Agentic AI Engineer

Also known as

The market uses several titles for the same role:

How it changes by seniority

• Junior — Implements agents with tools and prompts against a given design.
• Mid — Designs evaluation and guardrails, and ships agents to production.
• Senior — Owns reliability and cost of critical agents; sets design patterns.
• Lead — Leads the agent team and the platform they’re built on.

Interview questions

1. 1. You’re building an agent with access to the CRM and email. What limits do you set from the design?

   What to look for: Least-privilege per task, what it must NOT do, and brakes where it can do damage; security by design.

2. 2. How do you choose the model for each step of the agent?

   What to look for: Model per task (cost vs. reliability), not the biggest for everything.

3. 3. How do you know your agent is right before putting it in front of a customer?

   What to look for: Builds evals with real cases and a threshold; doesn’t trust "it works in my tests".

4. 4. Your agent has to call tools and sometimes picks the wrong one. How do you control it?

   What to look for: Scaffolding with error handling, retries and clear autonomy limits.

Scorecard

• Agent design — Defines goal, tools, data and limits; knows what it must NOT be able to do.
• Model choice — Tunes model per task by cost and reliability, not defaulting to the largest.
• Evaluation — Measures accuracy with real evals before production and on every change.
• Security by design — Least-privilege and defence against misuse and injection, not bolted on at the end.

Red flags

• — Gives the agent broad access "so it can do anything" with no limits.
• — Ships with no way to measure whether it’s right.
• — Uses the most expensive model for everything and ignores cost per run.