AI Impact: 90% per Step, 5% per Cycle

Avenir Voronov, adoption director at Veai, explains how CTO experience and experiments with AI in software delivery became a product and an implementation practice. Showing management a new agent is not enough; a team must prove the business effect and connect it to engineers' actual work.

The team builds a metric pyramid: P&L and executive goals at the top, flow speed in the middle, and signals such as DORA and tests at the bottom. Saved time is estimated counterfactually by comparing an agent's work with a human-effort estimate, while keeping the ability to drill into a task.

Telemetry covers code, project exploration, documentation, and questions asked of the system. Adoption moves through stages of trust, from avoiding production to broader use cases. Acceptance rate and diff size are balanced with rework, quality, releases, and user feedback instead of becoming one target metric.

Every measurement needs a purpose: more prompts do not prove value and may conceal a declining ability to frame tasks. Mature teams move AI toward intent and business hypotheses, then extend it into operations. Success is defined by the customer's goal and a demonstrated business change, not one universal acceleration percentage.