Every week there's a new headline about AI making web developers obsolete. No-code AI tools that build entire websites from a text prompt. GPT-4 writing entire applications. The era of human-written code is over.
We write code every day. Here's what's actually changed — and what hasn't.
What AI Has Genuinely Improved
Boilerplate and scaffolding
The most tedious part of starting a new project is the boilerplate. Setting up routing, writing CRUD operations, configuring build tools, scaffolding authentication flows — this is necessary but largely formulaic work.
AI code assistants (GitHub Copilot, Cursor, Claude) have dramatically accelerated this. What used to take a half-day of mechanical typing now takes an hour of review and adjustment. The developer's job shifts from typing boilerplate to evaluating what the AI generated.
This is unambiguously good. Experienced developers spend more time on architecture, business logic, and edge cases — the parts that actually require judgment.
Documentation lookup
Before AI assistants, a developer might spend 20 minutes finding the right MDN article, reading the relevant section, and synthesising an answer. Now, most API questions can be answered in seconds with a context-aware response that includes code examples.
Stack Overflow is still valuable. Documentation still matters. But the friction of looking things up has nearly disappeared.
Code review at speed
AI can review a pull request, identify obvious bugs, flag security issues, and suggest improvements — instantly. It's not a replacement for human code review (it misses context, business logic bugs, and architectural concerns constantly) but it catches the low-hanging fruit before a human has to.
Accessibility and testing
AI tools are increasingly good at flagging missing alt attributes, poor contrast ratios, missing ARIA labels, and other accessibility issues that human reviewers sometimes overlook. Automated accessibility scanning with AI-assisted remediation suggestions is a meaningful improvement.
What AI Has Not Changed
Judgment about what to build
AI can tell you how to implement a feature. It cannot tell you whether you should build it, whether it solves the actual user problem, or whether there's a simpler approach you're missing.
Product thinking, user research, and architectural decision-making require understanding business context, constraints, and goals that AI simply doesn't have access to. The best developers we know are spending more time on this thinking, not less — because AI handles the mechanical parts, the thinking becomes the distinguishing contribution.
Security thinking
AI-generated code has well-documented security failure modes. It will introduce SQL injection vulnerabilities, generate insecure random number generation, suggest deprecated cryptographic functions, and create authentication flows with subtle flaws — all with complete confidence.
A developer who doesn't understand security cannot rely on AI to make their code secure. The security layer still requires expertise.
Performance architecture
Choosing the right data structure, understanding database query plans, deciding when to cache and what to invalidate, designing for horizontal scale — these decisions happen before a line of code is written. AI can help implement a decision; it struggles to make one with incomplete information.
Client communication
Understanding what a client actually needs (as opposed to what they've asked for) is a human skill. Translating vague business requirements into a technical specification, managing expectations, explaining trade-offs in plain language — none of this is automated.
The "AI Website Builder" Question
Clients regularly ask us whether they should just use Wix AI, Squarespace AI, or one of the new AI website generators.
Our honest answer: for a simple brochure site with no custom functionality, they can produce a reasonable starting point. For anything that needs to:
- Integrate with external systems
- Have custom authentication or user accounts
- Handle transactions
- Perform well in search
- Be maintainable over time
- Do anything genuinely unique to your business
...AI website builders fall short. They produce generic, template-constrained output that looks similar to everything else the AI has generated, performs poorly under load, and creates technical debt that's expensive to unwind.
The gap between "a website that exists" and "a website that works for your business" is where developers add value. AI has not closed that gap.
The Developer's Changing Role
The practical effect of AI on our work is this: we write fewer lines of boilerplate and spend more time on:
- Understanding the problem deeply before writing anything
- Reviewing and validating AI-generated code rather than trusting it
- Architecture and system design
- Security review
- Performance profiling
- Client communication and requirements clarification
The developer who thrives with AI tools is not the one who lets the AI write everything. It's the one who uses AI to accelerate the mechanical parts and redirects that time toward the parts that require genuine expertise.
What This Means for Businesses Commissioning Web Development
AI tools have made good developers faster. They have not made non-developers into good developers.
If you're considering commissioning a web project in 2025, the AI tools available to your development team will likely mean:
- Faster turnaround on boilerplate and standard features
- More time available for custom, business-specific logic
- Potentially lower cost for straightforward features
What you should still scrutinise carefully:
- Security of the authentication and data handling
- Performance under real load
- Code quality and maintainability (AI output needs review)
- Whether the solution actually fits your specific business need
The fundamentals haven't changed. AI is a better tool, not a different discipline.
Talk to us about your project — we use AI tools where they help, and our judgment everywhere else.