AI-assisted delivery, and what it changed
I rebuilt the handoff around design intent that could reach code, with designers validating generated specs and engineering reviewing the build. The merged Org Strategy study follows the recovered capacity upstream into product shaping. This site turns that approach toward two readers, a person checking the work and an agent reading the same facts, with a review gate before an application goes out. The public template below contains only the agent layer, not the whole site or application pipeline.
The delivery model in full: two MCPs in one context window, a spec that got generated rather than written, and the stakeholder group that was still only partially convinced when Jason left.
Arcos, IncHow restructuring the design-to-dev pipeline at Arcos cut kickoff time from two weeks to three days: one shared design language for UX, engineering, and AI.
At Arcos, design specs were thorough but not executable. Developers made hundreds of small judgment calls per feature, and AI entering the pipeline automated the ambiguity instead of solving it. I led Harmony’s architecture and contributed hands-on at Arcos, including work with a senior designer on the token architecture. Other contributors owned or completed parts of the component and implementation work. Harmony was intended to carry design intent to designers, engineers and AI coding agents. The context-rich npm package was paired with Storybook and Zeplin MCPs. Across three piloted Arcos Control Tower cycles, kickoff compressed from two weeks to three days, and the crew-entry Timecards prototype I built after field testing took about 12 hours against an estimated ~40 for a wired Figma prototype.
What the pipeline revealed once it was running: the recovered time moved design upstream, and the ways-of-working documentation my team published became the training set for another pod’s agent.
Arcos, IncArcos had no UX function before I arrived. Credibility, a team that owned its domains, and a delivery model that freed time for shaping are what made design a necessary voice in what gets built, not just how it looks.
Arcos had tried to build a Timesheet capability once before UX existed, and it missed. The second attempt was shaped with aggregated user data and a coded prototype tested with roughly 400 field users before engineering committed, and it shipped in beta in June 2026, reaching production in September 2026 after I left. Along the way I built a team of three Senior Designers who owned their domains, and a delivery model that recovered half a sprint to a full sprint per cycle. The recovered time showed where design’s real work was: upstream, in purpose and the shape of the data.
- Artifact
agent-layer-template
qaContext, llms.txt generation, and the eval harness this site’s own agent layer runs — extracted and published as a GitHub template repo, so it’s a thing you can fork rather than a paragraph you have to take on faith.
(opens in a new tab)