Berkeley paper argues autonomous software development will reshape the entire SDLC

mboehme_ · x · 2026-07-28

Autonomous software development is framed as a structural shift, not just faster coding

The post points to a new position paper, Towards Autonomous Software Development, arguing that software engineering is entering a deeper transition rather than a simple productivity upgrade.

What the paper claims

It introduces three autonomy levels for software development.
It identifies six structural shifts in how software is specified, built, verified, deployed, and governed.
It lays out ten predictions for the future of software engineering.
It proposes a research roadmap toward trustworthy autonomous software development.

Core thesis

Frontier coding agents can already do more than code generation: they can reason across repos, write and run tests, find vulnerabilities, and coordinate multi-stage work. But the paper argues current agents are still stuck at the lowest level of the SDLC automation ladder, especially when measured on continuous evolution rather than isolated tasks.

The broader implication

The authors predict that once AI becomes central to software production, the bottleneck shifts from writing code to defining specifications, evaluating agents, governing trust, and organizing AI-native workflows. In other words, software may become more abundant, while trust and certification become scarcer and more valuable.

Original post →

Berkeley paper argues autonomous software development will reshape the entire SDLC

More from coding & agent