Agentic Coding Workshop for Public Sector IT: Procurement, Sovereignty

How public sector IT teams can run agentic coding workshops that respect procurement rules, data protection and digital sovereignty requirements.

Short answer

Public sector teams can train on coding agents today if they separate the learning environment from production data and document the decisions procurement and data protection officers will ask about. A good workshop covers tool options from cloud to self-hosted, works on non-sensitive code and ends with a pilot proposal that fits public procurement.

On this page
  1. Why public sector IT is different
  2. Separate learning from production
  3. Tool options through a sovereignty lens
  4. Workshop agenda for public sector teams
  5. Procurement realities
  6. Beyond coding: governed AI across the organization
  7. What participants take home
  8. FAQ
  9. Tools from the jocapps ecosystem

Why public sector IT is different

Engineers in ministries, municipalities, public utilities and their IT service providers face the same pressure as everyone else: legacy systems, a shortage of skilled staff and growing demand for digital services. Coding agents can help with exactly that. But the path to using them is shaped by constraints that private companies feel less strongly:

  • Procurement law governs how tools and services are bought, often with thresholds and tender requirements
  • Data protection officers and information security officers must approve data flows, sometimes with formal documentation
  • Digital sovereignty expectations favour EU hosting, open standards and avoiding lock-in
  • Accessibility obligations apply to the software that is built
  • Staff councils or works councils may need to be involved when new tools change how people work

A workshop that ignores these constraints produces enthusiasm and then a long stall. A workshop designed around them produces skills and a realistic route to approval. At neooutput we design public sector sessions with that second outcome in mind.

Separate learning from production

The most effective move is to decouple training from the approval of production use. Engineers can learn agentic workflows on code that carries no risk while the formal process for production use runs in parallel.

Workshop settingSuitable whenTypical code base
Open-source or sample repositoryNo tool is approved yetPublic projects similar to your stack
Sanitized copy of an internal projectData protection agrees on a test setupInternal code without secrets or personal data
Approved tool on internal repositoriesA pilot has been approvedReal projects with defined rules
Self-hosted or sovereign model setupRequirements exclude external providersAny internal code within the approved environment

Many administrations publish code on open platforms already. Those repositories make excellent workshop material because they resemble the real stack and carry no confidentiality concerns.

Tool options through a sovereignty lens

Participants should understand the spectrum of options, not just one product:

  • Commercial cloud agents such as Claude Code, OpenAI Codex or Cursor, with enterprise terms and, where offered, regional data processing
  • Access via an API gateway operated by the organization or its IT service provider, which can log usage and enforce policies
  • Open-source agents combined with models hosted in an EU data centre or on premises
  • Hybrid setups where non-sensitive work uses cloud tools and sensitive work uses self-hosted models

We do not pretend that every option delivers the same quality. Part of the workshop is an honest comparison on the same tasks, so decision makers can weigh capability against sovereignty with evidence. Our article on open-source and self-hosted coding agents goes deeper into that option.

Workshop agenda for public sector teams

Module 1: Framework and rules of the game (90 minutes)

What coding agents do, what data they see, and which decisions the organization needs to make. We invite the data protection officer and information security officer to this module. Their questions shape the rest of the day.

Module 2: Hands-on with agentic workflows (half day)

Participants work through realistic tasks: understanding a legacy module, writing tests for it, fixing a bug, documenting an interface. They write a context file (AGENTS.md or CLAUDE.md) that captures the project's conventions.

Module 3: Quality, accessibility and documentation (90 minutes)

Public software must be maintainable for years and often accessible. We use agents to check components against accessibility criteria, generate documentation that survives staff turnover and produce test coverage that makes future changes safer.

Module 4: Security and data protection in practice (90 minutes)

Permission settings, handling of secrets, prompt injection risks and what must never enter a prompt. We map these to the organization's existing information security framework so nothing has to be invented from scratch.

Module 5: From workshop to pilot proposal (60 minutes)

Participants draft a pilot proposal: scope, tool option, data classes, success criteria, duration and the documentation needed for procurement and data protection. This document is often the most valuable output, because it gives decision makers something concrete to approve.

Procurement realities

We cannot give legal advice on procurement, but some patterns help:

  • Training services and tool licences are often procured separately. Plan for both.
  • A clearly bounded pilot with a defined end date is usually easier to approve than open-ended use.
  • Framework agreements held by IT service providers can sometimes cover tool access. Ask early.
  • Documentation of data flows, processing locations and subprocessors speeds up the data protection review.

The broader context of building AI solutions in government, including accessibility and sovereignty, is covered in our article on AI development for the public sector.

Beyond coding: governed AI across the organization

Once IT teams have experience with coding agents, other departments usually ask for AI agents in their own workflows. That calls for audit trails, role-based access, integration with existing systems and EU data residency. kalAIDoCreate provides exactly that as an orchestration platform with an expert team, built for enterprises and the public sector.

What participants take home

  • Practical skill with at least one coding agent on realistic code
  • A context file template adapted to their projects
  • A comparison of tool options on the same tasks
  • A pilot proposal ready for internal approval

neooutput delivers these workshops in German or English, on site at your authority or remotely, and can tailor content for IT service providers that serve several public clients.

FAQ

Can public sector teams use cloud-based coding agents at all?

It depends on the data involved and the organization's assessment. Many teams start with non-sensitive or open-source code, where the risk is low, while the data protection review for broader use runs in parallel.

Do we need a tender to run a workshop?

That depends on your procurement rules, the value of the contract and existing framework agreements. We provide the documentation your procurement team needs to choose the right route.

Is a self-hosted model good enough for coding agents?

For some tasks, yes. Capability differs between models, so we recommend testing self-hosted options against your own tasks in the workshop before deciding.

Should the staff council be involved?

If new tools change how work is done or could be used to monitor performance, involving the staff council early is usually wise. A transparent workshop is a good way to start that conversation.

Tools from the jocapps ecosystem

Want this done right, and fast?

Talk to our AI Code Engineers and AI architects. In a free 30-minute call we look at your idea, prototype or roadmap and tell you honestly what it takes to ship it.

24/7 call support in English and German.

Free call

Free intro call

Tell us which task an AI agent should take responsibility for. We show how our platform, AI Code Engineers and operations plug into your stack.

  • 30 minutes, no obligation
  • Honest feasibility check
  • Concrete next step
Rather talk now? 24/7 call support in English and German+49 761 519 751 50