Blog

Announcement: Deterministic PR Refactoring Agents

Adam Tornhill — Thu, 28 May 2026 18:23:03 GMT

CodeScene’s new PR Refactoring Agent analyzes the branch using deterministic Code Health data, performs targeted refactorings, and commits improvements back to the PR.

The result is code that’s easier to review, AI-friendly, and more token efficient. And yes, we support all those promises with data. Your code deserves that.

Fix Code Health degradations via a single click in the PR.

Unhealthy code is burning your token usage - here's the data

Adam Tornhill — Thu, 28 May 2026 16:14:00 GMT

Our research across C++, Java, and Python shows that agents working on unhealthy codebases consume up to 50% more tokens to complete the same tasks. That’s in addition to the increased defect risk covered in earlier research. You're not just getting worse output. You're paying significantly more for it.

Knowing that your unhealthy code burns excess tokens is useful. Having a workflow that prevents it from getting worse with every agent commit is what changes the outcome.

Making Legacy Code AI-Ready: Benchmarks on Agentic Refactoring

Adam Tornhill — Wed, 18 Mar 2026 15:53:36 GMT

Most enterprises are now on a trajectory towards agentic AI. The general assumption is that faster coding will translate into increased productivity across the board.

However, there’s a potential clash between expectations and reality:

The Code Health benchmarks show that the average Hotspot Code Health in the IT sector is 5.15 on a 10.0 scale.
The threshold for healthy code that is easy for a human to understand is 9.0.
AI requires even better code quality, not less: research on AI-friendly code shows that you need at least 9.4 to keep AI-induced bugs in check.

Given that the average industry Code Health is only 5.15, naive AI adoption is unlikely to deliver the expected benefits. Not because the agent’s underlying LLMs are weak, but because the average code structure is too unhealthy for an AI to comprehend and safely modify.

This implies that legacy code is a key bottleneck for enterprise adoption of agentic coding tools. Before AI can accelerate feature work, code health needs to improve. Let’s explore how AI can guide that uplift.

Agentic AI Coding: Best Practice Patterns for Speed with Quality

Adam Tornhill — Fri, 20 Feb 2026 12:28:33 GMT

This year has marked an interesting and fundamental shift. I’ve been coding for almost 40 years, 30 of those professionally, and yet I’m now at a point where 100% of my production code is written and maintained by AI agents. I’m not alone: companies at the forefront of the agentic AI revolution report the same pattern.

But let’s not get carried away by hype. This shift didn’t come for free. Coding via agents requires more rigor, more structure, more code quality, not less.

In this article I’ll share what worked for my team, as well as the general patterns I’ve observed in the industry. (I’m in the fortunate situation of regularly meeting and working with organizations across the globe in all states of AI adoption).

Strengthening the Inner Developer Loop: Turn AI Into a Reliable Engineering Partner

Adam Tornhill — Tue, 02 Dec 2025 13:30:47 GMT

What if AI could accelerate development without leaving a trail of complexity, defects, and hard-to-reason-about code that even your own AI agents fail to understand?

That’s the promise of a well-designed inner developer loop: a feedback system that keeps AI productive and predictable. In this article, we present a practical and research-backed framework for building codebases that stay understandable to both humans and machines, today and for the years to come.

Human, What Must I Tell You?

Markus Borg — Tue, 23 Sep 2025 13:48:42 GMT

Requirements on Technical Debt: Dare to Specify Them!

Markus Borg — Tue, 23 Sep 2025 13:48:25 GMT

Requirements for Organizational Resilience: Engineering Developer Happiness

Markus Borg — Tue, 23 Sep 2025 13:47:17 GMT

Cyclomatic complexity - a fresh look at code complexity｜Blog

Adam Tornhill — Mon, 22 Sep 2025 14:45:59 GMT

Complex code slows down development productivity, increases the delivery risk, and leads to excess maintenance costs.

Architectural Analysis Simplified and growing Programming Language support

Adam Tornhill — Mon, 22 Sep 2025 12:12:25 GMT

We’re proud to announce a new release of CodeScene on-prem, version 4.2.