Two kinds of outcomes show up here. Custom AI software — pipelines, platforms, and backends built from scratch and handed over with the source code. And AI environments built for teams: the toolchains, agent configs, and knowledge layers that change how a company's engineers and operators actually work. Both shapes are shown below. Case studies are short, concrete, and honest about what was shipped versus what is still in flight.
A gachapon game — the pull-collect-and-build loop — built from scratch and now in alpha at rumbleofages.com. A different lane from the AI work, same discipline: a real product shipped solo and put live for players to break. Alpha means what it says — playable and improving in the open, not a finished launch.
Play the alpha →A booking platform for a creator who needed a workflow that did not exist in any off-the-shelf tool. AI-assisted matching between event briefs and the creator's availability, a clean public booking surface, and an admin panel the operator actually wants to open every morning. Started life as a Zapier mess; ended as a typed Next.js app.
Visit the site →Marketing site plus the content automation backend that feeds it. Scheduled generation, prompt library, output review queue, and a small dashboard for the team running the content calendar. Boring infrastructure done right so the AI work on top has somewhere stable to live.
Visit the site →Small content team's shared Claude Code environment — custom skills for the team's writing rules, MCP wiring into the internal CMS and asset library, and a knowledge layer over the brand docs. Role configs so writers and editors get different surfaces over the same stack. Case study writeup pending client sign-off.
One developer's setup, dialed in over about a week. Claude Code as the terminal-first agent, Cursor for IDE-level edits, both pointed at the same project context and knowledge layer so they hand off cleanly instead of fighting. Writeup pending — the engagement is recent.
The tag on each case study — Software build, Tuneup, Workspace — names the recognizable shape the engagement actually took, not a tier in a SaaS price grid. A Software build is what it sounds like: a new thing that needed to exist. A Tuneup is one person's AI environment, dialed in. A Workspace is a small team's shared setup. A Rollout is the agency-scale version. A Quick fix is a single-day job with one deliverable.
You do not pick a shape in advance. The duck reads what you describe on the call and names the shape that fits. Most engagements end up being one shape, cleanly. Some are a build plus a Tuneup on the same project. The duck quotes whichever combination is honest.
The full shape breakdown lives on the services page. The case studies marked "coming soon" will fill in as engagements wrap and clients sign off on writeups.
Describe what your team is doing, what is broken, what you wish was true. The duck reads the shape, quotes the work, and either books the next step or sends you somewhere better. Thirty minutes. No scoping form. No slide deck. Real numbers on the call.