So when AI started changing how we build software, I expected it to clash with all that. Agile is about people, conversations and adapting as you go. AI sounds like the opposite: automation, tools, machines doing the work.

It turns out I had it backwards. The manifesto was written in 2001, long before anyone was asking an AI to write code, and it's built around four values. Every one of them is stronger now than it was before.

It's worth remembering what the authors actually said, too. They didn't say processes, documentation, contracts and plans were worthless. They said there's value in those things, but they value the people side more. That nuance matters, and I'll come back to it. Here's how each value stacks up.

Individuals and interactions over processes and tools

You'd think AI would push us towards tools and away from people. In practice it's done the reverse.

When the build takes less time, the time goes somewhere else. For us, it goes into conversations. Understanding what the client actually needs, challenging the brief when it needs challenging, and getting the decisions right before anyone writes a line.

The tools do the grinding. The people do the thinking. That's exactly the balance the manifesto was asking for.

Working software over comprehensive documentation

This is the one that surprises people, because we write more specs now than we ever did.

But there's documentation and there's documentation. The manifesto was pushing back on hundred-page requirements documents that nobody reads and everybody ignores. Our specs are nothing like that. They're short, focused, and written for one job: telling the build exactly what to do. If a spec doesn't change what gets built, it doesn't get written.

And the payoff is working software, sooner and more often. Every sprint ends with something the client can click on and try, not a progress report. If anything, we're further from comprehensive documentation than ever, and closer to working software.

Customer collaboration over contract negotiation

Hourly billing quietly turns every project into a negotiation. How many hours? Why did that take so long? Is this change in scope or out of scope?

We've moved to pricing per sprint, and it's taken most of that friction away. We agree with the client what goes into the next fortnight, at a fixed price, and then we get on with it together. The conversation is about what to build next, not about the invoice.

AI is what makes that possible. Sprints are faster and more predictable, so we can price them with confidence instead of padding estimates to cover the unknowns.

Responding to change over following a plan

Change used to be expensive. A client would realise halfway through a project that they needed something different, and everyone would wince at what it meant for the timeline.

Now each sprint is defined and priced on its own. Something more important comes up? It goes in the next sprint. Turns out a feature isn't needed after all? It drops out. Nobody has to tear up a six-month plan.

Because the build moves faster, the cost of being wrong is smaller too. Trying an idea, seeing it working and then changing course is cheap. That's agile the way it was always meant to work.

The catch

None of this happens by itself. AI without discipline is just vibe coding: fast, fun, and a headache six months later. Remember the manifesto's point that there's still value in processes, documentation and plans? This is where it counts. The agile values only get stronger if you keep the structure around them. Proper specs, proper review, proper testing, and people who own the result.

That's the bit we care about most. If you're curious how it changes what you pay, I've written about that too: Why we've stopped quoting by the hour.

And if you'd like to talk about a project, drop us a line.

Cheers, Rob