How AI changes scoping, pricing and delivery.

AI coding assistants are at their best on well-specified implementation work: turning a design into components, wiring up a CMS, writing repetitive markup, building variations, scaffolding pages and fixing routine bugs. Work that used to take days of focused development can often be produced in hours, then refined.

They also make iteration cheaper. Trying three versions of a layout or animation, or rebuilding a section after feedback, costs far less than it used to. That changes how much exploration a project can afford.

The parts of a website project that depend on judgement and on other people move at roughly the same speed as before:

• Discovery and positioning — understanding the client's business, audience and goals.
• Content — gathering real information from the client, writing it well and getting it approved.
• Design direction — deciding what the site should feel like and making it distinctive.
• Stakeholder feedback and approvals.
• QA — accessibility, performance, cross-device testing and content accuracy.

Because implementation is faster, the quality of the specification matters more. A clear brief lets an AI-assisted team produce good work quickly; a vague one produces confident-looking work that misses the point just as quickly.

1. 01Define the outcome: what the site must achieve and how it will be measured.
2. 02Agree the sitemap and the purpose of every template.
3. 03Specify content sources and owners for each page before design starts.
4. 04Set the design direction with references and a small set of approved key screens.
5. 05Write acceptance criteria for performance, accessibility, SEO and structured data.
6. 06Fix the number of feedback rounds and what counts as a change request.

If a site is priced on development hours and those hours shrink, the price shrinks with them. That rewards the agency for being slow. Pricing on deliverables and outcomes is more sustainable for both sides.

• Package by scope: number of templates, integrations, motion level, content support and launch services.
• Use some of the time saved to add value the client can see — more polish, more motion, better content structure, GEO-ready markup — rather than only cutting the fee.
• Price discovery, content and strategy as distinct line items; they are where much of the value now sits.
• Offer ongoing improvement retainers. Faster iteration makes continuous optimisation practical in a way it rarely was before.

AI-assisted builds work best in short loops: specify a piece of work, generate it, review it, refine it, commit it. Each loop should end with a person checking the result against the specification, not just checking that it runs.

Keep the codebase conventional and well organised. A clear component structure, design tokens and consistent patterns make AI tools more accurate and make it easier for a developer to step in later.

Speed makes it tempting to skip review. Don't. The common failure modes of AI-generated code are subtle: accessibility gaps, inconsistent spacing, missing edge cases, duplicated logic and plausible but wrong content.

• Run automated checks: linting, type checking, accessibility scanning and performance budgets.
• Test on real devices, with keyboard navigation and with reduced-motion settings enabled.
• Review every page's copy for accuracy — never publish claims, statistics or testimonials that haven't been verified.
• Check structured data and metadata on every template before launch.