Most mid-sized companies now have AI running somewhere. A finance analyst uses a copilot. Customer service tests automated case summaries. Operations is trying a workflow tool. Each effort makes sense alone. The trouble starts when nobody owns how they fit together.
That gap is where a new kind of hire is showing up: the person who designs how AI agents work inside your systems, not just whether one tool performs well in a demo. The title varies. Some employers call it an AI architect, some fold it into solutions architecture or an automation lead. The job matters more than the label.
What the work actually is
An AI agent completes a task, or a string of tasks, within limits you set. It reads from a system, takes a step, and hands off to a person or to another agent. Designing that well means deciding:
- Which workflows are worth automating, and which are not
- What data and systems each agent can touch
- Where a human must review, approve or take over
- What gets logged so you can audit a decision later
- How you will measure whether it saved time or reduced errors
None of that is model building. It is closer to systems design with a strong dose of process knowledge and risk judgment.
Signs you need this capability
You do not need a dedicated hire for every AI project. You probably do when:
- Several departments are buying or building AI tools on their own
- A pilot worked but nobody can say how to put it into production
- More than one agent has to pass work to another
- Your security, privacy or audit people are asking questions IT cannot answer yet
- Leadership has promised results from AI and no one is accountable for delivering them
If only one of those is true, a strong senior developer or analyst with clear scope may be enough. If three or more are true, you have an ownership problem, not a tooling problem.
Why pilots stall
The pattern is familiar. A pilot runs on clean sample data with a keen team. Then it meets the real environment: inconsistent records, older systems with no clean integration, staff who do not trust the output, and no agreed rule on who is responsible when it gets something wrong.
The model is rarely the reason it stalls. Data access, workflow fit, governance and unclear value are. Those are design problems, and they are cheaper to settle before build than after.
How this differs from your existing technical roles
A technical or enterprise architect looks after the overall environment: cloud, data platforms, integrations, security. The agent-focused designer works inside that environment and decides how agents behave across workflows. In a smaller company one person may cover both. In a larger one they sit side by side.
Do not expect a single candidate to be strong everywhere. Decide which half you are weaker in and hire toward it.
Where it shows up in the six areas we recruit for
Operations. Reporting, coordination and exception handling. The design question is what an agent can resolve alone and what goes to a supervisor.
Finance. Invoice processing, reconciliation and variance analysis. Controllers should be in the room, since approval limits and audit trails are theirs to define.
Manufacturing. Purchase orders, supplier follow-up, shipment status and maintenance scheduling. Plant systems are often old, so integration experience counts for more than AI buzzwords.
Sales. Lead routing, quote preparation and CRM upkeep. The risk is bad data going to customers, so review steps matter.
Contract and project work. Many employers bring in a short-term specialist to scope and stand up the first production use case, then hand it to an internal owner.
AI and data. Engineers and analysts who build the pieces need a clear design to build against.
What to look for when hiring
The market for this exact title is thin, so screen for the underlying skills:
- Has shipped something into a real business workflow, not only prototypes
- Can explain a failure and what they changed afterward
- Understands your data sources and the integration work involved
- Talks comfortably about access controls, privacy and audit
- Can sit with a controller or plant manager and speak in their terms
- Defines success in business measures such as cycle time, error rate or cost per transaction
Be wary of candidates who describe only tools, or who promise full automation. Good designers are specific about where a person stays in the loop.
Employee or contractor
If you have one use case and a defined end date, a contract specialist is often the sensible route. If AI is becoming part of how several teams work, you want someone permanent who will live with the consequences of the design. Many employers do both: a contractor to get the first build right, then a permanent hire to own it.
What to do next
List every AI effort currently underway in your company, name who owns each one, and mark where they overlap or depend on the same data. The gaps in that list tell you what role to hire for. If you want help defining that role or finding people who have done it in production, a Traverse recruiter can help.