systemix docs

The meta-loop

Systemix is the context engine for your design system — the one place decisions, tokens, and evidence are written down, so every person and every agent works from the same source. A design system has always been the connective tissue between the people who build a product; this one is AI-native, and it improves itself on a loop.

1. Hypothesisqueue.jsonthe next bet, framedBizMktDesign2. Buildexperiments/<id>.mdxprototype it, togetherEngDesignBiz3. MeasurePostHog evidencereal numbers, pulledMktBizDesign4. Evaluateevidence + thresholdsdoes it clear the bar?MktAIDesignEngBiz5. IdeateLEARNINGS.mdwhat we learned, what's nextBizDesignMkt6. DocumentLEARNINGS.md · contractsynthesized, written backEngAIDesignyou approveyou approvethe notebook seeds the next idea
Biz BusinessDesign DesignEng EngineeringMkt MarketingAI AI agents

Six steps, five owners

The meta-loop is how the system learns. Each step is owned by the people (and agents) closest to it:

  1. Hypothesis — the next bet, framed. Business · Marketing · Design.
  2. Build — prototype it, together. Engineering · Design · Business.
  3. Measure — real numbers, pulled. Marketing · Business · Design.
  4. Evaluate — does it clear the bar? everyone, including the agents.
  5. Ideate — what we learned, what's next. Business · Design · Marketing.
  6. Document — synthesized, written back. Engineering · AI agents · Design.

Then it feeds its own next hypothesis. Two steps wait for a human: you approve a bet before it's built, and again before a decision is recorded. Everything the loop wants from you arrives as a card in .systemix/queue.json.

Five personas, one loop

Each persona has its own job, its own operational loop, its own setup, and its own data-flow — and each feeds the shared meta-loop:

  • Business frames the bets and makes the calls, from receipts.
  • Design holds the rationale and catches the drift.
  • Engineering builds the thing and keeps the code true to the system.
  • Marketing runs the experiments and reads the numbers.
  • AI agents are a first-class participant — they read the context, act on it, and write evidence back, with a human closing every decision that matters.

The operational loop

The 6-step meta-loop is the whole picture. The operational loop — Propose → Build → Measure → Learn — is one turn of it, and it's the part that runs on a daily cron today. Start there for the mechanics of a single experiment.

On this page