What is AI changing in software development? A new report from Sii

AI is now a permanent part of software development teams. It happened faster than many companies expected, and some are still not ready to make the most of it. Developers now work with AI assistants every day, and AI agents are taking on increasingly complex tasks. The next challenge is using these capabilities at scale, and knowing whether AI is actually improving software delivery. That question is at the heart of Sii’s new report, Software Delivery in the Age of AI, which draws on market analysis, hands-on project experience, and the team’s own experiments. 

Faster coding doesn’t mean faster delivery 

Not long ago, the question was whether AI could meaningfully help developers. Today, there is little doubt that it can. The more interesting question is what is what happens to the results of their work. 

With AI, code can be written in minutes, only to sit for hours or even days waiting for review, testing, or approval. A developer may finish a task faster, but that doesn’t mean the change reaches users any sooner. 

The authors examine what they call the productivity trap: why gains in individual productivity don’t always translate into better results for the whole team. They also look at how to measure the impact of AI adoption and where to find the bottlenecks that can cancel out the benefits of new tools.

What did Sii’s experiment show? 

A central part of the report covers an 11-day experiment run on three live production projects. 

The Sii team used Claude Code and built an agent harness around it: custom tools, integrations, and mechanisms for orchestrating and controlling AI agents. The goal was to see how much an agent’s effectiveness depends on its environment and on how its tasks are organized and supervised. 

The results are promising and thought-provoking. With the harness, 60 of 70 tickets (86%) went end to end with no human input outside the planned checkpoints. With the baseline setup, the success rate was just 33%. 

What explains the gap? How did the agents work, and where did people still need to step in? The authors walk through the experiment in detail, including what tasks cost and where the approach falls short. The findings are also a starting point for rolling out similar setups across larger development teams.

How can organizations prepare to work with AI Agents? 

The report also presents Sii’s approach to building an AI-native software development life cycle (SDLC). At its core is a three-stage maturity model that helps companies assess how ready they are for broader AI adoption. 

The authors explain the difference between using AI for individual tasks and building it into development workflows, and when it makes sense to hand whole processes over to agents. They also cover security, data access, and quality control. A separate chapter covers testing and where AI brings measurable gains for QA teams. 

Finally, the report outlines a 6-12-month roadmap, with practical guidance on assessing current AI use, choosing the right areas for pilots, and gradually scaling what has already proven to work. 

About the report 

Software Delivery in the Age of AI​: Sii’s perspective on scaling AI across the software development lifecycle​ is written for technology leaders and the people responsible for building and delivering software. 

The authors share market insights and lessons from their own work with AI agents, along with approaches that help teams move beyond early experiments toward a more structured use of AI in everyday development. 

What new opportunities do AI agents create for software development teams, where do their limitations lie, and how can organizations prepare for broader adoption? Discover the answers in Sii’s full report. Download here.