Open · Remote
AI Sales Engineer
Remote (US) · Remote
Owns the AI sales pipeline as product. Tunes the Anthropic prompts behind /api/demo-request, /api/quote, /api/cutover-plan so the AI's output reads like Doug wrote it. Every prompt drift, every misfire, every 'sounds robotic' is your call to fix.
Compensation
Base $140-180k DOE + equity track for foundational hires. Bonus on AI-quality metrics (eval-set accuracy + voice fidelity) — paid on shipped improvements, not on customer count.
Why this role exists
We're betting that AI can run the cannabis-SaaS sales motion well enough that 1 Industry Connector + 1 AI Sales Engineer + 1 Cutover Lead can do what Dutchie hires 12 people for. You're the lever that makes that math real. Foundational hire — equity reflects the bet.
What you’d do
- Tune Anthropic prompts so AI-generated proposals + cutover plans + demo scripts read like a Doug-voiced operator-to-operator email, not a SaaS chatbot
- Watch every AI-generated proposal in the first 60 days. Flag the ones that drift. Retune. Ship the prompt update with the changelog entry naming what shifted.
- Build the eval harness — golden-set of 50 prospect inquiries scored on fit + accuracy + voice. Run it on every prompt change. Block ships that regress the score.
- Own the demo deploy — reseed cadence, persona quality, walkthrough script accuracy, what each persona reveals. The demo IS the sales pitch.
- Build the 'explain-to-an-operator' surfaces — ROI calculator, migration playbook, cutover checklist — all rendered server-side from the AI's structured output, not chat-style
What we’re looking for
- 5+ years shipping production code. At least 1 of those years building LLM-powered features that are NOT chatbots
- Strong taste for voice — you can read a generated paragraph and tell whether it sounds like a real human or a SaaS-speak template
- Comfortable owning the prompt-engineering, eval, AND production code paths. This isn't a 'prompt engineer' role; you ship the surrounding infrastructure too
- Direct-style communication. You can defend a prompt change in front of Doug without hiding behind 'the model said.'
- Preference for systems where the AI's output is the product (not a feature inside a larger product)
What we’re NOT looking for
- Engineers who treat AI as 'just another API.' This role is the AI/UX boundary; the boundary IS the product
- Prompt engineers without production-shipping muscle. We need eval harnesses + monitoring + deploys, not just .md files
- Folks who want to optimize one model service forever. Each engagement is a fresh prompt-set
First 90 days
- Day 1-14: read every AI-generated proposal in the v0.21+ /api/demo-request log. Identify the 5 most common voice + accuracy drifts. Write the eval harness.
- Day 15-45: ship the first 3 prompt-tuning rounds against the golden set. Score each ship. First measurable lift in eval-set accuracy by day 45.
- Day 46-90: extend the AI surface to /api/quote + /api/cutover-plan. Reseed the demo deploy on a 30-day cadence so personas don't go stale.
- By day 90: AI-generated proposals score 85%+ on the eval set, voice-drift caught in pre-ship not post-ship, demo deploy refreshes automatically.
Apply directly to Doug.
No portal, no recruiting funnel. One email.
careers@cannagent.ai →