We shipped a full website migration in two weeks
Earlier this year, we took on a project for XL Biotec, a biotech supplier here in Thailand. They had a WordPress/WooCommerce site with over 440 products across 27 suppliers. The site was slow. Managing products was painful. Their CEO described WordPress as "a slog. A lot of clicking and waiting."
We proposed migrating the entire thing to Next.js with Sanity CMS, deployed on Vercel. A project like this, with data migration, SEO preservation, bilingual infrastructure, and a complete frontend rebuild, would normally take us four to five weeks. We estimated that in our proposal.
We delivered it in two.
AI was a big part of why. I used Claude Code throughout the build for scaffolding components, writing migration scripts to parse and clean 440 products from a WooCommerce CSV export, setting up the Sanity schemas, and handling the tedious but critical work of mapping Yoast SEO metadata to Next.js. The kind of work that used to eat days got compressed into hours.
But here's the thing: the reason it worked is because I knew what needed to be built. I wrote the PRD. I mapped out the data model. I decided that the WooCommerce checkout should be replaced with a Request a Quote flow because that's how 90% of their customers actually buy. AI didn't make those calls. I did, based on conversations with the client and understanding his business.
When Joe saw the preview, he said: "Wow. That seems so much FASTER!! This is great." His team's reaction was the same. "Everyone loves it so far. So fast. Everyone was shocked."
AI helped me build it faster. It didn't tell me what to build.
We built a privacy-first course platform from scratch
Another recent project was Intimia Institut, a German-language online platform for sexual education. This one was technically demanding, but the real challenge was the sensitivity of the subject matter.
Every design decision had to balance three things: privacy (users needed real anonymity, not just a checkbox), credibility (the platform had to feel like a serious educational institute, not a SaaS app), and commerce (live course bookings with Stripe, confirmation emails, reminders). Getting all three right at the same time, in a way that feels coherent, was the actual design problem.
I used AI tools heavily throughout the build. Claude Code helped with the backend architecture, the Stripe webhook integration, the magic-link authentication flow. When the data model needed to evolve mid-project to support multiple session dates per course, AI helped me think through the migration strategy and write the backfill logic.
But the decision to make anonymity a first-class design constraint from day one? That was a human call. The choice of an editorial design language, cream backgrounds, wine-red accents, serif headlines, to signal professionalism and warmth? That came from understanding the brief and the audience. The restraint in every confirmation email, keeping the tone short, friendly, and signed by the institute? That was judgment, not generation.
AI made me a faster engineer on this project. It didn't make me a better designer or a more empathetic product thinker. I had to bring that myself.
I built our own internal HR system
This one is personal. At AMS, we'd been using third-party HR tools for leave management and team coordination. They were clunky, generic, and didn't fit how we actually work. So I built our own system from scratch: LIFE AMS. It handles leave requests, team calendars, approval routing by department, employee profiles, and work-from-home quotas.
I'm not a backend engineer by training. I'm a web project manager who learned frontend along the way. A year ago, building a full internal tool with role-based access, notification routing, and PDPA compliance would have been beyond what I could reasonably tackle alone.
With AI-assisted development, using tools like Claude Code, Lovable, and Replit, I built it. Not by asking AI to "make me an HR app" and getting a finished product. It took weeks of iteration, feedback sessions with my boss, testing with our team, and constant refinement. The meeting notes from our review sessions have 14 action items across onboarding flows, approval routing logic, calendar views, and export features. That's real product work.
AI gave me the ability to punch above my weight technically. But I still had to manage the project, gather requirements, prioritize the backlog, and make judgment calls about what our team actually needs. The tool didn't replace the need for a project manager. It made the project manager more capable.
What AI actually does for us day to day
Let me be specific about where AI shows up in my daily workflow, because the vague "AI helps with everything" narrative is useless.
Writing migration scripts and data transformations. Parsing a WooCommerce CSV with 440 rows of messy HTML descriptions, cleaning it, and uploading it to a new CMS is exactly the kind of task where AI saves hours. The logic isn't complex, but the volume and edge cases are tedious. AI handles this well.
Scaffolding and boilerplate. Setting up a Next.js project with App Router, connecting a Sanity client, wiring up Zustand for cart state. These patterns exist in thousands of codebases. AI generates them quickly and I adjust to fit the project.
Debugging and problem-solving. When something breaks, I can describe the error and get targeted suggestions instead of scrolling through Stack Overflow threads from 2019. The feedback loop is tighter.
Content creation. Blog posts, case studies, LinkedIn content, client communications. I still write the ideas and the voice is mine, but AI helps me get from rough thoughts to polished drafts faster.
Image generation for mockups. We use AI image tools to create professional mockup environments for client presentations. Instead of spending hours in Photoshop compositing a screenshot into a lifestyle scene, I write a prompt and generate it in minutes.
What AI does not do
It doesn't talk to clients. It doesn't sit in a meeting and read the room when a CEO is frustrated with his slow WordPress site. It doesn't know that a biotech supplier's customers mostly pay by purchase order, so a checkout flow is wasted complexity. It doesn't understand that a sexual education platform needs an editorial magazine aesthetic to establish credibility.
It doesn't manage a project timeline. It doesn't negotiate scope when requirements change mid-build. It doesn't decide which features to cut when you're trying to ship in two weeks instead of five.
It doesn't know your team. It doesn't know that one colleague is the right person to test the HR system because she understands both the admin and employee perspectives. It doesn't know that your boss needs to approve the WFH policy before it goes into the system.
These are all things that happen every single day at a working agency. And they're all things that require a human being who understands the context, the relationships, and the business.
The real shift
The conversation about AI replacing jobs misses the point. The better question is: what does your job look like when the mechanical parts get faster?
For me, it means I spend less time writing boilerplate and more time thinking about product decisions. Less time wrestling with data migration scripts and more time talking to clients about what they actually need. Less time on the tedious parts of code and more time on the parts that require taste, judgment, and experience.
The work didn't shrink. If anything, it expanded. Because when you can build faster, you take on more ambitious projects. A year ago, I wouldn't have attempted to build an internal HR tool for our company. Now it's live and our team uses it every day. That's not a story about replacement. That's a story about what becomes possible when the tools get better.
AI didn't make our jobs unnecessary. It made us better at them.
Restons en contact.
Suivez Asia Media Studio pour des conseils en matière de design et des mises à jour sur Facebook, Instagram et LinkedIn.
