Four roles. Four people.

Founding Engineer

You will own at least one packages/* directory end-to-end. You are the person who decides whether the tool registry, the credit meter, or the case engine lives or dies this quarter.

• Ship the Chrome extension v1 (autofill, not submit — CICC red line).
• Own the AI router: CA-default, US-opt-in, split-policy on fail.
• Live in Drizzle + Supabase RLS all day; write integration tests that assert cross-tenant denial.
• Pair with the founder on hard architecture calls. Disagree and commit when you have to.

Not ideal for · You want a spec handed to you. You treat RLS as someone else's problem. You think 1M-row migrations are rare.

Founding Designer

You are the hand on every pixel of DESIGN.md and the person who notices when a pattern slides into slop. You draw, you write, you ship.

• Own the CRM UI (linear-dense) and the marketing site (editorial-spacious).
• Design the DRAFT stamp, the PII scrubber preview, the audit-log viewer.
• Write copy as well as you draw; retire anything that reads like "AI slop".
• Run visual QA after every ship — before the CEO notices.

Not ideal for · You need a design system handed to you. You ship figma files, not PRs. You believe "spec" and "hand-off" are the same thing.

GTM Lead

Our customers are RCICs and their paralegals. You learn the CICC code, the IRCC portal quirks, the WhatsApp-first reality of their day — and you sell against that, not against a deck.

• Run the first 100 firm conversations yourself. Don't hire SDRs until you have beat the number solo.
• Own the content studio pipeline: IRCC news → a short, practitioner-grade write-up a day.
• Partner with CICC-adjacent communities without ever implying endorsement.
• Carry a demo. Run the CRS calc on Zoom. Know the numbers cold.

Not ideal for · You want a marketing budget. You think the category is "Canadian SaaS". You are allergic to talking to customers.

Applied AI Engineer

You will write the evals, tune the router, and make sure the SOP Generator doesn't invent a UBC faculty. You carry the eval suite in your head.

• Own the golden-output test harness (pnpm test:evals).
• Ship per-tool eval suites before the tool itself ships (eval-first).
• Tune prompt caching, cost per run, and the split-policy fallback.
• Work with the founding engineer on the MCP server and Claude-as-judge.

Not ideal for · You want to build a chatbot. You measure progress in model release days. You think "hallucination" is someone else's problem.