AI is accelerating a major shift in software delivery, reshaping how software is built, planned and governed, according to Digital.ai’s 18th State of Agile Report.
The survey — based on global responses from teams across industries—finds AI has become central to how software is envisioned and delivered, ushering in a “Fourth Wave of Software Delivery” where intelligent systems don’t just support teams but increasingly orchestrate the full lifecycle.
The data reflects growing operational pressure, with more than three-quarters (76%) of respondents reporting heightened expectations for measurable ROI, while 79% say they’re being asked to do more with fewer resources and 77% feel pressure to accelerate innovation.
Agile isn’t fading, but rather evolving — shifting toward predictive, data-driven practices as developers and AI agents begin working in tandem. Customer satisfaction remains the top success metric at 52%, followed by efficiency and cost reduction at 40%.
The report also found that despite budget constraints, organizations continue to invest: 41% increased Agile spending over the past two years.
Terry Densmore, director of Agile product, Digital.ai, says AI can help organizations finally connect Agile practices to business outcomes by pulling signals from planning, development, testing and production into a single, continuous data model.
“When AI operates inside that shared system, it can trace how work moves through the pipeline, show which decisions improve customer experience or reduce cost, and reveal where teams are investing effort without delivering value,” he says.
From Densmore’s perspective, the key is to put AI inside the delivery workflow rather than bolting it on as a separate tool.
“If AI is allowed to operate on fragmented data or isolated dashboards, it will simply create new blind spots instead of closing the existing ones,” he says.
Adoption patterns are also changing, with 74% using hybrid or blended models rather than sticking to a single framework. Roles are expanding as well, with nearly one-third of respondents now responsible for linking Agile execution directly to business outcomes, signaling a move away from activity metrics toward strategic impact.
One of the report’s most striking findings is the speed of adoption of agentic systems. More than a quarter of AI users are already experimenting with Agentic AI — autonomous systems that can make decisions, coordinate work across tools, and handle tasks such as workflow execution, risk detection, governance and planning.
“Once autonomous agents begin executing workflows and making risk calls, enterprises need the same safety controls they expect for any engineer with production access,” Densmore says.
That starts with strict permission boundaries, full audit trails, and human-in-the-loop checkpoints for anything that touches security, compliance, or release decisions.
“Teams also need real-time observability into what the agent is doing, why it acted, and what data it used to make the decision,” he adds.
Finally, every agent must have a clean rollback path, override mechanisms, and a clear escalation process so teams can quickly stop or reverse actions when something looks off.
“Without these safeguards, an autonomous agent can turn a small mistake into a system-wide failure,” Densmore says.
AI adoption overall has surged from 68% to 84%, and 41% of organizations now deploy AI tools in a coordinated way across teams. Yet only 49% have formal governance guardrails in place, highlighting a growing risk as automation advances faster than oversight structures.
Densmore says the real danger is that autonomous systems will make mistakes faster, deeper in the stack, and without the observability that teams need to catch them.
“Every autonomous agent needs the same things you’d expect from a human engineer: A clear scope, audit logs, SLAs and an escalation path,” he cautions. “Without that, you’re flying blind at machine speed.”