GenAI Isn't Magic. Adoption Is a Literacy Challenge, and States That Treat It That Way Will Win.

Artificial intelligence has just claimed the top spot on NASCIO’s State CIO Top Ten for 2026, the first time in the list’s twenty-year history, and 90 percent of state CIOs report GenAI pilot projects underway, with proofs of concept close behind. The technology, though still evolving, has raced far ahead of end users’ conceptions of what it is and isn’t. The adoption question is wide open. So let us state the thesis plainly: GenAI is neither magic nor menace. It is a tool, and done right it is a dramatic productivity accelerator in service of a goal that has nothing to do with technology: helping civil servants do the best job they can for the people they serve. What separates the programs that deliver from the pilots that fade is not the model. It is whether a state builds GenAI literacy across its workforce, audience by audience, with each owning its part.

One lever matters above all: an accurate understanding of what GenAI is and isn’t. It is a tool that advances productivity and thinking, but it will not do anyone’s job for them. That understanding shows up as learned behaviors. The first is prompt craft: today’s models are capable enough for most use cases that model choice is rarely the constraint, so the largest predictor of effective, enthusiastic adoption versus skeptical abandonment is the user’s ability to describe what they want, with context, constraints, and clear objectives. The second is productive skepticism: pushing back on outputs, questioning conclusions, and verifying anything that matters. The third is judgment about fit: knowing which tasks belong to the tool (synthesis, drafting, exploration, first passes at analysis) and which do not. The fourth is data discipline: sensitive data never enters the tool unless it is operating within a well-constructed private LLM. Literacy also means treating the first output as a first draft: iterating, supplying the context the model cannot know, and recognizing that a confident tone is not evidence of a correct answer. When users internalize these behaviors and discover through their own experience that better prompts and sharper questions produce markedly better outcomes, adoption takes a sharp upward turn. A literate workforce is also a protected one: people who understand what the tool is and is not are far harder for hype, or for fear, to push around. Treat this full GenAI literacy as a first-order craft skill, developed through prompt libraries, role-specific tutorials, and peer learning. But literacy lands differently across a government workforce, and each audience faces a distinct challenge with a distinct role to play.

Technology and Data Staff

Technology and data staff carry the first wave of challenges, and it’s rarely discussed: the foundational data, infrastructure, and governance work that precedes any deployment, often well before a model ever touches operational data. When a state commits publicly to GenAI without a parallel conversation about this prework, technology and data teams react with informed alarm; they understand exactly what gap they’ll be expected to close. The remedy is twofold: honest, early framing about realistic timelines and resources, and accelerators (prebuilt APIs, data mapping tools, design maps, reference use cases) that compress the scaffolding work so these teams can focus on the agency-specific complexity that genuinely requires custom effort. No group’s candor matters more: when they say the foundation isn’t ready, that is expertise speaking, not resistance, and their early seat at the planning table is what keeps timelines honest.

Executives

Executives face an entirely different challenge. Seasoned public-sector executives are no strangers to technology’s limits or to implementation timelines that run longer than hoped. What is new is the combination GenAI presents: a genuinely generational change in how government works, arriving at the same moment it becomes a political hotspot. A sponsor working from inflated expectations may quietly disengage at the exact moment the program needs sponsorship most, a dynamic that can kill GenAI initiatives faster than technical failure ever will and one that public scrutiny only accelerates. Now is the time for full candor, and executives are the ones positioned to demand it: insist on seeing the prework laid out plainly, expect an iterative improvement curve rather than a launch event, and hold to the principle that distinguishes GenAI from prior enterprise technology: value compounds with use. The sponsorship only executives can provide, sustained and visible through the unglamorous middle, is what carries a program from pilot to practice. The curve can also be steepened: bounding models against organizational data rather than the open internet substantially accelerates time-to-value, and meaningful returns can arrive in months rather than the multi-year horizon skeptics assume.

Managers

Managers may be the most consequential audience of all, and the most overlooked. Administrations arrive and depart; managers remain, carrying the institutional memory of every initiative that came before, including the ones that quietly died. They are the layer where executive ambition either becomes daily practice or gets politely waited out, and their stance cascades: a manager who models GenAI literacy and a practical, skeptical enthusiasm produces a team that does the same. They also hold rational concerns of their own, since much of what GenAI does well (synthesizing, summarizing, reporting upward) resembles the connective work managers perform. That is why managers belong at the table first, not as recipients of an adoption strategy but as co-designers of it. Initiatives move through government because managers move them, and engaging managers early and in genuine collaboration honors that reality. It also gives them what they need most: a real hand in shaping how GenAI relieves the reporting burden, so more of their time goes to the coaching and judgment only they can provide.

