Insight Report · April 2026

A Practical Toolkit for Deploying Agentic AI in Government

Identifying high-impact opportunities for agentic AI.

An interactive resource based on insights from the report Making Agentic AI Work for Government: A Readiness Framework, written in collaboration with the World Economic Forum, Capgemini and the Global Government Technology Centre Berlin.

The ambition for agentic AI already exists – this report reveals the full scale of the opportunity

90% of constituents are willing to use AI agents for public sector interactions (Salesforce, 2025; based on a global survey of 11,750 constituents across several countries).
90% of public institutions plan to explore or deploy agentic AI within 2–3 years (Capgemini Research Institute, 2025; based on a survey of 350 public sector organizations across 13 countries).
50% of government activities may be suitable for automation with agentic AI (World Economic Forum, 2026; based on the analysis in this report, highlighting activities with significant potential and manageable complexity).

While the case for action is strong, common pitfalls slow down progress

40%+ – Gartner® Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027.

Three factors derail government initiatives:

Using an outdated lens – Traditional assessment methods don't fit agentic AI, leaving governments unsure where to start or how to sequence action.
Failing to prioritize – Without clear criteria, governments default to opportunistic use cases instead of investing where they will get the best results.
Implementing in silos – Pilots are confined to individual departments, making it hard for governments to reuse and scale them across institutions.

A new perspective for deciding where to start

Rather than thinking in departments, think in functions: recurring activities that use at least one end-to-end workflow to deliver a clear outcome.

New function-based approach – Workflow-oriented, cross-domain, reusable. Reflects operational reality, works universally across all governments, aligns naturally with agentic AI automation, and enables reuse and scaling across departments.

Traditional department-based approach – Siloed, hierarchical, department-defined. Neglects similar workflows across institutions, follows organizational structures not operational reality, and results in highly tailored pilots with limited cross-agency transferability.

Millions of global government activities captured in a concise yet comprehensive catalogue

At the heart of this approach is a new catalogue that organizes public sector activities into 70 core functions across nine categories:

Public services
Citizen communication and interaction
Security and risk response
Compliance and oversight
Policy execution
Policy planning and strategy
Operations and resources
Organization and workforce
IT infrastructure and data governance

Assessing potential against complexity reveals the highest-value starting points for agentic AI

Identifying functions is only the first step. To deploy agentic AI where it can make a real impact, you need to understand which functions combine the most promise with the least risk.

Agentic AI potential – Can agentic AI do this well and is the effort worth it? Assessed via: Potential for automation, Agent requirement, Volume and impact.

Implementation complexity – How hard is it to deploy agentic AI and what are the risks? Assessed via: Data type and quality, Technical integration complexity, Internal resistance, Ethical threat, Error consequence and function criticality, Privacy and data issues.

Mapping the functions across three levels of agentic AI readiness: high, medium and low

The scores reflect a global assessment of cross-country baseline estimates; local feasibility depends on context. Functions are mapped into three zones: Action zone (high potential, lower complexity), Exploration zone (moderate readiness) and Observation zone (lower potential or higher complexity).

Please enable JavaScript to use the interactive topography chart and explore the full assessment.

How to implement agentic AI in the public sector

The global assessment revealed a core set of recommendations that applies across all functions and contexts:

Prioritize public services functions as prime targets for agentic automation. Two-thirds of public services functions are positioned in the high-readiness area, with repeatable, rule-based workflows that deliver visible citizen-facing improvements.
Build on mature data foundations to accelerate agentic IT operations. Governments with strong IT operations and mature data structures can find early wins in their own technology management functions.
Focus on information-driven functions before tackling judgement-based ones. Functions that present, aggregate or monitor information are lower risk and easier to implement than functions where agents would influence sensitive decision-making.
Target fully automatable functions for early deployment. Functions that can be entirely automated face fewer organizational, ethical and error-related barriers and are strong initial candidates.
Improve multi-agency coordination to unlock long-term agentic value. Cross-agency automation can enable compounding value through multiagent systems.

Six steps for translating the global scores into a local implementation plan

The scores for each function reflect baseline estimates across countries, not country-specific recommendations. Follow these steps to create a tailored plan:

Establish local baselines – Assess local conditions that shape feasibility.
Develop risk-mitigation strategies – Address factors that raise complexity.
Reassess function-level scores – Use local knowledge to adjust global scores.
Sequence implementation – Start with high-readiness functions.
Validate through testing – Run small-scale pilots with real users and data.
Iterate assessment – Scale successful pilots and reassess regularly.

This is what agentic AI really looks like: Explore where it's making an impact in the public sector today

Ukraine: Diia.AI National AI Assistant – A national AI assistant enabling citizens to access over 200 public services through conversational interaction, used by more than 200,000 citizens since launch. The system follows a zero-trust approach and does not grant the model direct access to personal data.
Germany: AI-based construction permit system (BMDS) – An AI system screening large-scale infrastructure permit submissions on the same day, drafting legally grounded decisions under human oversight.
United Arab Emirates: HR AI Agent (FAHR) – An agentic HR interface autonomously resolving over 80% of HR legislation and policy inquiries for more than 50,000 federal employees.
Germany: Jira ticket creation (Federal Employment Agency) – An agentic system converting requests for change and user stories into compliant Jira tickets, saving approximately 150 hours per month.

Built for government realities: An actionable framework grounded in public-sector practice

70 core government functions · 2 agentic AI assessment dimensions · 3 levels of readiness

Five main takeaways for governments worldwide:

Think in functions, not departments – Agentic AI operates in workflows that cut across organizational boundaries. Shifting from department-based to function-based thinking enables cross-domain learning and reuse.
Balance ambition with feasibility – High agentic AI potential should be weighed against implementation complexity before operationalizing at scale.
Start where you have the best odds – Build capability and confidence with high-readiness functions. Early success creates the foundation and trust needed for more complex functions.
Tailor the results to your local context – Global scores are baselines. Local infrastructure, regulatory environments and cultural norms determine what is possible.
Reassess regularly – The assessment is dynamic and should be revisited as conditions evolve.

About this platform

This site was developed by Capgemini based on "Making Agentic AI Work for Government: A Readiness Framework". The report was written in collaboration with the Global Government Technology Centre Berlin and the World Economic Forum.

The content provides a high-level introduction to deciding where to start implementing agentic AI. For detailed guidance, insights, and concrete next steps, explore the full report.