In-house, for SAP teams: 1 prep day + 1 workshop day
Smart use of AI in SAP teams. One real ticket done by 5 pm.
2,400/3,900 EUR excl. VAT
Group average under 8 out of 10: everyone gets the full amount back.
The SAP AI Workshop is a one-day, in-house workshop for SAP development teams of up to 12 people, run on your own ABAP code exported to files and on your own backlog, with no connection from AI into your SAP system. It is built on what cut implementation cycle time by 71% in an 18-person SAP team. Every participant leaves with a setup where Claude Code reads the exported SAP code without copy-paste, and one real ticket finished with AI. For the whole IT team there is a 50-person format in two tracks.
Not for
Who it is for
Marian prepares the SAP specifics together with your SAP lead. Without this day the workshop stays generic and the first hour goes on installs. Many on-premise SAP IT teams allow no AI connection into SAP, and the workshop does not need one. Where your IT approves it, we also test SAP’s ADT MCP server as an optional second route.
With you and your SAP lead: the landscape (ECC, S/4HANA, BW, HANA), Jira or Azure DevOps, which AI licences people have, what may leave your network, and whether your IT allows any AI connection into SAP.
Picked from your backlog. For each one we write down how long that kind of ticket takes today. That is the baseline.
The packages we will work on, exported through abapGit or as Z-code and DDIC structures in a repo, plus a Jira or Azure DevOps export. We check that Claude Code reads them on a developer laptop. Whatever will not work, we know before the day.
A CLAUDE.md with your conventions: Z/Y naming, Clean ABAP, authority checks, banned patterns.
Sent a week before: licences, access, installs.
The program
9:00 to 9:30
One of your tickets solved by pasting into a chat, then again with the exported package and your house rules loaded. The difference shows on your own code.
9:30 to 10:30
IDE with Claude Code over the exported code, house rules in CLAUDE.md, the Jira or Azure DevOps export at hand. The same setup on every laptop. Where your IT approves it, SAP’s ADT MCP server in Eclipse as an optional second route.
10:45 to 12:00
AI reads the exported code and sharpens the spec for the new feature, then scores it against the 8 of 10 gate. Stories, tasks and test cases ready for Jira or Azure DevOps.
12:45 to 14:15
Pairs take their own ticket from spec to code. A bug fixed from an ST22 dump or an SQL trace. ABAP Unit tests for Z-code that never had any. The code goes back into SAP by a person; nothing on the day writes into your system.
14:30 to 15:30
A code review agent checking against Clean ABAP and ATC results. A documentation agent writing the documentation for Confluence. The team keeps both.
15:30 to 16:15
What AI sees in your exports and what it does not. The team writes its own rules: what never goes into AI, what an export may contain (no production data), how AI code gets reviewed. Transports stay with a person.
16:15 to 17:00
Three or four pairs show what worked and what did not. Each person picks one practice for four weeks and we agree the metric.
Whole IT team format, track A1
For the ABAP developers. One shared tooling setup for all of them: IDE with Claude Code, shared DDIC context and conventions, reusable skills. Real old Z-code taken to modern ABAP with ABAP Unit tests and a Clean ABAP check. Then one real spec all the way through: spec check, design with two options, code with tests, a person approving between steps. Outcome: a faster development cycle and predefined tooling across the SAP developers.
Whole IT team format, track A2
For SAP consultants, helpdesk and infrastructure, on a 6-month Jira export. Scores spec quality against the 8 of 10 gate, evaluates the last sprints against plan, builds a sprint morning brief, and traces one helpdesk incident from its log to severity, probable root cause and, in the IDE, the related objects in the exported Z-code and a proposed fix. A person decides. Outcome: 3 agreed metrics for productivity and quality of sprint delivery.
A developer pastes a piece of ABAP into a chat. The model sees one snippet, none of your Z-tables, DDIC structures, includes or naming rules, and writes textbook ABAP that does not fit your system.
Claude Code reads the whole exported package and your house rules, and the same model writes code that compiles against your system. No AI connects into SAP for that.
Each developer pastes snippets into their own chat window, in their own way.
One shared setup on every laptop, a CLAUDE.md with your house rules, and two reusable agents for code review and documentation.
Specs reach the team unscored. In the 18-person team the average spec scored 3.3 out of 10 at the start.
A spec scorer on your Jira or Azure DevOps export. Below 8 out of 10, a spec does not go to the team.
Developers buy AI themselves and run company code through private accounts the company cannot see or control.
Team rules for AI, written by the people who follow them, and a licence option the company owns.
There is no number to say whether AI moved anything.
A baseline measured on the prep day, and the cycle time of that ticket type compared with it four weeks later.
Format
Standard: up to 12 people
2,400 EUR excl. VAT
1 prep day + 1 workshop day. Optional assistant for 13 to 14 people: +1,000 EUR.
Whole IT team: up to 50 people
3,900 EUR excl. VAT
1 prep day + 1 workshop day in two tracks, one lecturer each, run with an assistant.
Optional review after four weeks
+600 EUR excl. VAT
Half a day on the metric, four weeks after the workshop.
