An AI data center for SAP development
How Laverana uses Claude Code and a direct SAP connection to reduce the time needed to understand unfamiliar code from days to minutes, with 2,006 objects documented automatically.
The facts
At
a glance
- 2,006 custom development objects fully inventoried and automatically described in functional terms, every single one
- One day of development for the S/4HANA analysis cockpit, built with Claude Code
- 20 minutes for a complete transaction analysis that would previously have taken an estimated two to three days
- Complete network isolation: access only through a dedicated VPN, a mirrored SAP development system and test data only
- Zero Data Retention: no storage and no training on Laverana data
The starting
point
A mature
custom code based
An SAP team of this size relies on a code base that has grown over time. More than 2,000 custom development objects have accumulated at Laverana over the years: transactions, reports, includes, classes and Web Dynpro applications. Much of it was created by developers who are no longer with the company, and it is documented in the way ABAP code often is: briefly, if at all.
“For a small object, this took two to three hours. For a large program, it could take several days, before the first line had even been changed.” — Thorben Warnebold, Team Lead SAP & Enterprise Systems, Laverana GmbH & Co. KG
Two tasks consumed a particularly large amount of time. The first was troubleshooting across call hierarchies, includes, enhancements and BAdIs. Where-used lists show structure, but not intent. The second was finding suitable function modules in the SAP standard, spread across naming conventions from several decades. The upcoming S/4HANA transformation added custom code analysis, adaptation to the new data model and ATC findings for more than 2,000 objects.
The
decision
Why not simply use
a chat window
The obvious approach would have been to give developers an AI assistant in the browser: paste code in and copy the answer out. That is precisely what Laverana did not want. A model that sees ABAP code only as text knows neither the Data Dictionary nor the customizing, neither the where-used relationships nor the actual data. It can discuss code, but it cannot assess it in the context of the system.
Laverana therefore chose Claude Code with a direct connection to the SAP system through MCP. Claude Code works agentically: it can research independently, call tools and complete multi-step tasks instead of returning one answer to one prompt. Through MCP, it receives read and write access to code as well as SAP content and settings. This changes the task from ‘explain this code’ to ‘check the system.’
The
solution
An isolated AI environment
called adam
adam is the AI data center that Laverana built together with SK. It combines four components in a fully isolated network segment:
- Windows Terminal Server as jump host. The only access point to the environment, available exclusively through a dedicated VPN.
- Debian host with Claude Code. The main working environment for developers, who connect via SSH.
- Custom-built chat application. Provides business users with the same system access through a web interface, including user management, model selection, usage overview, logging and feedback. The application itself was built with Claude Code.
- Mirrored SAP ECC development system. A one-to-one copy of the development environment containing test data only. This makes the MCP server’s full access unproblematic from a data protection perspective.
The
framework
Security and
data protection
From the outset, Laverana was clear that a tool with write access to SAP could not run in an arbitrary cloud environment:
- Network isolation. A dedicated segment accessible only through its own VPN and from there through the Windows Terminal Server as the sole jump host.
- No production data. The mirrored development system contains test data only. Neither customer nor employee data leaves the company.
- Zero Data Retention. Model access through the Anthropic API with Zero Data Retention configured. Prompts and responses are not stored.
- No training on company data. Contractually guaranteed.
- Traceability. Usage, runs and logs are recorded in the chat application and can be reviewed by administrators.
In use
What runs
through adam today
On-demand analyses. Requests that once required a development ticket, an effort estimate and a place in the queue are now submitted and answered directly. The route from a business requirement to an answer has shortened from days to minutes.
Checking tickets against the live system. Incoming tickets are entered directly into the chat application. adam checks them against the SAP system, including the customizing settings actually in place. A model without a system connection can speculate; adam checks.
S/4HANA transformation. This is the most demanding use case and the one that shows the approach most clearly. To assess the custom development portfolio, the team integrated its own S/4HANA analysis cockpit into the chat application. It was built with Claude Code in one day and inventories the entire custom code base:
- Transactions
- 704
- Reports
- 547
- Includes
- 524
- Classes
- 188
- Web Dynpro components
- 17
- Web Dynpro applications
- 15
- Subroutine pools
- 7
- Module pools
- 4
- Total/strong>
- 2006
The objects are distributed across 67 packages. All 2,006 objects have an automatically generated functional description, not the short text from the repository, but a description of what each object does from a business perspective. The cockpit also assigns every object to an end-to-end process such as Plan to Produce, Order to Cash, Warehouse and Shipping, Quality or Product Passport. It prequalifies the scope by identifying and excluding 295 objects behind transactions, 158 third-party software objects and 94 local objects in the $TMP package. Each rule also receives a risk assessment, with a runtime of roughly 30 seconds per rule.
The
result
What adam changes in
day-to-day operations
“We completed a full transaction analysis in 20 minutes today. Based on our experience, this would previously have taken us two to three days.” — Thorben Warnebold, Team Lead SAP & Enterprise Systems
This figure comes from a specific, recent case. The comparison value is an experience-based estimate from the period before adam, not a measured metric. The team is currently expanding its collection of cases.
The tool builds its own tools. Both the chat application and the S/4HANA analysis cockpit were developed with Claude Code, with the cockpit completed in a single day. For a team of four internal employees, this is the real leverage: custom developments that would previously have been impossible because of limited capacity become realistic on this scale.
Documentation as a by-product. The 2,006 automatically generated functional descriptions are not an intermediate result, but a lasting asset. A manual inventory on this scale would have taken months and, based on experience, would never have been completed.
Who benefits most. adam has the greatest effect where system knowledge is missing: developers new to the code base, external colleagues familiarizing themselves with unfamiliar programs, and business users who can ask the system questions through the chat application without knowing ABAP.
The
outlook
The pilot becomes
a working model
- Expanding usage. The role of AI in SAP development at Laverana is growing continuously. In the team’s view, day-to-day work without it is no longer imaginable.
- Rollout to the BPM team. Five additional users will be added next. This takes adam beyond the immediate SAP development team and turns it into a tool for the process side of the organization.
- Expanding the S/4 transformation. The next step is to address the findings, including adaptation to the S/4 data model, ATC findings and Simplification Items.
- Dedicated S/4 development system in the environment. Alongside the mirrored ECC system, the plan is to provide an S/4 system in adam so development and testing can take place directly in the target environment.
- From analysis to implementation. The MCP server already has write access, so the technical path from automated assessment to automated adaptation is open.
The conclusion
The difference lies
in the connection
With adam, Laverana has not created an AI pilot installation, but a working environment: isolated, connected to the real system and used every day. The team is small, but the tasks are not. More than 2,000 custom development objects and an upcoming S/4HANA transformation must be managed. The decisive difference from most AI projects in the SAP environment is the connection. adam does not merely read code; it works within the system.
“Our SAP development is no longer conceivable without AI. Its role will continue to grow, and the next step is a dedicated S/4 system in this environment.” — Thorben Warnebold, Team Lead SAP & Enterprise Systems, Laverana GmbH & Co. KG
SK
view
“adam shows what becomes possible when SAP expertise and infrastructure expertise come from a single source: not a cloud black box, but an environment for which the customer retains responsibility and in which AI works where the data actually resides. That is exactly what matters to us at SK.” — Gürkan Kondu and Stefan Schrott, Managing Directors of SK
As of 3 August 2026. All figures are taken from the pilot operation at Laverana GmbH & Co. KG. Values identified as estimates are based on the team’s experience and were not measured.
Next
step
What would your
success story look like?
Let’s assess where your SAP Basis operations stand today and which operating model can turn a functioning system landscape into one that is reliably predictable.