AI was supposed to make projects simpler.
In many ways, it has. It can summarize meetings, organize requirements, identify technical considerations, write documentation, brainstorm functionality and help teams think through problems much faster than before.
But we’re beginning to see an interesting side effect in website projects:
AI is making relatively simple projects dramatically more complicated.
At Vinbound, we’ve started referring to this phenomenon as AI Scope Inflation (or the AI Complexity Spiral).
It happens when the cost of generating ideas, requirements, questions and documentation becomes so low that the volume of all four begins to expand.
A client who might have sent a five-paragraph website brief a few years ago can now provide a 10-page AI-assisted specification covering customer journeys, SEO requirements, analytics events, accessibility, mobile behavior, ecommerce scenarios, integrations and dozens of potential edge cases.
The agency receives that document and does the natural thing: uses AI to analyze it. That analysis uncovers another 25 questions.
The client puts those questions back into AI and generates detailed answers, additional requirements and several new ideas.
The agency responds with another round of analysis.
Before long, a relatively straightforward website redesign has developed the documentation and discovery process of an enterprise software implementation.
Everyone involved was trying to be thorough, but the project quietly became something very different from the project everyone originally agreed to build.
The Documentation Arms Race
Before generative AI, producing complexity required effort. If someone wanted to create a 15-page requirements document, they had to actually sit down and write a 15-page requirements document.
That friction served an unintended but useful purpose: it forced people to prioritize. Most clients focused on the things that mattered most:
- What pages do we need?
- What functionality does the website need?
- What does the brand need to communicate?
- How does ecommerce work?
- What content needs to migrate?
Today, generating an exhaustive list of considerations takes minutes.
Ask an AI system: “What should we consider when redesigning a winery website with ecommerce, a wine club, reservations and events?” and it can immediately produce dozens—or hundreds—of considerations. Many of those considerations are perfectly valid.
The problem is that valid does not necessarily mean necessary.
A project can theoretically account for every user journey, edge case, integration possibility, future marketing initiative, accessibility scenario, analytics event and technical optimization. But doing so has a cost. Every requirement must eventually be interpreted, designed, developed, tested, documented or explicitly excluded. AI has made it nearly free to propose complexity. Implementing that complexity is still very much not free.
The AI Complexity Spiral
This can create a feedback loop. The client uses AI to create a highly detailed brief. The agency uses AI to identify gaps in the brief. The client uses AI to answer those questions. Those answers introduce additional possibilities. The agency analyzes those possibilities and discovers additional dependencies.
Soon, both sides are producing extremely sophisticated documentation for functionality that may never materially improve the website. The irony is that AI can make everyone involved feel more productive while simultaneously making the project harder to complete. More documentation creates more decisions. More decisions create more dependencies. More dependencies create more testing. More testing creates more opportunities for revisions. And all of that can quietly expand the project far beyond its original scope.
This is where AI Complexity Inflation turns into AI-assisted scope creep.
Not Every Good Idea Belongs in the Project
This may be the most important principle for modern website projects: A good idea is not automatically a project requirement.
AI is exceptionally good at generating plausible improvements.
- Should the website support advanced event tracking? Probably.
- Should every button have a unique analytics event? Maybe.
- Should the project include structured data for every possible content type? Potentially.
- Should we build sophisticated customer segmentation into the website? That could be useful.
- Should we account for a future loyalty program the business might launch three years from now? Now we’re starting to drift.
The question should not be: “Could this make the website better?”
Almost anything could.
The better question is: “Is this important enough to justify adding time, cost and complexity to this specific project?”
That distinction is becoming increasingly important in the AI era.
How We Keep AI From Overengineering Website Projects
AI should improve the quality of a website project without allowing the project to expand indefinitely.
There are several ways we can accomplish that.
1. Establish the project boundary before deep discovery begins
Every website should have a clearly defined center of gravity.
For example: Redesign the winery website using the existing brand, migrate approximately 30 pages of content, integrate Commerce7 ecommerce, rebuild the primary tasting and events pages, and improve overall SEO structure.
