Intel / video

The State of AI in Software Development: Data from 400+ Orgs — Justin Reock, DX

Source: youtube.com
Author: AI Engineer
Date: Published Sep 30, 2026

Context

DX's data suggests that faster code generation does not translate directly into a matching increase in completed engineering work. In a conference talk, DX's deputy CTO Justin Reock, according to the talk's description, cites median gains in pull-request throughput of around 7.7%, with even top performers short of 2x, and argues that code generation was never the bottleneck. Pull-request throughput is the flow of proposed code changes through a team, so the measure concerns delivery activity, not how quickly a model writes code.

In a separate discussion of its velocity research, developer-analytics company DX describes following organization-level throughput as AI adoption matured, while acknowledging that it is hard to separate AI from other confounding factors. The discussion also says new tools introduce new bottlenecks, and its show notes list planning, reviews, testing, documentation and coordination as still consuming most engineering time. The two sources are distinct and may not use the same sample. Together they support measuring the whole development process alongside AI use, rather than assuming time saved on coding becomes an equivalent delivery gain.

Key takeaways · AI-distilled

DX data on about 200,000 engineers shows deployment frequency rising while change failure rate has become far more volatile, and maintainability is up while change confidence is down.
PRs grew from about 44 to 72 lines, and Reock says incremental delivery is suffering as PR size grows.
Juniors use AI the most, but staff+ engineers save as much time while using fewer , according to DX's data.
Median PR throughput gains are around 7.7% and even top performers did not reach 2x, which Reock attributes to code generation never having been the bottleneck.
Case studies Reock cites: Morgan Stanley saving 300K hours a year, Zapier getting 15% more value per engineer while still hiring, and Spotify running an SRE incident .

Terms in this piece · Glossary

token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.

Read the source www.youtube.com