The Execution Layer That Gets Agents Past the 97% Pilot Wall
Partners build the AI layer on BTP. The hard part is not building agents, it is getting them past the pilot wall and into production. CreateOS is the governed runtime that closes that gap, so agents reach production on BTP, act on Business Data Cloud, and register into the SAP AI Agent Hub.
- Built for BTP (Kyma)
- A2A-compliant
- ISO 27001 compliant
- SOC 2 Type II in progress
~97%
of enterprise agents never reach production, the wall the modernization wave meets next.
~3%
of SAP customers run Joule in production today.
Named gap
liability, auditability, and approval frameworks for agents remain unsolved.
Q3 2026
SAP AI Agent Hub GA. The window to have the production pattern ready is now.
The governed agent runtime
A governed runtime with a full agent lifecycle: intent, generate, build, govern, deploy, observe, heal. Self-healing execution (detect, diagnose, patch, redeploy) keeps agents reliable after go-live. Built to run on BTP (Kyma), act on BDC data products, and register into the SAP AI Agent Hub over A2A.
Cited, not generated
A deterministic core computes and reconciles every figure; the model only narrates over cited findings. The runtime is architecturally incapable of inventing a number. That is the direct answer to the liability and auditability gap.
Complement, never compete
Not a marketplace, the Agent Hub is. Not a foundation, BTP AI Foundation is. Not orchestration, Joule is. CreateOS is the factory and reliability layer that feeds the Hub over A2A and makes the BTP investment more valuable.
Where CreateOS Sits in the SAP Business AI Stack
CreateOS slots into the one layer SAP leaves to partners: execution and reliability. It reads the system of record, it never replaces it.
| Layer | Owned by | CreateOS posture |
|---|---|---|
Context Business Data Cloud, Knowledge Graph, domain models | SAP | Agents consume BDC data products as grounded truth. Read, never replicate; the system of record stays untouched. |
Build Joule Studio, Agent Builder, MCP | SAP + partners | Full interop over MCP and A2A. Partner-built agents deploy onto CreateOS instead of rebuilding plumbing per project. |
Governance catalog SAP AI Agent Hub, GA Q3 2026 | SAP | A2A-compliant. CreateOS agents register into the Hub and run under SAP's discovery, verification, and monetization catalog. |
Execution and reliability The open layer, no named SAP capability | CreateOS | Self-healing runtime. RBAC, PII, and policy enforced at runtime. Pilot to production in weeks. Runs across SAP and non-SAP estates, sovereign or multi-cloud. |
Cited, reconciled answers, behind enterprise auth
For a multi-site industrial operator, CreateOS deployed a governed intelligence layer for finance and business development. It turned 20 years of unstandardized cost reports across 70+ sites into a governed source of truth the teams can simply ask questions of. Finance gets cited, reconciled answers in plain language: the full P&L reconciled line by line, every anomaly ranked by financial impact, the exact site and cost driver behind a negative-margin month found in seconds instead of a week of digging. Business development prices and plans against true site-level cost and margin, not stale extracts. The model narrates; it never computes.
20 years
of data ingested and normalized
70+ sites
75 labels resolved to entities
6 weeks
from pilot to production
One Pattern, Not a Snowflake
Take a manufacturer midway through ECC to S/4HANA. BDC already publishes clean Supplier-Master, Open-PO, and Inventory data products. The pain is acting on them. An agent scores every open PO against supplier health and disruption signals, ranks exposure by revenue impact, and drafts replacements: expedite, dual-source, reschedule. Weeks, not months, because nobody rebuilds deployment, auth, or metering. SAP keeps the system of record and gains stickier BDC consumption.
SAP
System of record and governed BDC data products.
Joule
The conversation. The buyer asks, Joule routes.
SI / Partner
The agent logic, onboarded with SAP teams.
CreateOS
Runtime, governance, observability, per-use metering.
What the Pattern Returns
AI velocity
Weeks-to-production partner agents, without internal build cycles.
Stickier BDC consumption
Every agent makes the data foundation more load-bearing.
Agents feeding the Hub
A2A-registered partner agents in the catalog from GA day one.
A pattern for the base
Repeatable across the modernization base, not a snowflake.
How We Engage
- 01Weeks 0 to 2
Align
Select the lighthouse account and vertical. Co-define the BTP (Kyma) and BDC topology and the A2A path into the Agent Hub.
- 02Weeks 2 to 8
Prove
An industry agent live in production on the named account, running on BTP with BDC data products, governed and observable end to end.
- 03Q3 2026
Scale
Certify, register into the Agent Hub at GA, and make the pattern repeatable across the base.
What we do not claim. No native BTP runtime or pre-built BDC integration today. CreateOS is designed for BTP Kyma, BDC data products, and A2A; the exact topology is co-defined with your teams. CreateOS does not replace Joule, the Generative AI Hub, or the Agent Hub.
Bring one stalled agent to the pilot wall
Pick one modernization account where agents-on-BDC is the bottleneck. We will run it on BTP, govern it, and stand it up in production in weeks.