The 8/10 rule
Both prices follow one rate card: 1,200 EUR per day for the lead trainer, the assistant on top. Travel billed at cost. Paid before the workshop. Invoiced by Engineering Leaders Academy: Czech VAT at 21% for Czech clients, reverse charge for EU businesses with a VAT ID.
Faculty
You learn from people who did the job. Our lecturers have built multiple unicorn companies and helped organizations scale from startup to scale-up to established corporate. No academic theories, no empty certificates.
Hard to copy
The day runs on your exported packages and on 4 to 6 tickets from your own backlog, picked on the prep day. People work in pairs on their own tickets.
Many on-premise SAP IT teams allow no AI connection into SAP, and the workshop does not need one: abapGit or a repo export is enough. SAP’s ADT MCP server stays an optional second route where your IT approves it. If neither route reads your code reliably on the prep day, GitHub Copilot in Eclipse is the fallback.
Targets sit well below what the 18-person team reached: 30% instead of 71%. If cycle time moves less than 10% by week 4, we look at which step broke: the setup (the agent cannot read your code), the context (house rules missing) or the review (seniors rewrite what AI produced).
The skills, the spec evaluator and the installation guide are one plugin. The ten base skills are open source on GitHub (sap-ai-skills); on the workshop they get your team’s own rules.
“SAP is not the problem and neither is the model. The model gets one snippet.”
Proof
The 71% cannot be split between the spec gate, the skills, the pairing and the incident agent; they started in the same weeks. The team’s own answer is the spec gate.
Sample workshop deck
The deck for a workshop day with an SAP team on NetWeaver 7.40, 73 slides: a seven-step install guide, a five-minute write test, how to split into four groups and work as a mob, one slide on how each AI skill works before the group runs it, the exact prompts step by step, and how the team writes its SDLC guide and onboarding guide at the end.
Open the sample deck →
Questions
Yes. In an independent benchmark of 180 ABAP tasks, Claude Opus 4.6 solved 84% of them after compiler feedback rounds. The weak point is context: a pasted snippet carries none of your Z-tables, DDIC structures or naming rules, so the model writes generic ABAP. Giving it the exported package and your house rules fixes that, with no connection into SAP.
No. The workshop runs on code exported to files, through abapGit or as packages exported to a repo, and on a Jira export. Many on-premise SAP IT teams allow no AI connection into SAP, and none is needed. Where your IT approves it, SAP’s ADT MCP server is an optional second route: SAP’s own connector that lets AI agents search, check, test and activate ABAP objects from Eclipse. It shipped in 2026, most of its tools cost nothing to run, and SAP states support down to release 7.31 SP04, on-premise and cloud, ECC included. From ADT 3.62 it also reads and edits source code; on ADT 3.60 and 3.61 a community extension adds reading.
Because the snippet is all it sees. Your Z-tables, DDIC structures, includes and conventions stay inside the SAP system, so the model guesses and writes textbook ABAP. The workshop replaces pasting with the whole package, exported through abapGit to files and loaded together with your house rules. Where your IT approves it, SAP’s ADT MCP server can read the code live instead.
Yes. The larger format takes up to 50 people, SAP and IT infrastructure or helpdesk together, run by Marian and an assistant. A 75-minute opening for everyone, then two tracks: AI in ABAP development on exported Z-code, and efficiency metrics on a 6-month Jira export. The team leaves with one shared tooling setup for its ABAP developers and 3 agreed metrics. 3,900 EUR excluding VAT, travel at cost.
Partly. BW objects such as aDSOs, transformations and DTPs have no text source and no public API, so AI reads only the ABAP and AMDP routines inside them. HANA calculation views and HDI artifacts are text and XML files in Business Application Studio or VS Code, and AI reads those well. HANA Studio is not recommended: SAP stopped developing it in 2017. SAP GUI has no AI assistant for developers; AI-assisted development runs in Eclipse with ADT, VS Code or BAS.
No. The workshop runs on the licences your team already has, and Claude Pro is enough because it includes Claude Code. For company code we recommend Claude Team, from 20 USD per user per month (claude.com/pricing, October 2026): the company owns the accounts and chats are not used for training by default.
Not by default. On private Claude accounts (Free, Pro, Max), chats are used for model training since 28 August 2025 unless the user switches it off, and are then kept for up to five years. The company cannot see that setting or enforce it, and when someone leaves, the history with your code stays on their private account. ST22 dumps and table extracts can also contain personal data. The minimum, today: switch training off on every private account and keep production data and dumps with personal data out of AI.
An example for a team of 12: going from 5 to 3.5 days saves 1.5 days per ticket. If each developer closes one such ticket every three weeks, that is half a day a week per developer, or 6 days a week for the team. At an internal cost of 400 EUR a developer day (an assumption, your number will differ), that is 2,400 EUR a week, so the workshop pays back in its first week. On the prep day your real baseline replaces the example and we agree the targets together.
Marian Kamenistak. He is not an ABAP developer, and the workshop does not pretend otherwise. The SAP specifics come from your SAP lead on the prep day; Marian brings the AI setup, the agents and the way to make a team use them.
Tell us your SAP landscape, how many people would join (12 developers or the whole IT team) and which AI licences they use today. You get the fit and a date proposal on the call.
Ready for the full operating rebuild? The AI-Native Engineering Manager →
