Establishing the Colorado Applied AI Workforce Initiative
The use of AI in the workplace is expanding and outpacing access to applied training, which remains inconsistent or unavailable altogether. Jobseekers, employees at smaller organizations, and local public workers need more than access to a model or a vendor course. They need practice choosing useful tasks, protecting data, testing outputs, and explaining where human judgment remains necessary. The Colorado Applied AI Workforce Initiative would provide 1,000 participants in one metropolitan and one rural region training through real or realistic workplace tasks. Each learner in the two-region pilot would build a bounded agent and receive a portable record of independently assessed skills. The result would be a broader pool of workers who can use AI responsibly, smaller employers better able to adopt useful tools, and public agencies better prepared to improve services without giving up accountability.
Challenge and Opportunity
In Q2 2026, Gallup reported that 52 percent of U.S. employees used AI in their role. That adoption creates an opportunity to improve efficiency, but it also creates risk when employee access to training, approved workflows, and review controls is either inconsistent or not available at all.
Access to a large language model is not evidence of job-ready skill. A county employee, for example, might use a bounded agent to retrieve policy, draft a sourced resident notice, flag missing facts, and stop for approval. The employee still has to decide whether AI belongs in the task, protect sensitive information, test difficult cases, correct errors, and explain the decision. That is the opportunity and the gap: people can use AI to handle routine work, but they need training to decide when it is appropriate and remain accountable for the result.
Among Colorado employer businesses that reported using AI in a business function, only an estimated 15.1 percent had trained current staff in the previous six months. While Colorado’s Office of Information Technology (OIT) has published AI guidance and training, it is limited to state employees and contractors. Colorado workforce centers already provide jobseekers with computer and internet access and training, while the state’s Digital Navigator pilot demonstrated how libraries and rural workforce centers can be used to address device, connectivity, and digital-skills barriers. The Colorado Applied AI Workforce Initiative would fill this gap and create a common applied standard for local workers.
Colorado must act before unequal access becomes a durable workforce divide. In the San Luis Valley, 15.5 percent of households lacked home broadband service, so applied training must include accessible, low-bandwidth, and in-person options. Workforce centers, colleges, libraries, and local partners can bring training closer to residents. Connecticut’s Tech Talent Accelerator shows how colleges and employers can develop applied AI programs, while New Jersey’s emerging library network shows how public libraries can serve as community access points. Colorado can adapt both structures while testing one common standard across metropolitan and rural conditions.
The need is immediate; as of January 1, 2027, SB 26-189 will require developers and deployers of automated decision-making technology to provide documentation for how these tools are used. Developers must provide technical documentation describing the tool’s intended uses, categories of training data, known limitations, and instructions for appropriate use and human review, and deployers will be responsible for providing consumers with a description of the tool’s role within a consequential decision. Yet, this legislation creates no corresponding program to train employees to do this work.
Plan of Action
Recommendation 1. Establish the initiative and make an individually built AI agent the core training artifact.
The next Colorado Governor should establish the Colorado Applied AI Workforce Initiative and direct the Colorado Department of Labor and Employment (CDLE) to develop the applied AI training standard, eligibility requirements, records protections, and evaluation process within 90 days. Eligible training participants should include Workforce Innovation and Opportunity Act (WIOA) adults, youth, and dislocated workers, along with other jobseekers, workers at organizations with fewer than 250 employees and no comparable training, and local government workers.
During the 90-day design period, CDLE would collaborate with the Office of the Future of Work to bring together workers, small employers, local governments, workforce centers, colleges, libraries, rural partners, and disability-access experts. It would use that group to identify real workplace tasks, surface access barriers, and recruit regional partners. CDLE would retain control of the standard, funding, procurement, provider selection, records, compliance, and evaluation.
The training should cover six competencies needed to apply AI at work:
- Task selection: Define the user, workflow, baseline, result, and whether the task calls for no AI, a hosted model, or a managed open-weight model.
- Data protection: Classify information, restrict sources and tool permissions, and keep protected data out of the agent.
