Every year, the assessment and credentialing industry gathers at the ATP Innovations in Testing Conference and talks about the future. This year in New Orleans, something felt different.

The conversation had shifted. Not from pessimism to optimism. Not from caution to excitement. But from “what if” to “what now.”

AI was no longer the topic on the agenda. It was the assumption underneath every other topic. And that distinction, subtle as it sounds, changes everything about how credentialing organizations need to think, plan, and build.

The Pilot Era Is Over 

For most of the past decade, AI in assessment lived in a familiar place: the innovation lab. Promising. Exciting. Perpetually “almost ready.”

Organizations ran pilots. Published white papers. Debated ethics. And then, cautiously, ran more pilots. That era is closing fast. 

According to Deloitte’s 2026 State of AI in the Enterprise report, the number of companies with 40% or more of their AI projects in full production is expected to double within just six months. Meanwhile, NVIDIA’s 2026 State of AI report confirms what practitioners at ATP already know: enterprises have moved from experimentation to full-fledged deployment, touching everything from content development to compliance workflows. 

In short: the question is no longer whether to adopt AI. It’s whether your infrastructure is ready to support AI at the scale and standard that high-stakes credentialing demands.

That’s a fundamentally different problem. And it requires a fundamentally different solution.

Why Assessment Is the Hardest Place to Get AI Right 

Before we talk about solutions, let’s be precise about the challenge because the assessment industry is not like other industries. When a retail company deploys an AI tool that makes a suboptimal recommendation, a customer gets the wrong product suggestion. Annoying. Fixable.

When a credentialing body deploys an AI tool that generates biased, poorly calibrated, or technically unsound exam items, a candidate’s career hangs in the balance. A license is granted or denied. A professional’s competency is judged. An institution’s integrity is staked.

The margin for error isn’t slim. It’s essentially zero. This is why the research community has been unambiguous: AI in high-stakes assessment must be domain-specific, human-supervised, and rigorously validated. A human-in-the-loop framework isn’t a nice-to-have in many cases, it’s mandated by accreditation bodies including NCCA and the forthcoming updated ISO/IEC 17024 standard.

Generic AI tools, the kind built for broad content generation, simply cannot meet this bar. As Springer Nature’s research confirms, even the most capable large language models require specialized measurement frameworks to produce items that are valid, unbiased, and defensible at scale.

The assessment industry doesn’t need faster content. It needs better content, generated responsibly, governed carefully, and ready for the highest-stakes environments in the world.

The Three Shifts Reshaping Item Development in 2026 

Three structural changes are converging right now and together; they are forcing the issue for every organization that develops and delivers high-stakes assessments. 

1. Volume pressure is increasing

As credentials proliferate and workforce credentialing expands globally, the demand for exam content is outpacing the capacity of traditional SME-driven development cycles. The old model, convene a panel of experts, draft items manually, review over months cannot scale to meet what’s coming.

2. The talent gap is real

Experienced assessment professionals and item writers are a limited, expensive resource. Organizations that depend entirely on human development pipelines are vulnerable; to turnover, to cost inflation, and to the simple math of demand exceeding supply.

3. Speed-to-market is now a competitive differentiator

Credentialing programs that can bring new assessments to market faster, without compromising quality, have a structural advantage. Those that can’t are watching relevance erode in real time.

These three forces don’t just make AI adoption attractive. They make it inevitable. The question is which AI, built how, governed by whom.

What “Production-Ready” Actually Means 

Here’s where most of the industry conversation gets loose and where it matters most to be precise.

Not all AI is production-ready for assessment. The term gets applied to tools that are, at best, useful for drafting first-pass content that still requires extensive human rework. That’s not infrastructure. That’s a time-saving shortcut with a long quality-control tail.

True production-ready AI for item generation requires four non-negotiable capabilities:

1. Domain specificity

The model must understand the conceptual structure, terminology, and nuance of the subject domain, not just generate plausible-sounding text. Generic language models hallucinate. Domain-trained models calibrate. 

2. Measurement soundness

Items must meet established standards for difficulty, discrimination, and construct alignment. AI that generates items without rigorous validation isn’t generating exam-quality content, it’s generating risk.

3. Format versatility

Real assessment programs need more than multiple-choice questions. Case studies, scenario-based items, performance tasks, and complex constructed-response formats all require different generation logic.

4. Real-time feedback integration

The system must learn, continuously, from performance data, review cycles, and item-level analytics. Static AI produces static quality. Dynamic AI compounds its own improvement. 

Without all four, what you have is a pilot. With all four, you have infrastructure. 

Introducing GenQue: Built for Where Assessment Is Going 

This is exactly the problem OpenEyes Technologies built GenQue to solve. 

GenQue is a patent-protected AI engine for automatic item generation, purpose-built for the credentialing, certification, and high-stakes assessment industry. Not adapted from a general-purpose tool. Not bolted onto existing software. Engineered from the ground up for the standards this industry demands. 

1. It cuts development time by up to 70%

That’s not a projection, it’s the operational reality for organizations using GenQue today. Items that took weeks of SME time to develop are generated in seconds, freeing your subject matter experts to do what only they can do: validate, refine, and approve content with domain authority and professional judgment.

2. It uses domain-specific AI, not generic models

GenQue’s engine applies dynamic sequencing and context-aware generation trained specifically for assessment environments. The result is items that are tailored, relevant, and aligned to the highest testing standards, not generic content that happens to look like a question.

3. It integrates real-time feedback analysis

Every item GenQue generates is informed by ongoing performance data. The system doesn’t just produce, it learns. Quality improves continuously, not incrementally.

4. It supports multiple formats

From traditional MCQs to complex scenario-based items, GenQue handles the full breadth of assessment formats modern credentialing programs require.

5. It keeps humans in control

GenQue is designed to augment your experts, not replace them. AI generates. SMEs validate. That human-in-the-loop architecture isn’t a limitation, it’s the design principle that makes GenQue defensible before any accreditation body in the world.

The Window Is Narrowing 

Here’s the strategic reality that every credentialing leader needs to sit with: the organizations that move first on production-ready AI infrastructure will not just be more efficient. They will be structurally more capable of launching new credentials faster, responding to workforce changes quicker, and scaling quality in ways their competitors cannot match. 

According to IBM’s 2026 analysis, every AI program must now attach to clear KPIs and a defensible ROI model before scaling. The era of AI investment justified purely by its innovative potential is ending. What replaces it is the era of AI that delivers reliably, measurably, at scale. 

GenQue is built for that era. 

The industry talked about AI at ATP 2026. OpenEyes has been building it.