New analysis reveals builders are quickly adopting AI brokers, creating new challenges round coordination, management, and shipped software program.
GitKraken launched a framework for understanding how software program improvement is evolving as AI brokers turn into a core a part of the event course of, an idea referred to as Code Stream.
The announcement is accompanied by new analysis carried out by GitKraken in Could 2026 amongst greater than 550 software program builders, revealing that AI adoption has moved effectively past experimentation.
Key findings embody:
- 96% of respondents reported some stage of AI adoption on their crew
- 65% mentioned greater than half of their crew has adopted AI instruments
- 30% actively run a number of AI brokers in parallel, with one other 33% experimenting with this observe
- 28.5% are already permitting AI to function autonomously, both individually or alongside different brokers
“What’s altering isn’t simply how builders write code – it’s how they work,” mentioned Matt Johnston, CEO of GitKraken. “More and more builders are more and more managing a number of brokers on ever rising swaths of labor, coordinating work throughout repositories, and evaluating AI-generated output. The problem is not producing code. It’s flip agent-generated code into high-quality, shipped software program – from plan to predominant ”
As AI adoption accelerates, builders are starting to work in ways in which have been unthinkable simply 6-12 months in the past. As an alternative of implementing a function themselves, they could direct a number of brokers working in parallel, use completely different fashions for various duties, and spend extra time reviewing and coordinating work than producing it straight.
Additionally Learn: AiThority Interview with Matej Bukovinski, Chief Know-how Officer at Nutrient
GitKraken’s analysis discovered that the highest makes use of of AI in software program improvement right this moment are:
- Producing and refactoring code
- Troubleshooting and debugging
- Conducting code critiques
These modifications are creating a brand new set of challenges for engineering groups:
- Which fashions are greatest suited to several types of work?
- What number of brokers ought to be working concurrently?
- How ready is a repository for agent-driven improvement?
- How do groups keep requirements and context throughout parallel workstreams?
- How do organizations safely combine and merge dramatically bigger volumes of code?
GitKraken prospects coined the time period for this rising problem as Code Stream.
Code Stream describes how work strikes between builders, coding brokers, repositories, planning programs, critiques, branches, and manufacturing environments. It encompasses the coordination, visibility, governance, and integration required to rework AI-generated output into working software program.
Traditionally, software program improvement instruments have been constructed round a easy assumption: people wrote code and Git tracked the end result.
At present, builders might have a number of brokers working concurrently throughout a number of repositories, creating considerably extra branches, pull requests, commits, and merge exercise than groups managed just a few years in the past.
“The trade has spent years targeted on serving to builders write code sooner,” Johnston mentioned. “As AI adoption grows, organizations want higher methods to know what’s taking place throughout their improvement programs, the place work is getting caught and the way brokers are performing. The problem has turn into managing the exploding velocity and quantity of labor coding brokers can create. That’s what Code Stream is about.”
GitKraken acknowledges Code Stream will turn into an more and more essential self-discipline as organizations increase their use of AI brokers and autonomous improvement programs – and demand optimistic returns from their investments in AI developer tooling.