Is Agile Dead? XP Might Be the Future

First of all, sorry for the title, but I had to. We are living in a world where Everything is "dead". So I had to question Agile too.

I've never been a fan of Agile ceremonies. Stand-ups that drag on, sprint plannings that feel like theater, retros where nothing actually changes. The Agile Manifesto was beautiful in its simplicity. What we got instead was certification programs, expensive consultants, and a lot of meetings.

But then AI coding agents showed up. And they're making us rediscover practices that Agile to me seems outdated.

Enter Extreme Programming (again)

Kent Beck created XP in the late 90s. Pair programming, TDD, continuous integration, small releases. Back then, most of these practices were considered "too expensive" for normal teams. You need two developers for pair programming? That's double the cost! Writing tests before code? That's slower!

Except now, with AI agents, the cost equation has completely flipped.

In a recent podcast with Gergely Orosz, Kent Beck said something that stuck with me:

"The whole landscape of what's 'cheap' and what's 'expensive' has all just shifted. Things that we didn't do because we assumed they were going to be expensive or hard just got ridiculously cheap."

Pair programming? Your AI is your pair now. 24/7, no ego, infinite patience.

TDD? The agent writes tests faster than you can type the requirements.

Refactoring? Tell the agent to clean up the code and grab a coffee or do something else.

TDD is a superpower now

We are not freed from bugs. And let me tell you a liitle secret about AI coding: they introduce bugs too. They're confident, they're fast, and they'll happily break your existing features while implementing new ones.

TDD fixes this. The red/green cycle becomes your safety net.

Simon Willison wrote about this in his Agentic Engineering Patterns:

"Test-first development helps protect against both of these common mistakes, and also ensures a robust automated test suite that protects against future regressions."

The magic is in the "red" phase. You write a test, watch it fail, then let the agent implement. If you skip watching it fail, you might end up with tests that pass for the wrong reasons. The agent might write a test that already passes without actually testing anything.

What about Product Managers?

This is where it gets interesting. We've been talking about developers, but what about the rest of the team?
Product managers spend a lot of time trying to understand user flows, identify bugs, and spec out new features. They're often at the mercy of whatever engineers have time to investigate.

But now we have tools like Chrome DevTools MCP. This lets AI agents control a real browser, inspect network requests, check console errors, and test user flows end-to-end.

Imagine a PM asking their AI assistant: "Navigate through the checkout flow and tell me where users might get stuck." The agent opens the browser, clicks through the flow, checks for errors, analyzes load times, and reports back.

Or: "Find all the console errors on the dashboard page and categorize them by severity."

Or: "Compare the login flow on mobile vs desktop and document the differences."

PMs don't need to wait for engineering bandwidth to do basic investigation anymore. They can show up to meetings with actual data, actual screenshots, actual reproduction steps, but using AI too.

This bug report could have been a prompt

What XP got right was the focus on tight feedback loops. Write a test, implement, refactor. Small iterations, constant validation.

AI agents amplify this. The cost of writing tests drops to near zero. The cost of refactoring drops to near zero. The cost of trying something and throwing it away if it doesn't work? Near zero.

Maybe Agile isn't dead. Maybe we just need to go back to what worked before we turned it into a religion. Less ceremony, more coding. Less planning, more shipping. Less process, more tests.

Now it is more important than ever to have a clean project, perfectly organized, perfectly tested; The perfect feedback loop. The more tests, the more SOLID principles applied adapted to your needs, the better your AI Agent works. This of course, comes with more issues, like learn to be asynchronous with your tasks, and context switching, but that's for another blog post.

Kent Beck is 52 years into his programming career and he's more excited than ever. That's saying something.

What a time to be alive.