is our advantage: we deliver at the agreed price and carry the estimation risk, while you get earlier results and more iterations.
Bottom line: no blanket 10x, no discount promise. Instead, software that holds up, faster where speed is honestly possible, and a senior who stands behind it.
Let's plan your project.
Every successful project starts with clarity. Book a short call: no sales talk, just an honest read on your idea.