For the past few years, artificial intelligence (AI) in the public sector has mostly played an advisory role: analysing data, spotting patterns and helping people make better decisions. That chapter isn’t over, but the next one is already here. Agentic AI can monitor a process, detect an exception and complete an approved action without needing someone to click through every step along the way. But are governments ready to put agents to work inside public processes?
But workforce pressures alone aren't what's driving the conversation. AI readiness is emerging as the bigger challenge. Recent research from McKinsey found the public sector has the lowest AI maturity score of any sector measured—28 out of 100, well below the global average of 33² —while AI rose to the top spot in NASCIO's 2026 state chief information officers (CIO) priorities survey, displacing cybersecurity, which held first place for 12 consecutive years³.
Agencies are being asked to do more with constrained resources, while continuing to meet rising expectations from the communities they serve. Residents expect government services to be as responsive, accessible, and personalised as the digital experiences they encounter elsewhere. That makes AI readiness more than a technology issue. It's becoming an operational imperative.
This shift matters because the next generation of AI doesn't just advise—it acts.
The model itself isn’t the thing holding these agencies back. What matters is operational readiness: the workflows, data, controls, guardrails and oversight are in place before an agent is allowed to act.
Just 35% of respondents are confident AI use will be fair and unbiased⁵. Similar shares believe governments will be transparent about where AI is used and that personal information will stay protected. Respondents are nearly split on human oversight of critical decisions, with 37% confident that oversight will happen and 36% expecting it won’t⁵.
The public sector faces a higher bar than the private sector here—and for good reasons. Decisions about funding, procurement, benefits, licenses, permits and employment must be explainable, reviewable and auditable². Residents also tend to experience the government as a single system rather than a collection of separate organisations and departments, so when one agency gets AI wrong, the damage travels².
Staff haven’t fully bought in either. Only 1 in 5 public sector employees expects AI to meaningfully change their daily work, and just 3 out of 10 workers trust their employers to develop AI safely, compared with 71% across industries overall². Meanwhile, AI is already showing up in consequential processes without the controls that would make those decisions defensible. Among US state and local human resources (HR) managers surveyed in 2026, 64% said their organisation has no basic AI governance requirements in place, and only 17% require a staff member to review AI recommendations tied to hiring, discipline or benefit denials¹.
The gap between individual gains and organisational returns is the clearest evidence that speed isn't the variable. McKinsey's 2026 global AI survey found that 80% of respondents said AI has improved their productivity, and about half said it helps them make better decisions. Fewer than 40% report that AI contributes to their organisation's earnings, and a small group of genuinely high-performing organisations has stayed flat at roughly 6%⁷. Adoption went up, but returns didn't follow.
What separates the high performers from everyone else is rewiring processes. Nearly three-quarters of organisations fundamentally redesigned workflows because of their AI use, up from 55% in 2025⁷. The same pattern holds in government: About 70% of AI programs built around a specific government domain reach production, compared with just 30% built around a single isolated use case².
That’s the foundation organisations need before they can trust agents to act inside a process, and it’s the foundation Infor™ Velocity Suite is designed to sit on. Industry AI agents and process mining are grounded in Infor’s public sector data model, so they read grants, fund accounting and multi-organisation structures the way a public finance team already does, rather than treating government work like generic back-office processing.
Many organisations haven't drawn that line internally, either. More than half of all AI oversight work currently happens after a system is already in production². Staff concerns are running ahead of leadership: 20% of state and local employees are very or extremely concerned about AI modifying or eliminating jobs, while only 7% of HR leaders rate that as a very important issue¹.
These fears have outrun the evidence, however. In McKinsey's 2025 AI survey, one-third of respondents expected AI-driven workforce reductions over the following year. But when the 2026 survey measured actual workforce impacts, only 14% reported an AI-related decline in workforce size, and two-thirds reported little or no change at all⁷. The research is cross-sector rather than government-specific, but it gives leaders data to back reassurance: the reductions people braced for did not materialise at the expected scale.
Agencies should be explicit about this. The goal isn’t to take people out of government processes. It’s to give people more time for the work that needs judgement, expertise and public accountability.
Why the conversation is changing now
Staffing shortages are the usual explanation for governments’ AI push, but that’s only part of the story. More than 60% of agencies still see fewer qualified applicants than they have open skilled positions, according to the State and Local Government Workforce Survey Report 2026 from MissionSquare Research Institute¹. At the same time, hiring freezes are becoming more common as fiscal pressures are growing.But workforce pressures alone aren't what's driving the conversation. AI readiness is emerging as the bigger challenge. Recent research from McKinsey found the public sector has the lowest AI maturity score of any sector measured—28 out of 100, well below the global average of 33² —while AI rose to the top spot in NASCIO's 2026 state chief information officers (CIO) priorities survey, displacing cybersecurity, which held first place for 12 consecutive years³.
