PonTech Lab Inc. | AI startup site, built with AI

Building an AI startup’s public face, all the way, with AI

We produced the corporate site for PonTech Lab Inc., an AI startup in Tama, Tokyo. The company develops AI solutions and provides SES and PM/PMO support, and puts particular effort into hiring people with Japan Self-Defense Forces experience. Making the face of an AI company with AI — this project used our AI Website Production service (Claude Code) to the full.

The structure covers six pages including home, company, services, careers and contact, plus news and blog update functions. We chose Next.js (React) and put the site on Cloudflare’s edge network. We started in March 2026 and completed the launch and the analytics setup the following April.

The expression we cared about

A dark-and-purple palette expresses the brand’s technical strength and discipline, and we spent our time on the design of motion. Move between pages and tiled blocks spread in sequence from the lower left to the upper right, switching the screen. On a first visit, a mizuhiki cord unties and the screen opens in 3D. Enormous typography drifts slowly across the background as you scroll. For the parallax we went back and forth on real devices many times between moving so much it looks messy and not being able to tell it moves at all, and ended up at one third of the original amount.

For the careers page we handled the message to former Self-Defense Forces personnel from the planning stage. Almost no usable photography existed, so we generated every visual on the careers page with AI — from aerial shots of a garrison to portraits of personnel. AI image generation turns “we cannot build it because we have no assets” on its head.

Behind the build — a record of stumbles and decisions

On our work pages we publish not only what went well but the places where we actually stumbled. The following is taken straight from the build history with AI.

It started with a rebuild. We first built the site as an SPA (a site that runs as a single application), but with that approach search engines can barely see the content of each page, which hurts SEO. Midway through we migrated entirely to the Next.js App Router (static generation). The tile transition animation we had put so much into was the part most likely to break in a migration, but because the switching mechanism had been isolated in a single layer, it carried over intact.

We subtracted too much, and put it back. Feeling there was too much text, we cut heavily — and ended up short on information and out of step with the client’s intent, so we restored all of it. From then on we tested large changes in reversible stages. We also introduced a fashionable library (GSAP) for scroll effects at one point, but it interfered with the existing motion, so we decided to go back to a lightweight in-house implementation.

We killed the problems that only appear in production. News articles worked locally but vanished on the production server (Cloudflare Workers) — the cause was a structural constraint: files cannot be read at runtime in that environment. We fixed it at the root by changing the design so articles are baked in at build time. A form overflowing on a real iPhone and vertical labels rendering upside down were also fixed one by one, with a record kept for each.

We scored the whole thing before launch. Before release we scored maintainability, SEO and accessibility objectively, then worked through eight rounds of revisions along the findings: separating content management, unifying design variables, adding structured data, and accessibility support. It is a process that holds building fast with AI and guaranteeing quality with people together.

Operations after launch

Since launch, update requests arrive by email and Webharu applies them. News and blog posts go live simply by adding a single text file, and we set up a monthly analytics report. We had prepared review and revision phases, but this project barely used them — because we show the work only once it has been taken as far as it goes.

For details on the AI website production process and the deliverables, see our explainer article and service page.