That statement becomes the project’s anchor, and every new request can then be evaluated against it. Does this requirement help accomplish the project we agreed to build? Or does it represent a new initiative?
This makes scope conversations much easier because they become less subjective. The question isn’t whether an idea is good. The question is whether it belongs inside the agreed project.
2. Separate requirements from ideas
AI-generated briefs tend to flatten everything into one enormous list. But not everything on that list has equal importance. We prefer thinking in terms of three categories:
Required for launch
These are the things the website must do for the project to succeed.
Worth considering
These may improve the project, but they aren’t necessary to launch.
Future opportunities
These are good ideas that should be preserved without being allowed to complicate the current build.
That last category is especially important. Saying “not now” is very different from saying “no.”
Creating a future enhancements list allows everyone to capture good ideas without automatically turning those ideas into immediate development requirements.
3. Define complexity budgets
Projects traditionally have financial budgets and timelines. Increasingly, they also need a complexity budget.
If a $5,000 website is being built from an established template, the discovery and specification process should probably not resemble a six-month custom software project. The sophistication of the process should remain proportional to the sophistication of the product.
A useful question is: Are we spending more time discussing this feature than the feature is likely to save or earn the business?
If so, that is usually a warning sign.
4. Ask AI to prioritize, not just expand
Most people naturally use AI to ask: “What am I missing?”
That is useful but dangerous, because AI will almost always find something else.
A better prompt is: “Which five issues are most important to the success of this project? Ignore low-impact edge cases.”
Or: “Which of these requirements meaningfully affect launch, and which could safely be handled later?”
AI becomes much more valuable when it is used as an editor rather than an infinite brainstorming machine.
5. Require a reason for new requirements
When a new feature or requirement appears during a project, it should answer a simple question: What problem are we solving?
If the answer is clear, the requirement may be justified. If the answer is: “AI suggested this was a best practice,” that probably isn’t enough. Best practices need context. A feature that makes sense for a national ecommerce retailer may be unnecessary for a small winery selling a thousand cases per year.
Technical sophistication should serve business objectives, not exist for its own sake.
6. Recognize when a question doesn’t need an answer
This is surprisingly difficult because AI encourages us to resolve every ambiguity.
But successful projects have always contained reasonable assumptions.
If a low-impact decision can be handled using standard web conventions, experienced designers and developers should often simply make the decision. Not every button placement needs stakeholder discussion. Not every analytics event needs a meeting. Not every theoretical customer journey needs to be documented.
Professional judgment is partly the ability to know which questions matter enough to ask.
7. Create a clear change-control process
Some scope expansion is legitimate. Businesses change. New information emerges. Someone realizes an important requirement was overlooked.
When something meaningfully expands the project, everyone should understand:
- What is being added
- Why it is being added
- Whether it affects budget
- Whether it affects timeline
- Whether something else should be removed to compensate
This prevents scope creep from hiding inside dozens of seemingly harmless emails.
8. Optimize for outcomes, not documentation
The goal of a website project is not to produce the world’s most comprehensive website specification. The goal is to launch a website that generates revenue for the client with as little friction as possible.
Documentation should reduce uncertainty. Once it begins creating more work than clarity, it has stopped serving its purpose.
AI Is Still an Enormous Advantage
None of this is an argument against using AI in website projects. We use it constantly. It helps us analyze complicated requirements, think through integrations, identify SEO opportunities, and troubleshoot technical problems. It is an extraordinary tool.
But like most productivity technologies, its biggest weakness can be an extension of its greatest strength. AI makes thinking cheap. And when something becomes cheap, we tend to consume much more of it.
The challenge for agencies and clients will increasingly be deciding which thinking actually needs to become work. The best website projects of the AI era may not be the projects with the most documentation, the longest specifications or the greatest number of identified opportunities. They may be the projects where everyone involved becomes exceptionally good at separating what is possible from what is important.
AI can generate an almost unlimited number of ideas. Our job is still to decide which ones are worth building.