Professional Users

Professional users (attorneys, policy analysts, procurement specialists, finance and budget analysts, auditors and appeals staff, government relations, communications, HR, and the other professionals whose work supports every agency and its programs) are often the easy case. They frequently experience what the promise looks like in practice: dramatically faster research, drafting, and synthesis. Their challenge isn’t engagement but calibration: which outputs to trust and when to push back, which to verify, and when GenAI is the wrong tool entirely. Their role reaches beyond their own productivity: they become the institution’s true early adopters and champions, living proof that the tool works.

Front-Line Workers

Front-line workers are the hardest case, because their concern is legitimate: GenAI may eventually perform parts of their job, and evasive talk of “augmentation” reads correctly as evasion. The cultural narrative that casts AI as either magical or menacing hands front-line staff the menace framing, AI as job-taker, and they encounter it daily. The sentiment gap is measurable: Accenture research cited in NASCIO’s 2026 Agentic AI report found that only 55 percent of surveyed front-line government employees feel positive about integrating AI into their day-to-day work. But an honest reframe exists. The public will continue to demand live human interaction on matters affecting their benefits and families. What changes is which queries humans handle. When agentic AI absorbs routine status checks and procedural questions, front-line staff gain the breathing room to bring time, attention, and judgment to the complex, emotionally weighty queries they were trained for, with GenAI support. That message is genuinely different from “AI will replace you”, and case workers, call center reps, and other critical front-line workers can verify it against their own daily experience. They are also the people best positioned to say where the tools fall short, and that feedback should be actively sought.

Sequencing: Activity, Not Hierarchy

How these audiences come together matters as much as how each is engaged. The traditional government rollout, executives first, then technology staff, then managers and professionals, with front-line workers the last to know, is used up: it moves slowly, breeds rumor, and lets fear arrive before facts. A better frame is activity, not hierarchy. Executive commitment is a starting gun, not a first tier. The moment it sounds, literacy building begins across the entire workforce while technology and data teams lay the foundation in parallel, and managers join early to co-create the rollout strategy they will be asked to carry. When everyone is learning at once, the front line hears the real story from the program rather than a distorted one from the rumor mill.

Three Failure Modes

If literacy still sounds like training-catalog language, consider what its absence looks like: three failure modes recur across GenAI programs, and all are literacy failures. The hallucination credibility cliff: a single confident-but-wrong output, a fabricated citation or a non-existent regulation, can collapse trust far beyond the team. In government, administrative, legislative, and ultimately public trust are all on the line, and the verification habits literacy instills are the first defense. The guardrails that institutionalize them (governance frameworks, human review, scoped data access) deserve a piece of their own, as does the agent question: literacy governs how people use these tools, while what agencies permit autonomous agents to do on their own is a matter of governance, not training. Shadow GenAI: many agencies have placed absolute restrictions on accessing GenAI from state systems, a reasonable first response that has a shelf life: restriction is not prevention; remote workers can reach public tools from personal computers, and staff in the office can reach them from the phones in their pockets. Banning without alternatives produces compliance theater. The gap closes when the data discipline literacy teaches pairs with sanctioned alternatives that genuinely meet the needs that sent staff to public tools in the first place. And the second-day problem: initial enthusiasm comes easily, but sustained, effective use does not; it requires embedding the tool where people already work and keeping the value compounding through ongoing peer learning. None of what is outlined here is easy, and that is precisely the point.

GenAI isn’t magic. Its adoption is a literacy challenge, and treated as one it becomes something far more useful than magic: a dependable accelerator of the work civil servants do for the people they serve. The next piece in this series turns the argument into a work plan: a five-competency literacy baseline, a program component for each audience, and a timeline keyed to a state’s first chat assistant. The states that treat it that way will win.

---

Keith Cherry is the co-founder and Managing Partner of Barnaby Street, a senior advisory firm specializing in Generative AI strategy, governance, and organizational change management for state government. The firm pairs senior government practitioner experience with specialist implementation partners to help agencies move from GenAI ambition to adoption that lasts. If you found this useful, we’d welcome the conversation.