Dockett: AI Agents SaaS Website

Dockett is a concept website for a fictional startup whose AI agents handle back-office requests for SaaS companies: refunds, invoice matching, plan changes, and onboarding paperwork. The agents pick requests up from email, Slack, and the helpdesk, do the work in the company’s own tools, and ask a person before anything risky.

The buyer is a head of operations who is tired of “AI for everything” pitches. So the site drops the usual AI look, purple gradients, glowing orbs, chatbot bubbles, and robot faces, and speaks the language of the work itself: tickets, labels, trays, and stamps.

At the centre of the page is a tactile, toy-like 3D sorting machine. Requests arrive as labelled cards down chutes, and as you scroll the camera follows them through 4 steps:

Pick up: Cards drop in from labelled chutes.

Read: A card turns face-up and its key fields highlight.

Do: A small arm routes it to the right tool tray.

Check: Risky requests stop at an approval gate and wait for a person.

Then it’s your turn: pick a sample request, a refund, plan change, or missing invoice, and send it through the machine while a log beside it prints each step.

Light, warm, and calm, with one accent colour that always means the same thing. Clean, product-led layouts with generous spacing, the standard SaaS buyers compare against.

The machine is one persistent 3D canvas behind the key sections, while everything else is plain, fast HTML. Every animation runs on one shared clock, so the 3D, the step list, and the log never fall out of sync.

A dashboard preview with a request queue, agent log, and approvals waiting.

Use cases for support, finance, and people operations.

Integrations, security, and a book-a-demo page.

Research: Award-winning sites recorded and broken down frame by frame, including an AI product site that deliberately removed “AI” from its own look.

Build brief: The 4-step workflow story, the interactive “send a request” moment, phone layouts, and a performance budget.

Build: Implemented with React, Three.js, and GSAP by an AI coding agent from the brief, then deployed on Vercel.

Let’s talk.