we needed a SaaS engineer, not an AI engineer
We spent months looking for an “AI production engineer.” Someone to take our research prototypes (topic modeling with BERTopic/HDBSCAN, embedding-based clustering, misconception mapping with fine-tuned models) and make them production-ready. AI was our differentiator, so the role felt urgent.
Then we did an honest assessment:
| Area | Status |
|---|---|
| SaaS infrastructure | Manual billing, no customer lifecycle, no self-serve |
| AI/Product | 80% “good enough”, orphaned projects everywhere |
| QA | Stable, but engineers doing most testing |
| Design | New brand definition in progress |
The AI was 80% there. Not perfect, but shipping. Users were getting value. Topic modeling worked. Chat analysis worked. Transcription pipeline worked.
What wasn’t working: billing was manual. Customer onboarding was a spreadsheet. Support was Slack DMs. No lifecycle emails, no usage tracking, no churn indicators. We were building a B2B2B platform (Dembrane sells to consultancies who sell to municipalities) and had zero infrastructure for that sales motion.
The gap wasn’t AI capability. It was SaaS infrastructure.
So we pivoted the hire. Instead of an AI production engineer, we started looking for a SaaS engineer. Someone who’s built billing integrations (Stripe ideally), customer lifecycle systems, support tooling, and multi-tenant architectures.
Hard call because it felt like giving up on the thing that makes us special. But the AI work had a natural owner already. Jorim was leading it, with a data team member supporting. What nobody owned was the boring-but-critical plumbing that turns a product into a business.
The multi-workspace feature we’re building (P1 in our current sprint) is a good example. URL structure: app.dembrane.com/{workspace_slug}/projects/{project_id}. Users can belong to multiple workspaces. Sounds like a product feature but it’s actually SaaS infrastructure. It’s the foundation for billing per workspace, access control, and the B2B2B handoff flow where a consultancy manages multiple municipal clients.
Interview process changed too. Instead of asking about embedding dimensions and fine-tuning strategies:
- How they’d structure a multi-tenant billing system
- Experience with Stripe webhooks and subscription lifecycle
- How they think about customer onboarding flows
- Approach to support tooling and self-serve
Hire for your actual bottleneck, not your identity. We thought of ourselves as “an AI company” so we hired for AI. But we were actually a SaaS company with AI features, and the SaaS part was falling behind.