Agile4AIBlog

The hammer and the screwdriver

Asking whether Waterfall or Agile is better is like asking whether a hammer or a screwdriver is better. The question is wrong. And the same wrong question is being asked about AI right now.

Which is better — a hammer or a screwdriver?

It sounds like a silly question, because it is. They’re different tools for different jobs. The hammer is better for driving nails. The screwdriver is better for driving screws. Neither is better in the abstract. The question doesn’t have an answer because the question has the wrong shape.

Now ask: which is better, Waterfall or Agile?

Same wrong shape. Same missing answer. And yet the debate persists, with Agilists claiming Agile is better and traditionally-minded practitioners defending Waterfall, and the back-and-forth producing heat without useful light because neither side has located the actual question.

What each approach is for

Waterfall works for simple and complicated work: work that’s been done before, is well understood, and can be planned in advance with reasonable confidence. Building a house. Manufacturing a product. Running a payroll system. You know the steps, you know the sequence, you can create a detailed plan and execute it. Waterfall is not just acceptable for this work — it’s often optimal. Detailed upfront planning plus disciplined sequential execution produces predictable, efficient results when the work itself is predictable.

Agile works for complex and chaotic work: work that’s new, uncertain, or changing rapidly. Building a product nobody has built before. Responding to a market that’s shifting. Integrating a technology whose behavior you’re still learning — like AI. You can’t create a detailed upfront plan for this work because you don’t have the information that plan would require. You discover the path as you go, adapting to what you learn. Agile provides the structure for doing that intelligently rather than haphazardly.

Neither approach is better. They’re designed for different types of work, and using either on the wrong type produces predictable failure. Waterfall applied to complex work produces plans that become obsolete before they’re finished and implementations that are faithful to the wrong specification. Agile applied to simple work adds unnecessary complexity, removes the efficiency benefits that predictable work makes available, and leaves teams making decisions they didn’t need to make.

The real question is never “Waterfall or Agile?” It’s: what kind of work is this?

How both approaches arrived by accident

Here’s a piece of history worth knowing: both Waterfall and Agile were, in a meaningful sense, accidents.

Winston Royce wrote a paper in 1970 about managing software projects. His intent was to warn against sequential, one-pass development — to argue that it was risky and insufficient. The paper was misread as advocating for the very approach he was cautioning against. What we call Waterfall was extracted from a warning about Waterfall. The screwdriver was handed to someone who promptly used it as a hammer.

The Agile Manifesto’s authors weren’t trying to start a movement. Seventeen practitioners gathered in 2001 to find common ground between their different lightweight approaches — trying to popularize practices they’d already developed. What they discovered in the process of articulating why their practices worked was something larger than the practices: a set of values and a cultural orientation toward people, collaboration, and adaptability. They were promoting processes and accidentally uncovered a cultural revolution.

Neither Waterfall nor Agile was deliberately designed as a complete approach. Both emerged from practitioners trying to solve specific problems, and both were subsequently generalized beyond what their origins could support.

Hammering screws

The most common failure pattern in Agile adoption is giving organizations a screwdriver and watching them use it as a hammer.

Organizations that adopt Agile practices without adopting the Agile mindset — that implement standups, sprints, and retrospectives while continuing to operate with Waterfall thinking — are hammering screws. The practice is there; the purpose isn’t. The screwdriver makes a terrible hammer. The daily standup makes a terrible status meeting — because that’s not what it’s supposed to do. The output is frustration and a conclusion that “Agile doesn’t work,” when the actual problem is the misapplication.

This isn’t unique to any particular organization. It’s what happens when a new approach gets introduced before the thinking that makes it work has been established. The smartphone that gets used only to make calls. The jet engine attached to a matchbox car — you don’t get a faster car; you get an explosion.

The same error with AI

The Waterfall-versus-Agile debate has a contemporary parallel: organizations today are making the same classification error with AI.

The dominant organizational desire for AI is automation: take tasks humans currently do and have AI do them instead, faster and cheaper. This is a legitimate desire. And deterministic computing — traditional software — is the right approach to automation. Rules-based systems, scripted workflows, programmatic logic: these are excellent automation tools, exactly suited to work that can be fully specified in advance.

AI — probabilistic computing — is something different. Fully-specified, deterministic tasks are not where AI excels. What AI is good at is the complex, uncertain, contextual spaces that can’t be easily automated: synthesizing ambiguous information, generating options that weren’t pre-specified, reasoning through problems whose parameters keep changing.

Geoffrey Hinton — one of the foundational researchers in modern AI — puts that difference more starkly than most. He describes what we’re building as beings of a different kind. Whatever else that means, it isn’t a description of a tool. That is the distinction this piece has been making all along, pushed one level deeper: before choosing an approach, you ask what kind of work this is, and with AI you also have to ask what kind of collaborator you’re working with.

A tool is something you operate. Nobody reasons with a hammer. Automation wants repeatable, deterministic execution; AI’s usefulness lies in the variable and the contextual. Put AI on the droning, predictable work and the disappointment that follows is not a measure of AI’s capability — the request was aimed at the wrong kind of thing.

Organizations that insist on treating AI as automation are making the Waterfall-versus-Agile mistake in a new domain. They’re hammering screws. They’ll either fail to get the benefit that AI is actually capable of delivering, or they’ll pressure AI vendors to force AI into automation-shaped boxes — which is a different kind of misapplication with its own set of risks.

The question is the same one it’s always been: what kind of work is this? For automation, use the right automation approach. For the complex, uncertain, contextual work that actually benefits from AI’s capabilities, bring the mindset — and the culture — that complex work requires.

That’s the Agile mindset — suited to emergent, complex work, which is precisely the shape of the AI space. And that’s why Agile matters more now than it ever has.

Comments 0

Sort
Checking your sessionThe article is ready. Comment options will appear once your session check finishes.