Three Numbers Should Define Connecticut’s AI Workforce Push

by Gleb Tsipursky

Connecticut has committed itself to teaching people how to use artificial intelligence. The public deserves a way to tell whether that education leads to better work.

On October 28, state leaders highlighted the Tech Talent Accelerator’s emphasis on AI. Its implementation partner reports sixteen innovation projects at fourteen institutions, including programs in healthcare, cybersecurity and accounting. Those are sensible choices. The next announcement should lead with results rather than the number of people who attended training.

Three numbers would make that possible.

First, report the percentage of participants who demonstrate reliable independent judgment. Give trainees a second exercise in which a confident AI answer contains a wrong assumption, incomplete evidence or a misplaced recommendation. Ask them to find the problem, support a correction and explain where a human decision remains necessary. Publish the pass rate. Anyone can watch a demonstration; competence appears when the demonstration fails.

Second, report the percentage of graduates who obtain paid, supervised experience using the skills. Employers should help shape the assessment, because the difference between a polished exercise and a workplace task can be enormous. A student learning AI-assisted financial analysis should understand why an apparently accurate forecast may rest on the wrong input data. A future healthcare administrator should know which AI-generated summary needs verification before affecting patient services. These are illustrative scenarios, not claims about Connecticut programs.

Third, report the percentage of trainees who move into better-paid work six months after completion. Break this number down by prior education, age and occupation. Connecticut should know whether new opportunities reach incumbent workers and people outside traditional technology careers.

There is reason to ask for that equity check. Gallup’s October 6 workplace survey found that 40 percent of college graduates used AI daily or weekly, compared with 17 percent of employees without a college degree. Users reported gains in speed, but those self-assessments cannot establish measured increases in productivity.

The stakes are high for early-career workers. ADP Research and Stanford’s October 21 update showed a 3.7 percent year-over-year employment contraction for workers ages 22 to 25 in highly AI-exposed occupations in September. This comparison does not establish AI as the cause. It does make an exclusive focus on entry-level software operation a risky bet.

Connecticut already has institutions and employer relationships to test a stronger model. Each partner could publish a quarterly scorecard with these three numbers, the sample size and transparent definitions. Programs too small for reliable standalone comparisons could combine cohorts instead of producing misleading rankings.

At the same time, the state should ask graduates what they actually do with their skills after they leave class. Can they challenge bad outputs? Can they explain decisions to a supervisor? Did they have an opportunity to practice at work? Those responses provide context that a completion count cannot.

Training matters when it changes what people can responsibly do. Connecticut’s AI workforce initiative will be easier to defend, improve and expand when residents can see evidence that it did.

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