Agencies are being asked to do more with constrained resources, while continuing to meet rising expectations from the communities they serve. Residents expect government services to be as responsive, accessible, and personalised as the digital experiences they encounter elsewhere. That makes AI readiness more than a technology issue. It's becoming an operational imperative.
This shift matters because the next generation of AI doesn't just advise—it acts.
When recommendations become actions
Agentic AI raises the stakes because agents don’t stop at recommending. They coordinate activities across systems and complete approved actions on their own. Gartner predicts that at least 80% of governments will deploy AI agents to automate routine decision-making by 2028. The same research found that 41% of government respondents cited siloed strategies and 31% cited legacy systems as barriers to adopting digital technologies⁴.The model itself isn’t the thing holding these agencies back. What matters is operational readiness: the workflows, data, controls, guardrails and oversight are in place before an agent is allowed to act.
Where public confidence breaks down
People can picture AI making government faster, but they aren’t yet convinced it will make government fairer, clearer, or more answerable. The Organisation for Economic Co-operation and Development (OECD), an intergovernmental group that tracks public trust and policy performance among its members, surveyed residents about government AI. Across OECD member countries, only 4 out of 10 people⁵ are confident that government use of AI will deliver better-tailored services, while expectations around cost reduction are about the same. Confidence levels drop even further on the questions that determine legitimacy, such as fairness, transparency, privacy and human oversight.Just 35% of respondents are confident AI use will be fair and unbiased⁵. Similar shares believe governments will be transparent about where AI is used and that personal information will stay protected. Respondents are nearly split on human oversight of critical decisions, with 37% confident that oversight will happen and 36% expecting it won’t⁵.
The public sector faces a higher bar than the private sector here—and for good reasons. Decisions about funding, procurement, benefits, licenses, permits and employment must be explainable, reviewable and auditable². Residents also tend to experience the government as a single system rather than a collection of separate organisations and departments, so when one agency gets AI wrong, the damage travels².
Trust must be designed before go-live
The public hasn’t made up their minds yet about AI. Neutral and “don’t know” responses to the OECD’s AI trust questions reach 25% combined⁵, which is an open window rather than a verdict. Governance is what closes it.Staff haven’t fully bought in either. Only 1 in 5 public sector employees expects AI to meaningfully change their daily work, and just 3 out of 10 workers trust their employers to develop AI safely, compared with 71% across industries overall². Meanwhile, AI is already showing up in consequential processes without the controls that would make those decisions defensible. Among US state and local human resources (HR) managers surveyed in 2026, 64% said their organisation has no basic AI governance requirements in place, and only 17% require a staff member to review AI recommendations tied to hiring, discipline or benefit denials¹.
The gap between individual gains and organisational returns is the clearest evidence that speed isn't the variable. McKinsey's 2026 global AI survey found that 80% of respondents said AI has improved their productivity, and about half said it helps them make better decisions. Fewer than 40% report that AI contributes to their organisation's earnings, and a small group of genuinely high-performing organisations has stayed flat at roughly 6%⁷. Adoption went up, but returns didn't follow.
What separates the high performers from everyone else is rewiring processes. Nearly three-quarters of organisations fundamentally redesigned workflows because of their AI use, up from 55% in 2025⁷. The same pattern holds in government: About 70% of AI programs built around a specific government domain reach production, compared with just 30% built around a single isolated use case².
AI readiness starts with the operating foundation
Agencies do not become AI-ready simply by installing AI. They become AI-ready by modernising the systems, processes and governance models that make responsible automation possible. Elsinore Valley Municipal Water District did the work⁶. The California water district needed to keep pace with growth without adding a large team, so it rebuilt how work moved through the organisation by consolidating systems, making processes visible and creating a connected operating environment. The board gained visibility into process performance that used to take weeks of manual work to assemble, and invoice processing capacity rose 108%, from 75 to 156 invoices a week, handled by the same two staff⁶. None of this required an autonomous agent to make judgement calls. It required an agency that could see its own processes clearly enough to fix them.That’s the foundation organisations need before they can trust agents to act inside a process, and it’s the foundation Infor™ Velocity Suite is designed to sit on. Industry AI agents and process mining are grounded in Infor’s public sector data model, so they read grants, fund accounting and multi-organisation structures the way a public finance team already does, rather than treating government work like generic back-office processing.
