6 essential rules for unleashing AI on your software development process - and the No. 1 risk (www.zdnet.com)

Digital.ai’s 18th State of Agile Report — based on responses from nearly 350 Agile coaches and consultants at large enterprises — finds AI and AI agents are beginning to accelerate and potentially improve software delivery, offering a way to revive Agile practices that have been stuck on a decade-long plateau. With pressure mounting on tech teams to demonstrate ROI (76% of managers cite increased scrutiny) and only 13% saying Agile is deeply embedded, the report frames “agentic” AI not just as an assistive tool but as a potential orchestrator of the full SDLC: reasoning, deciding, and acting to improve flow, quality, and speed at scale. Early adoption today is mostly at the task level (code generation, docs, test creation, backlog triage and dashboards), not widespread autonomous agents. The report and experts stress practical rules and technical guardrails: start small (pilot low-risk functions like test-generation and documentation), map and govern all data sources before scaling, require explainability and human approval for merges/production pushes, and log prompts/outputs/approvals for auditability. Expect initial productivity dips while teams learn prompting and teaching agents. The No. 1 risk is data exposure — accidental leakage of credentials or customer data and “shadow AI” use that evades IT governance — so organizations must prioritize governance, traceability, and human-in-the-loop controls if AI is to become a reliable “teammate” rather than a brittle silver bullet.