The Starting Point: Catching Up in a Saturated Market
The market for correctional software is small and has been dominated for years by a handful of established providers. Their systems have evolved over a long period of time and, as a result, offer a broad range of features. New solutions therefore face the challenge of catching up to this lead in a relatively short period of time.
The technical foundation for this is Justalis’s architecture. The system was designed from the outset to be modular, incorporating the latest technological concepts and clearly separated functional modules. At the same time, the operational environment places high demands on the solution: high availability, strict security and data protection requirements, end-to-end traceability, accessibility as a contractual criterion, and operation on sovereign infrastructure.
Added to this are the requirements from public tenders. The specific functions required are not known until the relevant documents are published. Only then can it be assessed which requirements are already met and where improvements must be made within the tender period.
However, this requires a comprehensive analysis first. The request for proposal documents must be thoroughly reviewed, and each individual requirement must be compared with the system’s current status. Only on this basis can we reliably determine which functional gaps actually exist.
This process is both technically and time-consuming. With a traditional development process, it would have been nearly impossible to meet the deadlines set for this project under these circumstances.
Integrating AI into the Development Process
The first step was to use AI for code generation. Based on initial experiences, workflows, subagents, and rules were developed for Claude Code, which define the project’s conventions and processes.
These structures are continuously refined and reviewed with each sprint. Approaches that do not prove effective are adjusted, while successful processes are further refined and expanded.
In the next step, the workflow surrounding the backlog was also expanded. Claude was given direct access to the tickets and can process tasks from them all the way through to the finished pull request.
Humans remain involved in the process throughout. Between the ticket and the pull request, progress is regularly reviewed and corrected as needed. Responsibility for the code remains with the developers. Anyone who approves a pull request must be able to understand and defend the code it contains at the same level as if they had written it themselves.
In addition, the pipeline includes technical guardrails that apply regardless of whether AI is used. These include architectural tests, static analyses, security scans, and automated tests. They ensure that the project’s defined rules and quality standards are adhered to, regardless of whether the code was created by a human or by Claude.
From the tender to the Gap Analysis
After the first few sprints, the team had become sufficiently familiar with using Claude in the development process to expand its scope of application. Since then, product owners and stakeholders have also been using its capabilities alongside the developers when working on tenders.
Claude is used in two key areas: for the structured analysis of the documents and for addressing those requirements that are not yet fully covered by the current state of the system.
Claude Cowork and Claude Code handle different tasks in this process. Cowork reads the complete request for proposal documents and extracts the individual requirements from them. Claude Code knows the current state of the platform and compares these requirements with the existing solution.
The result is a gap analysis that shows, for each individual requirement, whether it has already been met, partially met, or not yet implemented. A process that previously took several days can now be completed in just a few hours.
The identified gaps are added to the backlog as new requirements. Since Claude Code understands the existing architecture, each of these work items already includes a description of how the required functionality integrates into the existing modules. They then go through the same process as all other tickets: implementation, pull request, and review.
In this way, the missing features were implemented within the deadline. Since the comparison is performed directly against the codebase, it is clear for each requirement from the request for proposal where it is already covered in the system and where it needs to be implemented.
Common standards make all the difference
At the beginning, each member of the team had individual access to Claude and used the tool according to their own work style. While this already saved time on individual tasks, it had only a limited impact on the development process as a whole.
The effect became significantly greater with the introduction of shared standards and clearly defined governance. Since then, workflows, rules, and subagents have applied to the entire team and are versioned centrally in the repository. This produces comparable results that can be verified in a traceable manner and utilized by all stakeholders.
Another lesson concerns the scope of individual changes. With AI, even larger features can be implemented in a single pass. However, this quickly results in extensive changes across numerous files, which become difficult to review in detail.
This point becomes particularly relevant during human reviews. A pull request with several hundred modified lines may be technically correct, but it can only be reliably verified with a correspondingly high level of effort.
To ensure that responsibility truly remains with humans, tickets must therefore be structured appropriately, and pull requests must be deliberately kept manageable. Only in this way is a thorough and traceable review possible.
Greater productivity creates room for new projects
“With EBCONT’s structured support, we have firmly integrated AI into our day-to-day development work: from the development of concepts and user stories to feature development and automated quality checks. This has significantly increased our productivity and enabled us to implement projects that were previously hardly feasible from a business perspective.”
Dietmar Stefani, Chairman of the Board of Directors, Justalis Holding AG
A Process That Grows Along With the Product
Justalis has quickly evolved into a competitive product. The development process that has emerged in the process now extends well beyond individual coding tasks. Today, AI supports the workflow from ticket to pull request, while architecture, security, quality, and accountability remain the responsibility of the team.
At the same time, the process remains intentionally adaptable. Workflows, rules, and guardrails are reviewed and refined with every sprint. New generations of models also continuously influence which tasks can be meaningfully supported or automated.
EBCONT also assists other companies in implementing such working methods, from strategic alignment and technical setup to playbooks and governance that are directly embedded in the development process and the code.
Learn more at EBCONT Agentic Engineering.