José Ochoa

José Ochoa · Product & web design · Data visualization

I design systems that stay coherent at AI speed.

On 1 November 2026, a bakery ordering system I designed went live at Community Food Co-op, built by AI agents I directed and checked by 6,089 automated assertions. It sits on the co-op’s design system, which I keep in parity across theme.json, CSS and Figma so an agent can’t hardcode a value without it surfacing.

Read the bakery case study How I work

Figure 1 · Commits per day, all branches

142 days from first commit to a live store

Daily commits on the bakery plugin, 12 June to 5 November 2026, with milestones. October alone holds 1,174 of 1,503.

The tallest bar is one night: 18 deli tickets built by an orchestrator and parallel builders, 307 commits dated 26 October.
Source: git log --all on joseocfc/cfc-bakery-orders, computed 2026-11-07 Data

1

Designer, 0 developers

One live store since 1 November 2026, for bakery and Thanksgiving deli orders.

git log --all: 1 author · STATUS.md:161

6,089

Automated assertions

In a full test run, with 1 known red that every report names.

tests/cli/README.md · 2026-11-05

1,594

Assumptions written down

Numbered A1 to A2121, each with a status: signed off, superseded or waiting on me.

docs/assumptions-register.md · 2026-11-05

94

Design tokens

1 source of truth, mirrored to Figma, with 0 unexplained deltas either way.

figma-parity-check.py · last run 2026-08-21

Selected work

Two systems in production and one study for Epoch

The Figma button component set: six variants, three sizes and five states in a grid, plus a loading state, with placement rules above.

Case study · Design system · 2026

One system in both directions

theme.json, CSS and Figma held in parity by a drift checker and a hook that stops agents hardcoding values.

  • Design systems
  • Tokens
  • Figma variables
Scatter chart of training compute for notable AI models by publication date, on a log scale. Scatter chart of training compute for notable AI models by publication date, on a log scale.

Study · Epoch AI data · 2026

Notable AI models, redrawn for a phone

Epoch’s CC-BY data rebuilt as a chart that reflows at 390 px instead of cropping.

  • Data visualization
  • D3
  • matplotlib

Two smaller builds, same method

Project Net

A community map of tension nets in Bellingham, with location privacy built into the data model.

  • Product design
  • Maps
  • Next.js

How I work

I specify, agents build, and nothing ships without proof

I don’t write the code. I specify it, direct the agents that do, and prove it works.

  1. Specify

    Every open question gets decided one at a time, with a recommended answer. Anything an agent has to assume becomes a numbered row before it commits.

  2. Direct

    An orchestrator splits the work into lanes, each builder works on its own branch, and a fresh reviewer scores every risky change. Below 9/10, it goes back.

  3. Prove

    Test output and screenshots, never summaries. Then I walk staging on my phone, and production moves only on my word.

The method, with the evidence

About

I work from Bellingham, Washington, on Pacific time

I trained as an industrial designer at Tec de Monterrey, then led brand and web design for startups in Mexico and the Bay Area. Since March 2025 I’ve been the web designer at Community Food Co-op.

More about me