TDD, AI Agents and Coding with Kent Beck – The Pragmatic Engineer

Kent Beck, creator of Extreme Programming and Test-Driven Development, joins Gergely Orosz to discuss how AI is changing software development and what it means for practices like TDD.

Kent Beck’s AI Journey

Kent has been actively experimenting with AI coding tools:

• Started skeptical, became increasingly impressed
• Uses AI daily for coding tasks
• Sees both opportunities and challenges

TDD in the AI Era

Does TDD Still Matter?

Kent’s perspective on TDD with AI tools:

• Tests as specification: Tests describe what you want, AI generates implementation
• Red-Green-Refactor evolves: The cycle changes but principles remain
• AI can write tests too: But human judgment on test quality matters

New Workflows

How TDD adapts to AI:

1. Write test (human)
2. AI generates implementation
3. Human reviews and refines
4. Iterate

AI Coding Agents

Kent’s thoughts on autonomous coding agents:

Current State

• Agents can handle simple, well-defined tasks
• Struggle with ambiguity and context
• Best for repetitive, mechanical work

Future Potential

• Will improve at understanding intent
• May handle larger chunks of work
• Human oversight remains essential

The “Taste” Problem

• AI lacks aesthetic judgment
• Can’t distinguish elegant from merely functional
• Humans provide the “taste” layer

Programming as Thinking

A key insight from Kent:

“Programming is thinking. AI can help with the typing, but the thinking is still ours.”

What AI Can’t Replace

• Understanding the problem domain
• Making architectural decisions
• Knowing what to build vs. how to build it

What AI Does Well

• Boilerplate generation
• Syntax and API lookup
• Pattern application

Practical Advice

For Individual Developers

1. Experiment actively - Try AI tools on real work
2. Maintain skills - Don’t let fundamentals atrophy
3. Focus on judgment - That’s your value-add

For Teams

1. Share learnings - What works, what doesn’t
2. Update practices - Adapt workflows to include AI
3. Keep testing - AI-generated code needs verification

On Software Craftsmanship

Kent reflects on craft in the AI age:

• Craft isn’t about typing speed
• Understanding and judgment define expertise
• AI raises the floor, humans raise the ceiling

The Economics of AI Coding

Interesting discussion on costs:

• AI tools have real costs (API calls, subscriptions)
• ROI varies by task type
• Not always cheaper than human coding

Looking Forward

Kent’s predictions:

• Short term: AI as sophisticated autocomplete
• Medium term: AI handling larger code chunks
• Long term: Fundamental shift in what “programming” means

But throughout: human judgment remains central.

Notable Quotes

“The best code is code you don’t have to write. AI helps with that.”

“Tests are a conversation with the future. AI doesn’t change that.”

“I’m more optimistic about AI than I expected to be, but also more cautious.”