Your first ML engineer is not a research hire. It's a shipping hire. Most startups get this backwards and pay for it in a year of runway and a system nobody can run.
The first ML engineer sets the bar for whether AI becomes a capability or a science fair at your company. The hire is high-leverage, and the most common mistake is optimizing for the wrong signal.
Seek to Understand: What the Role Actually Is
Understand the job before the JD. At a startup, the first ML engineer builds pipelines, owns evals, ships models to production, and debugs the integration — not trains SOTA architectures. If you hire a researcher for an engineer's job, you get a system that demos and doesn't run.
- Ships pipelines, not papers.
- Owns evals and observability, not just the model.
- Cares about the integration, not just the benchmark.
The Diligence Signals
- Have they taken a model to production and kept it running?
- Can they explain a production failure they owned, end to end?
- Do they talk about evals, drift, and cost — or only about model quality?
The Red Flags
- Optimizes on the benchmark, not the business metric.
- Treats production, evals, and observability as someone else's job.
- Can't name a model they shipped that failed and what they learned.
The Hiring Trap
The trap is hiring for prestige — the lab name, the paper count — instead of for shipping. A first ML engineer who can't own production leaves you with a model and no system, which is exactly where most stalled AI initiatives live.
How FACTA Frames It
FACTA's forward-deployed and startups work is built on engineers who ship and own — one real workflow in production in 4-8 weeks. The hiring bar we'd apply to a first ML engineer is the same one we apply to our own: ships, owns, and can defend the system in front of the board.
Conclusion
Your first ML engineer is a shipping hire. Screen for production ownership, evals, and integration — and treat pure research credentials as a yellow flag, not a green one.
About FACTA
FACTA helps startups and growth-stage teams turn AI into production systems that keep running — not demos that impress once.
We design the architecture around the parts that actually break under real usage: tooling you own, credentials you control, failover, cost controls, observability. The boring infrastructure that keeps a system alive after launch.
Led by Matías Baglieri and Carolina Fogliato, we focus on one thing:
AI leadership that builds. Not just advises.
Send us a candidate profile or your first-ML-engineer JD.
We'll tell you whether it's screened for shipping or for papers. See the fractional CAIO model for the leadership around the hire.
Explore AI Strategy