- Output verification: Test held-out cases, source grounding, repeated runs, abstention, and escalation.
- Error and bias recognition: Find security failures, factual errors, and counterfactual disparities, then correct and retest them.
- Documentation and human review: Record choices, sources, cost, corrections, residual risks, and the human approval point.
- Supervised workplace application: Build and defend a bounded AI agent that performs a real or realistic workplace task under independent assessment.
Each learner would build an AI assistant for one clearly defined workplace task. The provider would supply a secure workspace, approved source materials, and up to three managed tools. The assistant could retrieve information or simulate an action, but it could not change live records or make a decision on its own. The provider would set limits on tool calls and run time. The learner would test the result, explain the design, and obtain human approval before use. Examples could include a hosted service such as ChatGPT or Claude and a provider-managed open-weight model such as Qwen3 running through Ollama.
The training should consist of six weekly sessions totaling 24 contact hours: 12 hours of instruction, eight hours of guided construction and testing, and four hours of testing, demonstration, defense, and feedback. Sessions would progress from the problem, baseline, data, and deployment approach through configuration, security, reliability, bias, and outcome testing. Learners would be required to submit the artifact package before the final session.
Possible projects include a resident-notice agent based on county policy, a safety-procedure agent using approved manuals, or an occupation-research agent using workforce sources. Regardless of the task, every artifact package should contain five parts:
- The problem statement, affected user, baseline workflow, success measures, and a no-AI alternative (e.g., traditional automation).
- A working bounded agent with approved sources, limited tools, stop conditions, logs, and a human approval point.
- A comparison of a hosted proprietary environment, a hosted or provider-operated open-weight environment, and the no-AI baseline using the same cases.
- Functional, security, and fairness tests showing results, failures, corrections, and retests.
- Measured results and a live demonstration.
To pass, a learner must demonstrate all six competencies in a working artifact. An independent assessor would score the work during a live demonstration and brief oral defense. Providers would give managed access for one revision within five business days, without requiring a personal device, account, or home connection. The portable record would report pass or fail for each competency.
Recommendation 2. Run a two-region pilot and test whether the training standard works under different access and labor-market conditions.
CDLE should launch a pilot to test the training program over 12 months. The pilot should enroll 1,000 people across one metropolitan and one rural region, with at least 300 rural seats. Training should be run with cohorts of 20 to 25 participants, and should be delivered in-person or through an instructor-led hybrid format. Workforce centers, colleges, local governments, libraries, adult educators, and schools should serve as access points. CDLE should select providers based on population reach, task validation, instructional and assessment capacity, accessibility, connectivity, safeguards, and scheduling, without regard to vendor affiliation.
Both regions would use the same case material, rubric, and independent scoring process. Delivery must meet applicable Colorado WCAG 2.1 A and AA requirements and provide low-bandwidth options, assistive technology, accessible documents, and equivalent tool paths.
For procurement, CDLE would compare proprietary hosted, open-weight hosted, and provider-operated configurations under the same performance, security, and accessibility requirements, maintaining a tool-agnostic approach. The state should own publicly financed core training components or hold perpetual rights to reuse and adapt the training, excluding third-party intellectual property. Before pilot launch, CDLE must identify providers with a compliant design within their budget. Cost constraints should not weaken the training competencies, accessibility, security testing, or independent assessment.
Prior to enrollment, CDLE must define evaluation measures for the pilot. These measures should include employment status, wages, career level, opportunity to apply AI in the workplace, and workplace use. Wage and advancement results, measured through changes in career level, would be descriptive rather than continuation conditions. An opportunity must include paid or supervised work, approved tool access, and a pre-mapped task. Documented use requires employer confirmation or evidence that the participant applied a pilot competency to a real task in the workplace. An independent evaluator would collect these measures at enrollment and 120 days after a participant passes every competency, then report them statewide and by region.
Training expansion would require at least 75 percent completion among all enrollees statewide and in each region, usable 120-day follow-up for at least 70 percent of verified completers, an opportunity to apply AI for at least 50 percent of verified completers, and documented workplace use by at least 60 percent of opportunity-qualified completers. Only completers with usable follow-up may count as opportunity-qualified, and nonresponse would not count toward any numerator. An unresolved material safety or privacy incident would independently block expansion.
