How to Achieve +250% Velocity with Generative AI

On This Podcast

In this conversation, we examine how Elliott Beaty's team at Fruition restructured around AI by treating documentation as a first-class deliverable rather than an afterthought. He explains that strong Swagger definitions for their microservices, combined with per-folder CLAUDE.md markdown files describing architectural conventions, are what let agents like Claude Code and Windsurf work with minimal babysitting. We also discuss how the team went further and had agents generate their own documentation across the codebase, turning a job engineers traditionally dreaded into something automated and continuously improving. Elliott notes that markdown has effectively become the shared language between humans and agents, displacing earlier ideas about structured databases as the source of shared context.

In this conversation, we examine a concrete example of AI-driven development delivering real business value: Claude Code, connected to their Jira MCP server, diagnosed and fixed a race condition between a webhook and an HTTP controller in two unrelated parts of the codebase. Elliott recounts giving it nothing more than a ticket number, watching it pull context, trace the bug, write a fix and unit tests, and correct a CI pipeline failure after being handed the error message. We also discuss how this collapsed a task that would normally take two days, plus the overhead of backlog triage and resourcing, into roughly thirty minutes of unattended work. Elliott stresses this reliability is not yet the norm across all tasks, but for well-scoped problems it freed up a scarce senior engineer during a revenue-critical fire drill.

In this conversation, we examine the headline result of Fruition's AI adoption: engineering output, measured in story points through QA, is roughly 250% higher than in October of the previous year, even accounting for increasingly complex features. We also discuss where the strain has landed instead of engineering, with Elliott identifying QA as the team's clear bottleneck because testing tooling has not matured at the same pace as coding assistants, and critical thinking remains a distinctly human skill in test design. He describes the product team as similarly stretched to keep up with the new velocity, while marketing has adapted more easily because its work is already process-driven. Elliott frames this as a classic theory-of-constraints problem, where speeding up one part of the value stream simply exposes the next weakest link.