Oklahoma's AI workforce plan needs a front-desk test

Oklahoma is spending considerable energy preparing its government workforce for artificial intelligence. A new project will study how AI changes state jobs, while the state has already rolled out a shared platform for public employees. But an important question should be put to a citizen in Tahlequah, Muskogee or another northeastern Oklahoma community: will a person still be able to solve the problem when an automated answer is wrong?

That question belongs at the center of the workforce conversation. Technology can summarize regulations, draft letters and speed up routine paperwork. Residents, however, judge a public service by whether they receive a correct answer, a timely decision and an understandable way to appeal a mistake. New software cannot substitute for that responsibility.

In October, Holly and 17A announced a $1.1 million project, supported by national innovation funders, to help Oklahoma and Virginia assess how AI could change government job classifications, required skills, hiring and compensation. The project will run for 18 months and produce methods other states can use. It is a chance to study work as it actually happens instead of assuming that every task an algorithm can perform represents a position that can disappear.

Meanwhile, Oklahoma's July launch of BEACON established a common AI platform for state agencies. Its initial applications focus on modernizing administrative rules and helping public workers navigate procurement policies. The state describes the tools as supporting employees and keeping human oversight in place. That distinction matters in northeastern Oklahoma, where residents may need to sort through multiple offices, forms and rules just to obtain one service.

Suppose an AI assistant helps a clerk explain which form a family needs. The first answer might take seconds. If the system gives outdated directions, however, the family could make a wasted trip, miss a deadline or submit the wrong documents. A useful evaluation would count the entire process, including calls, corrected forms, rework and the employee time required to fix the error.

Public agencies should begin with a sample of anonymized real cases, including straightforward questions, incomplete applications and disputes. Have a supervised AI system prepare a response. Compare that with the existing process. Track accuracy, total processing time, escalations and whether the person receiving the service can get a human explanation. Publish the method and aggregate results. This would let the public distinguish a demonstration from improvement.

The workforce side deserves equal attention. In my consulting work on AI adoption, one recurring problem is that managers remove the simple assignments through which less-experienced employees learn. An employee who has never worked through routine records may struggle when the exceptional case arrives. Any state AI training program should therefore include supervised practice, not just short tutorials on composing prompts.

Recent private-sector evidence underscores the stakes, without pretending government service works exactly like a corporate call center. Gartner reported on November 5 that 21% of 275 surveyed service leaders had reduced human-agent headcount because of AI, while 7% had already rehired employees after earlier cuts. That finding is a reason to insist on evidence before reducing the people who know how to resolve exceptions.

Oklahoma already has a useful principle in its published AI policy: employees must verify AI-generated content and retain responsibility for final work. Local officials and employers should make that principle observable. Ask who has authority to correct an automated answer, what triggers escalation, and whether the organization records recurring mistakes rather than hiding them.

An effective AI workforce plan would help employees handle more complex work and give residents better service. Northeastern Oklahoma can help define success by asking for reliable answers, clear human responsibility and opportunities for newer workers to develop judgment. Those outcomes are more valuable than counting how many forms a computer can draft.

Gleb Tsipursky, PhD, a behavioral scientist, CEO of Disaster Avoidance Experts, and author of The Psychology of AI Adoption at Work: From Resistance to Results (Georgetown University Press, 2026). https://disasteravoidanceexperts.com/aibook