2026년 6월 10일

이 에피소드에 관해

Are your engineering decisions based on what your system actually does, or on what your team believes it does?

Most software teams are navigating in the dark. Developers spend more than half their time reading code, yet reading doesn't scale, and it rarely produces accurate decisions. As AI tools generate code faster than ever, the problem is getting worse, not better. This episode challenges the assumption that delegating understanding to AI is the answer.

Tudor Girba is the co-creator of Moldable Development and the Glamorous Toolkit, and a co-founder of Feenk, the research company behind both. With decades of experience working on some of the most complex legacy system challenges in enterprise software, Tudor brings a rigorous, engineering-discipline lens to a problem the industry has largely left unaddressed: how do you make a system legible, without reading it line by line?

Tudor is also co-authoring a book with Simon Wardley called Rewilding Software Engineering, written openly and available free at moldabledevelopment.com. The work sits at the intersection of software engineering, decision-making, and organisational change, and is increasingly relevant to anyone responsible for large or ageing codebases.

What We Cover

Why reading code is the single largest cost in software development, and why no one talks about optimising it
How moldable development works, and why it is closer in spirit to test-driven development than to conventional documentation
Why AI cannot replace human understanding of a system, and what happens when teams try to make it do so
How asking an LLM to build you a tool produces better results than asking it to answer your question directly
Why a diagram drawn by a human about a system is a belief, not a fact, and what the difference costs you
How Feenk helped an organisation find an entire system they did not know they had, two weeks into an engagement
Why only 26 percent of software modernisation efforts succeed, and what the outliers are doing differently
What it looks like to migrate more than 20 legacy systems in under a year, every time, without failure
How the Glamorous Toolkit enables thousands of custom, context-specific development tools to compound over time
Why Tudor believes we are nowhere near the peak of human ability, and why perception is the bottleneck, not intelligence

Whether you are a CTO navigating a legacy modernisation programme, a senior engineer trying to reduce the time your team spends reading unfamiliar code, or a technology leader being asked to rely more heavily on AI for engineering decisions, this episode will give you a clearer framework for where human understanding must remain, and the tools that make it possible.

Chapters

00:00 — Reading Code Is the Biggest Cost Nobody Is Optimising

02:26 — What Moldable Development Actually Solves

07:17 — Why LLMs Miss 20-30% of Your System, and What to Do Instead

09:23 — Diagrams Are Beliefs, Not Facts

10:12 — Make the System Draw Itself

13:26 — The User Interface Insight That Changes Everything

18:37 — Verifying AI Output Against Reality

21:04 — The Data Pipeline That Did Not Move for Three Years

24:03 — 26% Success Rate: Why Modernisation Efforts Fail

25:13 — The Bottleneck Is Perception, Not Intelligence

Resources