Improve AI Enablement
Most engineering orgs have AI coding tools rolled out, but usage is inconsistent: a few engineers use them well, most use them lightly or not at all, and there's no shared sense of where AI genuinely helps versus where it creates more review burden than it saves. The 2025 DORA research frames this precisely: AI isn't a neutral accelerator, it's an amplifier. It magnifies the strengths of a mature delivery system and intensifies the weaknesses of an immature one.
Why it matters
Ad-hoc AI adoption produces inconsistent code quality and no way to measure whether it's actually helping. DORA's 2024 report found 60% of organizations still report medium or low delivery performance despite widespread AI-driven productivity gains.
Without guidance, AI tools get used for the easy problems and ignored for the workflows where they'd help most, and unmanaged adoption tends to show up as two specific failure modes: an "illusion of speed" where generated code creates more debugging than it saves, and "review fatigue" where developers who write less code end up reviewing disproportionately more of it.
Apple's own research on LLM reasoning found accuracy can drop by up to 65% when the same underlying problem is phrased differently, a concrete reason AI output still needs human judgment in the loop, not just a glance at a diff.
60%
of organizations still report medium or low delivery performance despite widespread AI-driven productivity gains (DORA 2024)
How Pragmint helps
The Findings Report looks at how AI tools are actually being used across your teams today, not just whether a license was purchased, and at where your delivery system's real bottleneck already sits, since that's the constraint AI will pressurize first. Co-Dev Coaches pair directly with engineers to build shared workflows across the five dimensions that determine whether an AI coding agent helps or just generates more to review: context, guardrails, tools, process, and feedback. That includes establishing a clear, living AI-use policy instead of a legal document nobody reads, and making internal documentation AI-accessible without leaking data it shouldn't. The self-hosted analytics platform tracks adoption alongside its effect on delivery metrics like lead time and test confidence, not just how productive people feel.
Signals worth tracking
- AI-assisted pull requests don't show a matching spike in review time or rework.
- A living, versioned AI-use policy exists that developers can point to, rather than shadow AI use nobody talks about.
- Lead time and change failure rate move together with AI adoption, not just self-reported productivity.
Practices we draw from
From our open-source Open Practices library.
From our open-source Curious about another area? See the full list of outcomes we help teams improve, or read the FAQ.
Thanks for reaching out!
We've received your info and opened a scheduling page in a new tab — pick a time that works and we'll be in touch soon.