Using AI in programming
AI helps with understanding code, designing solutions, refactoring and creating tests. We use it as a development partner and review its suggestions in the context of the project.
AI / RETHINKING SOFTWARE ENGINEERING
AI can do a lot.
What helps your project?
We code with AI. But we start with the task rather than the model: what work can be supported, what knowledge can be accessed and what results can be reliably verified?
AI expands our toolkit.
Responsibility stays with people.
FROM POSSIBILITIES TO APPLICATIONS
Applications of AI
Effective AI use starts with a specific task. Together, we explore where intelligent tools can support your development or enable new features.
From answers to actions.
AI agents can break tasks into steps, use tools and process intermediate results. For suitable workflows, we examine which steps can be automated and where human approval remains necessary.
Integrating AI features into applications
Whether knowledge search, assistance features or document processing: we treat AI as part of an application. Interfaces, access rights and business processes are part of the picture from the start.
LOCAL LLMS & PRIVATE AI
Local language models and internal knowledge
We bring expertise in deploying local large language models. We help companies use AI on their own infrastructure and make internal information available for their applications in a targeted way.
The model comes to your data.
Local LLMs enable AI processing in an environment under your own control. We consider model selection, hardware requirements and response times alongside your task — for a solution that fits your business and technical needs.
Turning documents into assistance.
Technical documentation, internal knowledge or project documents: local models can be combined with targeted knowledge retrieval using RAG. This enables assistance features that use relevant context and respect existing access rights.
Integrating local models into existing systems
We connect local LLMs to existing applications and workflows. Interfaces, permissions and traceable quality checks are part of the solution. What matters is how reliably the model supports your specific task.
CONTEXT MATTERS
Good models need
the right information.
A language model does not automatically know your application. It becomes relevant through the right context: requirements, code, documentation and clearly defined tools.
Context Engineering
Provide the right information at the right time. This includes project rules, relevant code excerpts and a deliberately chosen scope for the model context.
RAG · Retrieval-Augmented Generation
Retrieve relevant content from a knowledge source and provide it to the model as a basis. This can ground answers more effectively; their accuracy still needs to be checked.
MCP · Model Context Protocol
A standardised way to connect AI applications with tools and data sources. Appropriate and permitted access is defined explicitly.
FROM EXPERIMENT TO RELIABLE RESULTS
Explore.
Evaluate.
Apply with purpose.
Value first. Then the model.
We clarify the task, available data and success criteria. A manageable prototype helps identify benefits and limitations before scaling the solution.
Assess quality and limitations
Representative test cases, expert reviews and clearly bounded permissions make quality assessable. Failure modes, runtime and costs also need to be considered.
Integration and knowledge transfer
AI must fit existing architectures and processes. Together, we consider integration, traceability and the knowledge your team needs to use it.
No autopilot
for responsibility.
AI-generated code is a suggestion. Architecture decisions, tests and reviews remain essential. For data and tool access, we take confidentiality, permissions and project-specific requirements into account.
QUESTIONS & ANSWERS
Good to know.
How does aformatik use AI in programming?
aformatik uses AI to support understanding code, designing solutions, refactoring and creating tests. Suggestions are reviewed in the project context; architecture decisions, tests and reviews remain part of software engineering.
What does Agentic AI mean in a software project?
Here, Agentic AI refers to AI-assisted workflows that can break tasks into steps and call tools. Which steps may run automatically and where human approval is required depends on the specific use case.
What is the difference between RAG and MCP?
RAG provides a model with relevant content from a knowledge source. MCP is a protocol that allows AI applications to connect to tools and data sources. Both can be building blocks of an application, but serve different purposes.