Presented by SAP
Generating code with AI is fast, but getting that code to run reliably within a large enterprise, integrate with active systems, be regulated for compliance, and maintainable for years requires critical work that most organizations underestimate.
While 81% of all organizations have a detailed strategy, only 12% to 16% achieve AI-powered executionsays Michael Ameling, CPO of SAP Business Technology Platform, and the reasons rarely come down to the quality of the generated code.
"Across industries, companies that have invested heavily in AI tools are hitting a wall when the generated code fits the reality of their existing environments, because generating code and putting it to work are not the same problem." says Ameling.
There are specific requirements for deploying AI-generated logic at enterprise scale: what the data and preparation for integration actually look like, how governance works as AI agents move from producing recommendations to executing workflows, and how development teams are changing their role as AI takes on more of the coding work.
Why AI code generation fails in enterprise production environments
The productivity gains from AI code generation are real and well-documented, but the ease of prototyping has given many organizations a misleading idea of how advanced they really are.
"Generating code is one thing," says Ameling. "Enterprise customers, including multinationals and large organizations, must ensure that compliance and security are not compromised. Code that runs reliably for ten or twenty years, as it does for many of SAP’s largest customers, also needs to be maintained, patched, and understood by whoever inherits it. In other words, lifecycle management does not happen on its own."
The problem is rarely the quality of the generation. Teams build something cool and then discover they lack access to the data it depends on, or the integrations it involves, or the permissions needed to run it in a real environment. The problem is essentially that AI amplifies the maturity of an organization’s existing processes and data, but it cannot replace them.
This dynamic intensifies as AI moves from producing code to executing actions. Latency, cost, and system load increase when logic runs continuously against live data instead of generating a single output. The performance requirements of an autonomous agent operating in the transaction systems of a multinational are categorically different from those of a co-pilot developer.
How to connect AI-generated logic to fragmented business systems
The architectural challenge that most enterprise AI projects underestimate is integration. Real enterprise environments are not a clean slate: they combine cloud systems, legacy on-premises infrastructure, fragmented data warehouses, and dozens of enterprise applications that were never designed to communicate with each other. Getting AI-generated logic to work reliably across all of them requires a layer that unifies data access, process context, and governance, and must be in place before any agent starts running. And organizations that see AI as a reason to postpone infrastructure modernization are making a mistake.
"The question is not whether to modernize or not. Of course you need to modernize," says Ameling. "But the value you get on top of this is much greater with AI. Federated data access and harmonized process layers are not alternatives to upgrading a fragmented landscape, they are what make the upgrade worthwhile."
At the platform level, this translates into a set of practical requirements: structured data integration, end-to-end process visibility, and the ability to discover and connect to APIs in both modern and legacy systems. SAP’s approach with the Business AI Platform relies on tools including its Joule Studio enterprise architecture layer, Integration Suite, Business Data Cloud, and SAP AI Agent Hub to provide that context. The goal is to give AI-generated logic accurate, current knowledge of what a company does and how, rather than simply accessing raw data.
AI agents handle large challenges by breaking them down into smaller autonomous tasks, where each agent is responsible for a specific domain and all coordinated toward a shared outcome. A financial close, for example, involves dozens of discrete subprocesses. Agents that handle each task in parallel, within defined constraints, can dramatically compress cycle times, but only if the underlying systems they interact with are coherent and accessible.
The governance and oversight that AI agents require in production
When AI moves from assistant to operational actor, governance issues become increasingly important, because agents that trigger workflows, update records, and interact with live business systems need the same accountability framework that applies to human employees, that is, identities, defined privileges, and auditable behavior.
There are two different models:
Parent propagation, where an agent acts on behalf of a user and inherits the permissions and scope of that user.
System-enabled agents, where the agent operates under their own identity and role-defined privileges, functioning more like an automated HR role than a personal assistant.
Both models require the same underlying infrastructure: an agent hub where operators can see what agents exist, what APIs they can access, and what they are authorized to do. Observability must also be properly operationalized for AI, combined with technical and business assessments.
"In production, openness is very important," says Ameling. "We use OpenTelemetry as a framework, so we can integrate with other solutions, for end-to-end tool observability, third-party agents, and the like."
On top of that, standard technical evaluations, which test whether an agent produces consistent results, are necessary but not sufficient. Business evaluations evaluate whether an agent is actually moving the performance indicators it was implemented to improve, but it has to work from one end to the other.
Where testing is performed is equally important. The traditional software development cycle in development, test, and production environments is disrupted when a model produces different results depending on whether it is run on test data or live data. Getting to trustworthy AI in production means accepting that validation is fundamentally different from what engineering teams have practiced for decades, with testing in live environments, including A/B/C testing to ensure results are reliable.
How AI-based code generation is changing the roles of software engineering
The role of the promoter does not disappear in this environment, but its center of gravity is changing. The productivity multiplier is significant when developers can run multiple coding agents in parallel on open terminals, each working on a separate problem and each taking several minutes to complete. But it introduces a new kind of cognitive demand, because humans have to stay informed. That means tracking context across concurrent workflows, evaluating results that span large code bases, and making architectural judgments that no agent can be trusted to make alone.
"The more specific and complete the prompt, the less intervention is required, and developers are learning that providing more context up front pays dividends by reducing trade-offs." says Ameling. "But it is still necessary to understand the result, not just accept it."
The competitive advantage will continue to be intellectual property, not tools. The companies that will come out ahead will be those that most effectively encode their domain knowledge into the systems they build.
"The process expertise of a manufacturer, the risk logic of a financial institution, the routing intelligence of a logistics company – these are the assets that AI can accelerate, but only if the organizations that own them do the work to make them accessible and usable." says Ameling. "Protect it and apply AI to accelerate your differentiation."
Sponsored articles are content produced by a company that pays to publish or has a business relationship with VentureBeat, and are always clearly marked. For more information, contact sales@venturebeat.com.
