Skip to main content
Back to blog
AI

What AI has changed in how we build software

By ···3 min read
On an oak shelf, a strip of tracing paper holds balsa sticks of irregular heights that become even after a band of teal tape.

Since late 2025, the way we build software has changed. AI coding agents are now part of our daily work, guided by a senior team that keeps control of every decision. Rather than simply assert this change, we chose to measure it: here are three years of our activity, compared over the same period.

Three years of data at a glance

Same period, three years

1 January to 16 October

  • 2024
  • 2025
  • 2026
Commits
20241,421
20252,017
202612,288
Merged pull requests
202474
2025142
20261,620
Repositories with commits
202415
202517
202641
Source code lines added
2024173,503
2025388,540
20262,667,698
Test share of added code
20241.2%
202511.1%
202631.8%

In two years, our activity has changed in scale. We run more workstreams in parallel, and almost a third of the code we add now serves to check the rest. These figures describe our activity; they cannot isolate the effect of AI, because our projects have changed too.

Above all, a change of method

Team commits per month, January 2024 – October 2026

Around 150 commits a month in 2024, then a rise from late 2025 up to 1,977 commits between 1 and 16 October 2026.

  1. June 2025: first written rules for coding agents in our projects.
2024
MonthCommits
Jan100
Feb123
Mar149
Apr50
May121
Jun102
Jul96
Aug390
Sep178
Oct182
Nov182
Dec150
2025
MonthCommits
Jan402
Feb164
Mar164
Apr104
May85
Jun130
Jul215
Aug373
Sep313
Oct257
Nov194
Dec394
2026
MonthCommits
Jan750
Feb453
Mar987
Apr1,362
May1,075
Jun1,018
Jul1,786
Aug1,288
Sep1,592
Oct (1–16)1,977

Our first written rules for coding agents appeared in June 2025. Activity picked up at the end of the year, once three principles were firmly in place.

A written framework for every project. Conventions, quality requirements and security rules are documented and shared. AI works within that framework, and every corrected mistake adds to it.

Nothing ships without verification. Every change is reviewed by the team and tested before it reaches your product. Agents attach proof of what they report: a passing test, a page checked on mobile and desktop. We describe this requirement in our article on how agents verify their work.

Quality is built in from the start. Tests accompany every feature, so your product can evolve without weakening what already works.

The tool alone does not make the difference: according to the 2025 Stack Overflow survey, 84% of developers use or plan to use AI. Google Cloud's 2025 DORA report describes AI as an amplifier of a team's strengths and weaknesses, and a randomised controlled trial by METR even found experienced developers 19% slower with early-2025 tools. These studies suggest that the method matters as much as the tool.

What it changes for your project

At Appik Studio, these productivity gains go straight to our clients: we can do considerably more, and faster than before. The time saved on writing code goes back into your project, so that your product can go further in the same timeframe or reach your users sooner.

  • A first version sooner, so choices can be validated with real users.
  • Tests from the start, so the product can evolve without breaking what works.
  • A senior team over time, with the same people as contacts.

AI can speed up writing code. Scoping, architecture, review and release to production remain with our team: these are the steps that determine the quality of a product.

Frequently asked questions

Does AI write all your code?

No. Agents help write the code, tests and documentation. The team decides what to build, reviews the changes and remains responsible for every release to production.

Is a project cheaper with AI?

Not in the proportions of these figures. They measure our activity, not the cost of a feature, and writing code is only part of a project: scoping, design, integrations, validation and release to production still take the time they need. The time saved goes first into delivering sooner, testing more and going further within the same budget. Our article on the cost of a mobile app in Switzerland details the factors that determine a budget.

Your project with a senior team

Do you have a mobile app, web platform or business tool to build? Let's talk about your project.

Sources and references

4 sources
Appik Studio · METR · Google Cloud · Stack Overflow
  1. Appik StudioGit history of the organisation's 62 repositories, team commits only, extracted on 16 October 2026github.com
  2. METRImpact of early-2025 AI tools on experienced open-source developersmetr.org
  3. Google Cloud2025 DORA State of AI-assisted Software Development reportcloud.google.com
  4. Stack Overflow2025 Developer Survey, AI usagesurvey.stackoverflow.co
Back to blogGaspard Chevassus · CEO, Appik Studio