Special: The DORA 2025 Critical Review

Jan 2, 2026

Dorota Parade, tech entrepreneur and CEO of Authress with engineering leadership chops, skewers the 2025 DORA Report. She argues it trades hard data for AI-flavored narrative. They debate LLMs that make engineers feel faster but do not shorten feedback loops, the limits of agentic tools, feature-flag pitfalls, and why 30% acceptance rates get celebrated.

Ask episode

AI Snips

Chapters

Transcript

Episode notes

ADVICE

Fix Bottlenecks Not Just Code Velocity

If you speed up one part of the pipeline, identify and relieve downstream bottlenecks like release engineering.
Warren warns more code without improving pull-request or testing throughput creates larger burdens elsewhere.

INSIGHT

Feature Flags Can Hide True Lead Time

Feature flags can distort metrics by counting unvalidated, behind-flag code as 'in production.'
They often become crutches for shipping untested work and hide real lead time until the flag is removed.

INSIGHT

Most LLM Use Is Low-Stakes Or Peripheral

LLM adoption is widespread but usage often targets low-stakes tasks like image generation, summarization, or PR-level help.
Both speakers note many 'AI uses' are linters, docs, or meme-making, not product-differentiating work.

Get the Snipd Podcast app to discover more snips from this episode

Get the app

Share Episode

"Those memes are not going to make themselves."

Dorota, CEO of Authress, joins us to roast the 2025 DORA Report, which she argues has replaced hard data with an AI-generated narrative. From the confusing disconnect between feeling productive and actually shipping code to the grim reality of a 30% acceptance rate, Warren and Dorota break down why this year's report smells a lot like manure.

We dissect the massive 142-page 2025 DORA Report. Dorota argues that the report, which is now rebranded as the "State of AI-Assisted Software Development", feels less like a scientific study of DevOps performance and more like a narrative written by an intern using an LLM prompt. The duo investigates the "stubborn results" where AI apparently makes everyone feel like a 10x developer, where the hard results tell a different story. AI actually increases software and product instability — failing to improve.

The conversation gets spicy as they debate the "pit of failure" that is feature flags (often used as a crutch for untested code) and the embarrassing reality that GitHub celebrates a mere 30% code acceptance rate as a "success." Dorota suggests that while AI raises the floor for average work, it completely fails when you need to solve complex problems or, you know, actually collaborate with another human being.

In a vivid analogy, Dorota compares reading this year's report to the Swiss Spring phenomenon — the time of year when farmers spray manure, leaving the beautiful landscape smelling...unique. The episode wraps up with a reality check on the physical limits of LLM context windows (more tokens, more problems) and a strong recommendation to ignore the AI hype cycle in favor of a much faster-growing organism: a kitchen countertop oyster mushroom kit.

💡 Notable Links:

🎯 Picks:

Warren - The Maximum Effective Context Window
Dorota - Mushroom Grow Kit