Everything about the work changed. Nothing about the business did.

One thing changed, and it changed everything

Asked what is different, Langlois starts by saying how much is not. “Much of the industry remains the same. We’re still building and managing websites for our clients.” Then the exception, which turns out to be the whole piece. “The biggest change has been the introduction of mainstream AI, which has had a significant impact on how we work and nearly every aspect of what we do.”

What clients ask for has grown rather than shifted. The new requests are AEO, GEO and LLM optimisation, plus automation work — all of it aimed, he says, at improving “how their websites are discovered, understood, and used by both people and AI-powered tools.” Nothing has dropped off the list to make room. “It’s less about clients asking for certain things less often, and more about them asking for additional services on top of what they already need from us.”

And the assumption that AI lets companies bring this work back in-house does not match what he sees. “No. In fact, I find clients are outsourcing more of this work now than they were two years ago.” They are looking for outside expertise, not trying to handle it internally.

A client banned AI in the contract

One client wrote a clause into their agreement prohibiting the use of AI altogether. It did not survive the conversation. “They removed it after I explained how broadly AI is now embedded in the tools and workflows we use, and that a blanket prohibition simply wasn’t practical if they wanted to work with us.” The door stayed open on the sensible version: “We were happy to discuss appropriate safeguards around how AI is used, but we couldn’t agree to not using it at all.”

Clients have also arrived with code they generated themselves. Twice, and both times it was bad. The team explained the risk of putting production code written by a non-technical person into a live site without understanding what it does. Both clients agreed and asked for it to be rebuilt properly.

The gain inside the studio is real and specific. Work that took a day can take a couple of hours. AI catches the small things in QA so people can spend their attention on the testing that needs judgement. The cost is the part nobody puts in a case study: reviewing its output, correcting it, trying new tools, and working out where it genuinely improves the workflow “rather than simply adding another layer to it.”

The experiment that failed was the obvious one. They had AI build an entire website. It “failed pretty miserably” — and confirmed what he already suspected. “The better the person using it understands the work, the more useful AI becomes.”

There is no mention of AI anywhere on the company’s website, and the new site they are building now will not carry much either. “We use AI extensively across our business, but I don’t see that as a selling feature or a meaningful differentiator. It’s a tool we use to do better work, not the reason someone should choose to work with us.”

The bill went up, and nothing came off it

Two things were added in two years: Vite as a build tool, and Cursor as an AI assistant. Almost everything else stayed. The effect on the invoice is the opposite of the one usually promised. “Definitely more expensive. AI hasn’t replaced any of our software, it’s just been added on.”

What did come off the bill came off for an older reason entirely. Adobe Creative Cloud had been their design tool for years, “but it just became too expensive.” They moved to open source. Notion went the same way: the internal documentation it used to hold now sits on a docs site they built themselves.

The tracker was part of choosing the tracker’s host

The workflow is four steps and no ceremony. A client request comes in, a team lead writes the issue and assigns a developer in Linear, it moves through To Do, QA and launch, and the developer replies to the client.

Linear replaced Notion, which “was just lacking basic functionality.” They looked at several alternatives and found most of them too complicated for what they needed. “We wanted something simple to match our workflow. Then we found Linear, which we love.”

Asked what he still does by hand and wishes were automated, he picks something closer to home than expected. “Our monthly timesheets are a bit of a pain. Don’t get me wrong, Everhour is fantastic. It just lacks a bit of flexibility in the way we download reports for our clients. That remains a manual process each month.”

“I wouldn’t expect to hear that it’s a lot of effort any more”

The team has not changed size and has always been remote. They hire intermediate and senior developers, and not many juniors.

His advice to anyone starting out has not moved in twenty years, and he is clear that it still holds: become expert in three languages, HTML, CSS and JavaScript. React, Vue and TypeScript are flavours of those. What he would add now is the other half of the job. “Understanding how to write prompts and guide AI to build good quality code is a must for anyone joining our team.”

What has changed is what counts as a hard problem. “Problems can always be solved. I wouldn’t expect to hear ‘that would be a significant amount of effort to build that’ any more.” The same goes for the things that used to get postponed to a second phase — responsive design, accessibility, technical SEO — which can now be handled during the build.

And the commercial side of the business, through all of it, has not moved. The company charges a monthly retainer for a set number of hours, “and it’s always been like that for 20+ years.”

Asked what humans will still be doing in five years, he does not hedge. “I honestly think developers will always be needed to guide and verify AI.”