Where the human loop matters most
None of this argues for removing people from the loop. It argues for being precise about where the loop earns its cost. Human sign-off should be defined by its consequence, not category. Benefit denials, license revocations or procurement anomalies need a person in the loop. A grammar check on a draft letter does not. Many agencies still default to reviewing everything or almost nothing because the line has never been drawn. Public trust rests on residents knowing a person can review, override and explain a decision that affects them.Many organisations haven't drawn that line internally, either. More than half of all AI oversight work currently happens after a system is already in production². Staff concerns are running ahead of leadership: 20% of state and local employees are very or extremely concerned about AI modifying or eliminating jobs, while only 7% of HR leaders rate that as a very important issue¹.
These fears have outrun the evidence, however. In McKinsey's 2025 AI survey, one-third of respondents expected AI-driven workforce reductions over the following year. But when the 2026 survey measured actual workforce impacts, only 14% reported an AI-related decline in workforce size, and two-thirds reported little or no change at all⁷. The research is cross-sector rather than government-specific, but it gives leaders data to back reassurance: the reductions people braced for did not materialise at the expected scale.
Agencies should be explicit about this. The goal isn’t to take people out of government processes. It’s to give people more time for the work that needs judgement, expertise and public accountability.
What separates AI adoption from AI value
Agencies that succeed with agentic AI will not be the ones that automate the fastest. They will be the ones that make automation trustworthy enough to scale. McKinsey's latest global AI survey points the same way: The small group of high-performing organisations capturing financial returns from AI is more likely than their peers to have decided where humans belong in the loop, and more likely to be actively working on explainability, fairness and unintended action⁷. In the public sector, accountability is not an obstacle to AI value. It’s what makes AI value possible.Five steps agencies can take now
- Map your decision inventory by consequence rather than by department, and flag which decisions legally or ethically require human sign-off before any agent touches them.
- Audit where AI is already running in HR, finance, or service delivery, including informally and check it against two questions: Is anyone reviewing the recommendations, and has anyone tested for bias?
- Identify the one or two workflows causing the most rework or delays, and redesign them before automating.
- Set a public standard for explainability. Define, in plain language, how a resident can ask for and receive review of an automated or AI-assisted decision.
- Budget for change management alongside the technology, not after it. McKinsey puts the ratio at roughly $5 US of change management for every $1 US of technology².
Most agencies don't know where they stand until something forces the question. Take Infor's AI readiness assessment to see where your governance, data and process foundation sits today.
Sources
1 MissionSquare Research Institute. (2026, August). State and Local Government Workforce Survey Report 2026. https://research.missionsq.org/posts/workforce/state-and-local-government-workforce-survey-report-2026
2 Vuppala, H., Fountaine, T., Ward, T., & D’Emidio, T. (2026, July 15). How can the public sector meet the AI moment?. McKinsey & Company. https://www.mckinsey.com/industries/public-sector/our-insights/rewiring-public-sector
3 NASCIO. (2025, December). 2026 State CIO Top 10 Priorities: Strategies, Policy Issues & Management Process. NASCIO. https://www.nascio.org/resource/state-cio-top-ten-policy-and-technology-priorities-for-2026/
4 Gartner. (2026, March 17). Gartner Predicts at Least 80% of Governments Will Deploy AI Agents to Automate Routine Decision-Making by 2028 [Press release]. Gartner. https://www.gartner.com/en/newsroom/press-releases/2026-03-17-gartner-predicts-at-least-80-percent-of-governments-will-deploy-ai-agents-to-automate-routine-decision-making-by-2028
5 OECD. (n.d.). OECD Survey on Drivers of Trust in Public Institutions 2026 Results. OECD Publishing. https://www.oecd.org/en/publications/2026/06/results-of-the-2025-oecd-survey-on-drivers-of-trust-in-public-institutions_96323a65.html
6 Patel, M. (n.d.) From reactive to proactive: How Elsinore Valley Municipal Water District is transforming finance operations with Infor Velocity Suite. Infor. https://www.infor.com/blog/elsinore-valley-municipal-water-district-finance-operations-transformation
7 Tinkoff, D., Van der Veken, L., Chui, M., & Balakrishnan, T. (2026, August 25). The state of AI in 2026: On the road to ROI. QuantumBlack, AI by McKinsey. https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai