Contents
- Why Agentic Coding Matters for Public Authorities and Public IT
- Special Conditions in the Public Sector
- AI Literacy Under the EU AI Act: Mind the Training Duty
- Which Operating Models for Coding Agents Are Possible
- Training Content for Administration and Public IT
- Step by Step to Public-Sector Training
- For Public-Sector IT Service Providers
- Conclusion
- Frequently Asked Questions
Agentic coding training for public administration and its IT service providers has to do more than an ordinary developer course: alongside productive use of coding agents, data sovereignty, information security, employee co-determination and traceability take centre stage. If these conditions are considered from the outset, the public sector can benefit from AI-assisted software development too. This article explains what matters, with a focus on the German and wider EU context.
Why Agentic Coding Matters for Public Authorities and Public IT
Public administrations, municipal data centres and public-sector IT providers are under considerable pressure: digitalisation projects must be delivered, specialist applications modernised and legacy systems replaced – often with scarce staff. Coding agents can help, for example with understanding old codebases, writing tests, migrations or documentation.
At the same time, the public sector has special requirements. Administrative data is often sensitive, decisions must be traceable and new tools go through formal approval processes. Training that ignores this reality fails at the latest when someone asks whether the tool being demonstrated may even be used in-house.
Special Conditions in the Public Sector
- Data protection and data sovereignty: The GDPR and regional data protection rules apply in full. There is often an additional expectation that data and source code are processed in the EU or in the organisation's own data centres.
- Information security: In Germany, many institutions follow the BSI IT-Grundschutz framework. New tools must fit into the existing security concept.
- Co-determination: Staff councils usually have to be involved when technical systems are introduced that could record performance or behaviour.
- Procurement: Tools and training are procured under public procurement rules, which affects timelines and selection.
- Traceability: Changes to specialist applications must be documented and verifiable – even when an agent proposed them.
This overview is not legal advice; specific requirements depend on the jurisdiction, institution and procedure.
AI Literacy Under the EU AI Act: Mind the Training Duty
The EU AI Act has been in force since 1 August 2024. Since 2 February 2025, Article 4 requires providers and deployers of AI systems to take measures so that their staff have a sufficient level of AI literacy. In principle this also applies to public bodies using AI tools. Practical training on coding agents can be one building block towards meeting this requirement – for example by covering how the systems in use work, their limits and their risks. Details on the literacy duty are covered in AI literacy under the EU AI Act. Again: not legal advice; have it reviewed by legal counsel if in doubt.
Which Operating Models for Coding Agents Are Possible
The choice of operating model determines what can be practised in training. An overview:
| Model | Advantages | Challenges |
|---|---|---|
| Cloud service with EU data processing | Powerful models, quick start | Contract review, acceptance by data protection and security |
| Self-hosted models (on-premise) | Full data control, no code leaves the building | Hardware, operations, often lower model performance |
| Hybrid | Non-critical tasks in the cloud, sensitive ones local | Clear rules needed on which code may go where |
| Practice environment only | Training possible without production code | Transfer to daily work is harder |
More on local options in local LLMs for software development. In many cases it makes sense to start training with synthetic code or non-critical modules while approval for production code is reviewed in parallel.
Training Content for Administration and Public IT
Content, format and scope are agreed individually. For the public sector, the following focus areas work well:
- Fundamentals: How coding agents work, what they can do and where their limits are.
- Secure operation: Permissions, sandboxing, approving commands and handling credentials – see also agentic coding and data security.
- Legacy and specialist applications: Using agents to understand, document and gradually modernise older systems.
- Quality assurance and traceability: Review processes, labelling agent-generated changes, documentation.
- Governance: Internal usage policies, roles and responsibilities. A useful companion: AI governance in development teams.
Step by Step to Public-Sector Training
- Clarify needs: Which teams, which systems, which goals?
- Involve stakeholders: Inform information security, data protection and the staff council early.
- Choose the operating model: Which tools are already approved or can be assessed?
- Prepare procurement: Write the service description and observe procurement requirements.
- Run the training: With a practice environment or approved code, on-site or remote.
- Embed policies: Turn the results into an internal guideline for using coding agents.
For Public-Sector IT Service Providers
External IT providers developing software for public authorities face a double challenge: they want to use agentic coding to deliver more efficiently but must comply with their public clients' requirements. It is important to clarify transparently with clients whether and how AI tools may be used and to record this contractually. Training can prepare for this specifically, for instance with templates for internal policies and a clear documentation process.
Through app-entwickeln-lassen.de we place AI developers from real-world practice as trainers. Workshops start from €799 net; you can request a concrete quote, for example as input for your procurement, by e-mail.
Planning an app? Request a free fixed-price quote in 2 minutesConclusion
Agentic coding can achieve a lot in public administration and among its IT providers, especially for modernisation and documentation. Training succeeds when it addresses data sovereignty, information security, co-determination and traceability from the start and takes the right operating model into account. That builds AI competence your teams are actually permitted to apply in daily work.
Frequently Asked Questions
May public authorities use cloud-based coding agents?
That depends on internal rules, the type of code and the provider's contractual terms. A review by information security and data protection is often required; this is not legal advice.
Does agentic coding training fulfil the AI literacy duty?
It can be an important building block if it covers how the systems in use work, their limits and risks. Each institution should assess for itself whether its measures are sufficient overall.
Can training take place without production code?
Yes, with practice projects or synthetic code. This is often a good start while approval for real systems is still being reviewed.
Does the staff council need to be involved?
Involving it early is usually advisable and often mandatory when new technical tools are introduced. The exact obligations depend on the applicable staff representation law.