“AI killed web design” makes a neat headline. Craig Burgess has a messier account from inside a working agency. He says he has built websites for 20 years, run an agency for 16, uses AI every day and is busier than ever. Some of that work now involves repairing sites clients generated themselves. His example is wonderfully unglamorous: a contact form that does not deliver enquiries.
AI can make a website faster to produce. It does not, by itself, establish that the site serves the right customer, delivers leads, meets relevant standards or has someone accountable when it breaks. Craig's busy inbox is evidence about his agency, not a measurement of the whole web-design market.
Watch Craig Burgess's Take
What His Experience Does and Does Not Show
Craig has heard the “your job is gone” line through several technology cycles. This time he concedes that AI has genuinely changed production: his team uses it to code, diagnose and repair websites. What he disputes is the jump from “a model can generate a site” to “a business no longer needs web expertise.” He describes AI as a junior developer. That is his working analogy, not a universal ceiling on model capability; the quality of agent-built work varies with the task, tools, supervision and test process.
He also reports a recent £10,000 client invoice. It shows that one client paid his agency; it does not establish what websites generally cost or whether demand is rising across the sector. Likewise, a do-it-yourself site can be a sensible choice for a small, low-risk project. The useful distinction is not “human versus AI” but who defines success, checks the result and owns the consequences.
The Rescue Job Is a Broken Contact Form
Craig says several clients came to him after building a site with ChatGPT or Claude because the form did not work. Their prompt histories even contained troubleshooting instructions. The missing piece was the ability to translate those suggestions into a working request, server-side handler or email delivery path. An attractive page can therefore fail at the one step a customer and business both care about: making contact.
A form is not “done” when a button animates or a success message appears. Test it on the deployed domain with a real address; confirm the data reaches the intended inbox; check validation, failure feedback, spam handling and the person responsible for follow-up. MDN's form-data guide explains the browser-to-server submission path and why server-side handling matters. Craig's examples are reported rescue jobs, not an audit of all AI-built sites.
What a Business-Ready Website Must Prove
Before accepting a launch, ask for a short demonstration of each important journey. The checks below are an editorial acceptance test, not a claim that a designer must perform every task manually.
| Decision or system | Evidence to request | Why it matters |
|---|---|---|
| Audience and offer | One named audience, one primary action and copy a real customer can understand | A polished template cannot choose the business objective |
| Enquiries and sales | Complete a test enquiry or purchase on mobile; verify delivery and follow-up | A visually finished page may not complete the transaction |
| Search discovery | Inspect public URLs, page titles, useful content and crawl access | Google's SEO Starter Guide focuses on helping search engines understand and users find content, not guaranteed rankings |
| Accessibility | Use keyboard-only navigation, readable labels and a relevant WCAG review | WCAG 2.2 provides testable accessibility criteria |
| Security and ownership | Review permissions, dependencies, updates, backups and the incident contact | OWASP ASVS supplies a framework for verifying web-application security controls |
Not every brochure site needs the same depth of assessment as an ecommerce store or an authenticated app. Scope the test to the actual risk, then put a name beside each untested item. “AI wrote the code” is neither a pass nor a fail.
A Better AI-and-Designer Workflow
Craig sees a legitimate use for AI before an agency engagement: review the current site, sketch a wireframe and turn vague preferences into a clearer brief. His team then uses AI again during implementation and diagnosis. That suggests a more productive workflow than the replacement debate:
- Define the job. Specify the customer, conversion action, content owner and constraints.
- Explore quickly. Use AI for alternatives, draft structure and a first implementation, keeping the source and assumptions visible.
- Review with expertise. Check the offer, copy, accessibility, search setup, integrations and security against the brief.
- Test real journeys. Run them on the deployed site, record failures and retest fixes.
- Assign ownership. Decide who updates content, monitors enquiries and responds when a service fails.
The designer's value may move from drawing every pixel or typing every line toward research, direction, integration and verification. That is an interpretation of Craig's experience, not a promise that every agency can preserve its old rates or workflow.
Questions Worth Asking Before You Buy or Build
If you are hiring: what customer action will the site improve, what will be tested before sign-off, who owns the domain and source files, and who fixes a failed form after launch? If you are building with AI yourself: can you run those tests, understand a failure report and maintain the result without relying on an unreviewed prompt chain?
Craig's argument is strongest when it stays concrete. A business is not buying a pile of HTML. It is buying a reliable way for the right people to understand an offer and act on it. AI can help produce that system. Someone still has to prove it works.
Video Chapters
| Time | Topic |
|---|---|
| 00:00 | The debate on AI and web design |
| 00:56 | Recurring technology hype cycles |
| 01:59 | AI as a coding assistant |
| 03:16 | The tractor analogy |
| 05:06 | Rescue jobs and practical AI uses |
| 07:00 | Professionalism and automation |
| 08:19 | What might change next |
Sources and Limitations
Based on Craig Burgess's video and its supplied transcript. His client stories and agency demand are first-person reports, not independently audited market data. The launch checklist is JQ AI SYSTEMS editorial guidance, informed by the primary technical sources below. Source pages checked 13 November 2026.