AI coding tool rollouts skip maturity stages, yield uneven results, TechGig

WorkAI.TV Editorial Desk
3 Min Read

Share with your CTO

Engineering organizations deploying AI coding tools are consistently hitting the same wall: they rush from initial experimentation straight to org-wide scaling, skipping the two stages where the real work happens. Vivek Raghunathan, SVP of Engineering at Snowflake, frames this through a five-stage maturity model and argues that pressure to show quick ROI is what drives teams to skip the discovery and codification phases. His team built roughly 14 named AI coding patterns before scaling adoption, treating them as a shared vocabulary across the engineering org.

What this means for your business

The recurring failure mode looks like this: a team buys GitHub Copilot or Cursor, runs a pilot, sees some developers love it and others ignore it, then declares adoption at 60 percent and moves on. What they’re actually measuring is license activation, not capability transfer. When the codification stage gets skipped, individual productivity gains stay siloed in the developers who figured things out themselves, and institutional learning never accumulates.

Raghunathan’s 14 named patterns are the analytical heart of this. “Plan in writing before you write code” sounds obvious until you realize most teams never make it a stated, teachable norm. Named patterns do something mandates can’t: they create a shared language that lets senior engineers teach practices without being physically present, lets managers measure adoption against meaningful behaviors, and lets new hires ramp to team-standard AI use rather than starting from scratch. This is how craft knowledge becomes organizational knowledge.

The metric problem compounds fast. Teams that skip codification default to measuring lines of code or merged pull requests because those numbers exist and look good in a board deck. Those proxies actively distort behavior: developers optimize for throughput on easy tasks and avoid using AI on complex, ambiguous work where it’s often most valuable. The signal worth watching is whether your engineering org can articulate, in writing, what good AI-assisted work actually looks like before it tries to scale it.

Concept deep-dive: Codified AI patterns

A codified AI pattern is a named, documented practice that describes when and how to use an AI tool effectively for a specific class of engineering task. The concept exists because AI tools are general-purpose: without guidance, every developer invents their own workflow independently. Think of it like a design pattern in software architecture, a reusable solution to a recurring problem, except here the problem is human-AI collaboration rather than code structure. Without these patterns, scaling AI adoption means scaling chaos rather than capability.

Based on reporting from AI coding tool rollouts skip maturity stages, yield uneven results, TechGig, originally published 2026-11-20 02:08:00.

Share This Article