YOUR OBJECTIVE COMES FIRST
Distinguish locating guidance from interpreting authority.
Start with a public objective and a reviewable use of AI.
Maryland Commerce highlights life sciences, cybersecurity and IT, advanced manufacturing, logistics and artificial intelligence among its key industries. Use sector context to ask employers and program partners better questions, rather than assume every organization has the same need. An employer discussion about research-support project intake can examine a project brief, approved source collection and access request and identify the people who own the work. Public objectives, approval authority and the agreed service scope determine what follows.
Maryland Department of Commerce: industry context ↗01Choose the task and its public purpose
Describe the administrative work, intended users and the decision the proposed tool supports. Distinguish drafting, search or intake assistance from an authoritative determination. The agency should establish the permissible use and accountable owner before implementation.
02Define information and access boundaries
Identify approved sources, sensitive records and the environments in which tools may operate. Confirm access and review requirements with agency owners. A proposed connection must fit the supported interfaces and procurement or security process rather than assume unrestricted data access. Working example: Approved source register and retrieval test set.
03Train staff on the work they will perform
Build practice around representative tasks, insufficient evidence and incorrect outputs. Explain source checking, appropriate information use and escalation. Training should distinguish a fluent answer from an answer supported by the approved material. For research-support project intake, agree how project leads, information owners and technical support contribute where those functions are involved.
04Plan a controlled introduction and evaluation
Agree a first user group, acceptance evidence, support owner and change process. Review output quality and staff effort, including corrections. Wider adoption should follow agency review rather than depend on demonstration quality alone. Review check: Review source support and unanswerable cases.