Before the final evaluation, CDLE would review the first 200 final artifact deadlines, including at least 60 learners in each region. New enrollment should pause if fewer than 150 participants have passed, either region falls below 75 percent completion, or projected spending exceeds the $2.8 million ceiling.
The pilot would have a $2.8 million spending ceiling. This proposal would cap any Workforce Innovation and Opportunity Act Governor’s Reserve contribution at $1.5 million, subject to available funds, participant and cost eligibility, lawful procurement, and required reporting. Signed nonfederal commitments would cover the balance.
Recommendation 3. Assign post-pilot training ownership if final evidence supports continuation.
If the pilot meets every continuation condition in Recommendation 2, the Colorado General Assembly should decide whether to authorize and fund statewide expansion. CDLE should lead the pilot unless later law assigns the initiative to another agency. HB 26-1317 directs its transition committee to recommend how specified workforce functions could move to the Department of Higher Education, but the act does not itself transfer CDLE programs or WIOA duties. Any continuation law should name the permanent owner. The designated agency should maintain the training standard, curriculum, rubric, records, model approvals, task and test banks, scaffold, evaluation measures, provider standards, provider roster, reporting, and evaluation. State-agency use would remain subject to Office of Information Technology review.
Timing matters because HB 22-1350’s Opportunity Now is scheduled to be repealed on July 1, 2028. If lawmakers extend it, that legislation could authorize continuation of this initiative and designate its permanent owner. Expansion funding should remain unavailable until final pilot results meet every continuation condition.
Conclusion
Colorado has a chance to shape an AI-enabled workforce before access to applied training becomes another source of inequality. A focused pilot would help workers build evidence they can carry across employers, help smaller businesses adopt useful tools without bearing the full cost alone, and help public agencies improve routine work while keeping human accountability.
It would also give the state a controlled way to learn what works in metropolitan and rural communities. If the evidence is strong, Colorado can expand a standard that workers, employers, and agencies can trust. If it is not, the state can stop before committing to a larger system.
Vendor courses are useful, but Colorado needs a common applied standard. The pilot requires a working agent, documented testing, and independent results that travel across tools and employers.
The 24 hours are structured contact time. A managed scaffold and limited tools keep the project narrow. Learners submit work before the final session, demonstrate it live, and may revise once after independent scoring.
No. Learners compare hosted, open-weight, and no-AI approaches on the same task. Choosing conventional methods is sound judgment when they perform better on accuracy, cost, accessibility, or risk.
At least 300 seats serve the rural region. Local access points provide devices, low-bandwidth delivery, assistive technology, and accessible tool paths. Participants need no personal device, vendor account, or home connection.
Expansion requires regional completion, sufficient follow-up, verified workplace opportunities, and documented use. Withdrawals and nonresponse remain visible. Any unresolved material safety or privacy incident blocks expansion.
The amount is a pilot spending ceiling, not tuition. WIOA may contribute up to $1.5 million for eligible costs. Nonfederal commitments cover the balance, and wages remain separate. The state retains rights to core materials, while legislation assigns a permanent owner only if the evidence supports continuation.
Jobseekers, employees at smaller organizations, and local public workers need more than a weekend crash course in AI. They need practice choosing useful tasks, protecting data, testing outputs, and explaining where human judgment remains necessary.
“What excites me is that it’s very tempting to be very discouraged, and say, ‘Oh, we’ve got these archaic institutions that are calcified and you could never change them.’ But I think we’re in the middle of a technological revolution that will upend lots of things, and does provide a window.”
As Congress considers broader packages to advance critical minerals production and supply chain resilience, science diplomacy vehicles must be part of that conversation, not as an afterthought, but as an intentional and foundational pillar of any strategy.
“Structured partnerships with our allies on critical minerals innovation can ensure that the best science and the best talent are working together towards shared security and economic prosperity.”