A popular piece of advice says you should only use AI when it does a better job than a person, and that saving time on its own is not a good enough reason.
I think that advice is wrong. It sets up a false choice between better work and faster work, and it misunderstands how time works in a professional services business. When AI takes over the mechanical parts of a job, the time saved does not vanish. It goes back into the parts that matter: planning, testing and getting the detail right.
We see this every day at Accent. AI can now help us produce production-ready code in days that would once have taken months. Yet we now spend more time planning projects, not less, because that is where the saved time is best spent, and it is what makes AI safe to use on live work.
That does not mean AI belongs everywhere. After 27 years of building websites and web applications, these are the three questions I ask before handing any task to it.
1. What is the cost of being 90% right?
AI is probabilistic. It produces the most likely answer, which is usually a very good answer, but not a guaranteed one.
For many tasks, 90% right is a brilliant result. A first draft of a page structure, a set of headings or example copy for a client to react to is far better than a blank page, and a person will refine it anyway.
For other tasks, 90% right is a disaster. Financial reconciliation, compliance and data migrations need to be 100% right. If an AI writes a 2,000-line migration script in five seconds, but an engineer then has to spend four hours auditing every line to make sure it will not lose a client's data, nothing has been saved. You have simply swapped writing code for the more stressful job of checking code nobody on the team wrote.
Where we learned this. For Kombat UK, we migrated more than 15,000 products, built from over 70,000 records in a bespoke legacy MySQL database, into a WooCommerce store running on MariaDB. AI is very good at writing migration scripts quickly, and the scripts it produced looked right. But a legacy database built up over years holds knowledge that is not written down anywhere: why a field exists, which records are exceptions, what the business actually relies on.
So our engineers designed the mapping, the data rules and the checks themselves, and AI was limited to well-defined supporting jobs such as tidying messy text, writing test data for known edge cases and producing repetitive boilerplate.
The lesson: AI writes for the happy path. Experienced engineers plan for the exceptions, and real business data is full of exceptions.
2. Is AI automating a mess, or refining a proven system?
Give AI a chaotic, poorly documented process and it will produce chaotic results, only faster. It works best as an accelerator for something people already do reliably.
Our website process worked long before AI was part of it, which is exactly why AI improves it:
- The brief. AI helps us turn meeting notes and emails into a structured brief and spots gaps in what the client has told us. Agreeing the brief, and understanding what the client actually means rather than just what they wrote, stays with us.
- Content-first blueprints. Every page is planned before anything is designed. AI drafts the structure, headings and placeholder copy, but the client supplies the real facts: prices, services, qualifications and history. That keeps invented claims off the live site.
- Sign-off. Everyone approves the blueprint before design starts.
- Design and build. We choose a visual direction, develop it in Claude Design, and Claude Code handles the build and later updates, with a person reviewing at every stage.
What this made possible. The new PS English website is fully bilingual in English and Japanese, with matching pages linked between the two languages and more than 900 redirects from the old site. Keeping design and content consistent across two languages is genuinely difficult. With a proven process underneath and AI handling the repetitive work, the result was more consistent and better performing than we could practically have achieved by hand, with no compromises on functionality. If we could think of it, we could build it properly.
What happens without the system. One long-standing client began generating parts of their own site with AI to move faster. It all looked fine, and that was the problem: AI output always looks plausible. Underneath were errors that only showed up once they were live, and tracing and fixing them cost more than doing the work properly in the first place. The AI was not really at fault. Nobody with the right experience had checked its work before it was published.
3. Does this make our work look like everyone else's?
AI drifts towards the average. Ask it for a design, a tone of voice or a strategy and you will get something sensible, competent and very similar to what everyone else is getting.
That matters, because the distinct viewpoint and craft are what clients pay a premium for.
We noticed this in our own work. Each design coming out of Claude Design was sensible on its own, but looking across a whole run of client projects, they were starting to resemble one another. No AI working one project at a time would ever have flagged that. It took a person looking across the whole portfolio. So we added a deliberate stage to explore several different visual directions first, and only then define the brand and take the chosen direction forward.
The same thinking applies to imagery. We use real photography wherever authenticity matters. AI imagery is great for illustrative and campaign work, but it should never stand in for real people, real products or real places.
Our house rules
These rules apply to every project, whatever tools we are using that month:
- A person reviews everything before it reaches a client or goes live: briefs, content, designs and code.
- The client signs off the brief and the content blueprint before design starts.
- We are open with clients about using AI in our process.
- No personal data goes into AI tools. Given the software and security work we do, that is non-negotiable.
- No AI reviews or testimonials. They come from real clients, in their own words, or not at all.
- Real images first. AI imagery is for illustration, never a substitute for the real thing.
What clients are really paying for
During one project review, a client noticed we had delivered a large set of automated tests and data transformation scripts in roughly a third of the expected time. Their Managing Director asked us outright whether we had used AI to write it.
We said yes, and showed them exactly how: AI had produced the repetitive scaffolding and test data, while every data rule and security check had been designed and audited by our team. In short, the AI did the typing and we did the thinking.
Their response summed it up. They did not mind how the code was typed, as long as an experienced engineer was accountable if anything went wrong at two in the morning. They were not paying for typing speed. They were paying for the confidence that it works.
Keeping up without chasing every new tool
Something genuinely useful appears almost every month, so keeping up is a team effort. Everyone at Accent is encouraged to try new tools and share what they learn. But a new tool earns its place by solving a problem we already have, and it is always tested on our own work before it goes anywhere near a client project.
We are also happy to drop things. We recently moved away from Cloudways hosting because it was slow to adopt new technology and, at the time, did not support MCP, the standard that lets AI tools work directly with other services.
The takeaway
AI is an accelerator for mechanical work, not a substitute for judgement and accountability.
Clients do not pay for hours spent typing boilerplate, formatting data or drafting routine copy. They pay for context, taste and someone willing to stand behind the result when things get complicated. Use AI to remove friction, not to hand over responsibility, and you can deliver faster without turning your business into a commodity.
So here is a question worth asking about your own business: where does your team spend the most time on purely mechanical work that could safely be sped up?
If you would like help answering it, talk to the Accent team.
