How is software actually built?
Behind every software project is a methodology — the rules for how the team plans, builds and ships. Here's every model you'll hear an agency mention, from Waterfall to Agile to AI-Native, and when each one actually makes sense.
What a methodology actually is.
Software development methodology, defined
A software development methodology is the framework a team uses to organise how software gets planned, built, tested and shipped. It defines the order of work, how often you see progress, and how easily the project can change direction — which is why the choice matters more to you as a client than any tool or language.
When an agency says “we work in two-week sprints” or “we'll need the full spec signed off first,” they're describing their methodology. It shapes three things you'll feel directly: when you first see something working, what it costs to change your mind, and how risk is shared between you and the team. The methodology also pairs with a commercial model — how you pay — which we cover separately in our guides to software pricing models and agency engagement models.
- What it decides
- The order of work — everything up front, or in short repeating cycles.
- What you feel
- How soon you see a working product, and what changing your mind costs.
- What it isn't
- A tool or a language — it's the process around them.
- Who should care
- Anyone hiring a team. Ask “how do you work?” before “what do you charge?”
The plan-first classics: Waterfall & Spiral.
The oldest approach is Waterfall: each stage must be finished before the next begins, like water flowing down steps. It's easy to plan and predictable — and painful when requirements change, because the customer doesn't see the product until late.
Waterfall still earns its keep where requirements genuinely are fixed — government systems, banking, medical software, construction. For big, high-stakes projects there's also the Spiral model: plan, analyse the risks, prototype, develop — then repeat the whole loop. It combines upfront planning with risk analysis, and shows up in aerospace, defence, and other very expensive builds.
The Agile family: ship, learn, repeat.
Agileflipped the model: instead of planning everything upfront, the team plans, builds, tests and ships in short cycles — usually one to four weeks — and the product improves continuously. It's probably the most common methodology today, and it spawned a whole family of frameworks.
DevOps & RAD: built to ship fast.
DevOpsisn't really a development methodology — it's a culture. Instead of developers handing finished code to a separate IT team, one team continuously ships software, with automation, CI/CD pipelines, monitoring and fast deployments. It's how modern products go from “release every quarter” to “release every day.”
Rapid Application Development (RAD)has one goal: build something quickly. Heavy use of prototypes, user feedback and iteration. Modern no-code tools and AI have made RAD more practical than it's ever been — and it's a direct ancestor of the AI-Native model below.
AI-Native development.
AI-Native development, defined
AI-Native development is an emerging methodology where the team and process are designed around AI from day one — not AI bolted onto an old workflow. Instead of months of upfront planning: discovery, a prototype in days, an early MVP, measurement, small targeted builds, and weekly improvement.
It isn't an official methodology yet — no manifesto, no certification — but it's the philosophy behind many AI-first teams today, and it changes the economics of everything earlier in this guide. Prototyping is so cheap that you test the idea before you fund it. Building is so much faster that the MVP launches while a Waterfall project would still be in requirements. And because small custom software is finally affordable, improvement happens through micro-builds — small features and mini apps shipped weekly — rather than infrequent, large rebuilds.
The honest caveat: AI accelerates, it doesn't engineer. Architecture, security and judgement still belong to humans — which is why AI-Native describes the process, not a replacement for the team.
Want this done for you?
We build AI-native — it's our default.
Train the AI on your business, prototype before you commit, launch in weeks, improve weekly. That's our AI development service, end to end.
Every model, side by side.
There's no single best methodology — only a best fit for how fixed your requirements are and how fast you need to learn. The honest summary:
| Method | Best for | Flexibility | Speed |
|---|---|---|---|
| Waterfall | Fixed, regulated projects | Low | Slow |
| Spiral | High-risk, expensive projects | Medium | Slow |
| Agile | Most modern software | High | Fast |
| Scrum | Product teams | High | Fast |
| Kanban | Maintenance & support | High | Continuous |
| Lean | Startups & MVPs | High | Fast |
| XP | Quality-critical software | High | Medium |
| DevOps | Continuous delivery | High | Very fast |
| RAD | Quick prototypes | High | Very fast |
| AI-Native | AI products & startups | Very high | Very fast |
Note: these are working characterisations, not lab measurements — real speed and flexibility depend on the team, the scope, and how clear the requirements are. Direction, not guarantees.
Methodologies, answered.
What is the difference between Waterfall and Agile?
Waterfall completes each stage — requirements, design, development, testing, deployment — before the next begins, so you see the product late and changes are expensive. Agile works in short repeating cycles, so the product improves continuously and priorities can change between cycles.
Waterfall suits fixed, regulated projects; Agile suits most modern software.
Is Scrum the same as Agile?
No — Scrum is a framework within Agile. Agile is the philosophy of short cycles and continuous improvement; Scrum is one specific way to run it, with fixed-length sprints, defined roles (Product Owner, Scrum Master), and rituals like planning, review and retrospective.
What is AI-Native development?
An emerging methodology where the team and process are built around AI from the start: discovery, a prototype in days, an early MVP, measurement, micro-builds, weekly improvement. It compresses months of planning into fast, evidence-led iterations.
It's how we work — see our AI development service for the done-for-you version.
Which methodology should I ask my agency about?
Ask three questions: when do I first see something working? What does it cost to change my mind? And how do we pay — which is really a question about pricing models and engagement models, covered in their own guides.
Does the methodology change what I pay?
Indirectly, yes. Plan-first methods pair naturally with fixed prices; iterative methods pair with time-and-materials, retainers or subscriptions, because the scope evolves.