---
title: "AI coding tools, but no faster delivery | QMates"
source: https://qmates.tech/en/services/engineering-transformation/software-delivery-assessment/ai-coding-tools-not-faster
description: "AI coding tools speed up writing code, yet releases stay rare. Find out where your team’s work stops, whether in code review, testing or the pipeline."
updated: 2026-11-21
---

AI in software development

# We use AI to write code and we’re no faster

With an AI assistant, writing code speeds up while the rest of the cycle stays as it was: code review takes longer, tests are missing, the pipeline caps how many changes reach production. Every larger pull request lengthens the review queue, and developers who use the assistant their own way, without team rules, produce larger ones.

## Does this sound familiar?

-   Developers say they write code faster, and the release frequency has stayed the same.
-   Pull requests have grown larger, and those waiting for code review pile up.
-   Everyone uses the assistant their own way, some on a personal account, and the team has no shared idea of what works.
-   Generated code passes review, then defects turn up in testing or in production.
-   Licences and tokens cost more and more, and it’s unclear what has changed in the work.

## What slows a team that writes code with AI?

Google Cloud’s 2025 DORA report describes AI as an amplifier of what an organisation already is, weaknesses included. The causes below are those weaknesses, and each of them adds to the time work spends waiting once the code is written.

1.  ### Work stops after the code is written

    Often the time goes after the code is written, when the pull request waits for someone to review it and testing starts once an environment is free. When the bottleneck is there, code written with the assistant arrives sooner and then waits just like the code before it. Following the whole cycle, from backlog to production, as the \[Software Delivery Assessment\](/servizi/trasformazione/assessment-velocita-sviluppo) does, shows where the time is lost.

    [

    Handled by

    Software Delivery Assessment](https://qmates.tech/en/services/engineering-transformation/software-delivery-assessment)

2.  ### Bigger changes, and few tests to protect them

    With an assistant it is easy to produce, in little time, a change that touches many files, and the reviewer has to understand all of it before approving. Without automated tests written alongside the code, review is the only safety net, and it slows down exactly where AI had saved time.

    Feeds into

    -   

    [

    Handled by

    Technical Coaching in XP and TDD](https://qmates.tech/en/services/engineering-transformation/technical-coaching-xp-tdd)

3.  ### The pipeline sets the limit

    A slow build or a release that needs manual steps determines how many changes can reach production, however fast the code is written. The gain from AI stops there, at the next release window.

    Feeds into

    -   

    [

    Handled by

    DevOps, Continuous Delivery and Platform Engineering](https://qmates.tech/en/services/engineering-transformation/devops-continuous-delivery)

4.  ### Everyone uses AI their own way

    Each developer has built their own method, with different tools and rules, sometimes outside any company policy, with no team decision on which work AI should affect or how to measure the effect. Everyone saves time for themselves, and the team’s flow stays as it was.

    Feeds into

    -   

    [

    Handled by

    AI Adoption Assessment](https://qmates.tech/en/services/ai-adoption-assessment)

- [The service that tackles it](https://qmates.tech/en/services/engineering-transformation/software-delivery-assessment)

## What changes when AI speeds up the team’s work

The team uses assistants and agents with shared rules, at the point where work was slowing down: small changes, tests written alongside the code, reviews that don’t turn into a queue. The time saved writing code carries through to release.

The pipeline data shows how often finished work reaches users, and decisions on tools, licences and training start from there. The rules for using AI are written down and shared, and anyone joining the team finds them ready.
