Presented by SAP
Building AI code is fast, but getting that code to run reliably across a large enterprise, integrated with live systems, managed for compliance, and maintainable over years requires groundwork that most organizations underestimate.
While 81% of all organizations have a detailed strategy, only 12-16% achieve AI-driven executionAnd the reasons rarely come down to the quality of the code being generated, says Michael Ameling, CPO of SAP’s SAP Business Technology Platform.
"Enterprises across all industries that have invested heavily in AI tools run into a wall when the generated code meets the realities of their existing environment, because creating code and running it are not the same problem." Ameling says.
There are specific requirements for deploying AI-generated logic at enterprise scale: what data and integration readiness actually look like, how governance works when AI agents move from making recommendations to executing workflows, and how development teams change their roles as AI takes over more of the coding work.
Why AI code generation fails in enterprise production environments
The productivity gains from building AI code are real and well-documented, but the ease of prototyping has given many organizations a false sense of how far they really are.
"Generating code is one thing," Ameling says. "Enterprise customers, including multinationals and large organizations, need to be assured that there is no compromise in compliance or security. Code that works reliably for ten or twenty years must be protected, patched, and understood by whoever inherits it, as is the case with many of SAP’s largest customers. Lifecycle management, in other words, does not create itself."
The problem is rarely generation quality. Teams build something compelling, then discover they don’t have access to the data it depends on, the integrations it accepts, or the permissions it requires to run in a real environment. The problem is that AI augments an organization’s existing data and process maturity, but it cannot replace it.
This dynamic is amplified as AI moves from producing code to performing actions. Latency, cost, and system overhead increase when logic runs continuously against live data rather than displaying a one-time output. The performance requirements of an autonomous agent operating in the transactional systems of a multinational company differ dramatically from those of a developer copilot.
How to connect AI-generated logic to fragmented enterprise systems
An architectural challenge that most enterprise AI projects underestimate is integration. True enterprise environments are not clean slates: they combine cloud systems, legacy on-premises infrastructure, fragmented data warehouses, and dozens of business applications that were never designed to talk to each other. A layer that integrates data access, process context, and control is required for AI-generated logic to work reliably across all of them, and it must be in place before any agent can begin executing. And organizations that see AI as a reason to delay infrastructure modernization are wrong.
"It is not about modernizing or not. Of course, it is necessary to modernize," Ameling says. "But on top of that, the value you get is much higher with AI. Federated data access and aligned process layers are not alternatives to improving a fragmented landscape, they are what make the improvement worthwhile."
At the platform level, this translates into a number of practical requirements: structured data integration, process completion, and the ability to discover and connect to APIs in both modern and legacy systems. SAP’s approach with the Business AI Platform uses tools including Joule Studio, Integration Suite, Business Data Cloud, and the SAP AI Agent Hub enterprise architecture layer to provide this context. The goal is to give AI-generated logic accurate, current knowledge of what the business is doing and how it’s doing it, rather than only having access to raw data.
Artificial intelligence agents solve large problems by breaking them down into smaller, autonomous tasks, each agent responsible for a specific area and all coordinated towards a common outcome. For example, financial closing involves dozens of discrete sub-processes. Agents running each task in parallel within specified constraints can dramatically compress cycle times, but only if the underlying systems with which they interact are connected and accessible.
The management and control that AI agents require in production
As AI moves from assistant to operational actor, governance questions loom large, as agents running workflows, updating records, and interacting with live business systems need the same accountability framework as human workers—namely, identities, defined privileges, and auditable behavior.
There are two different models:
Basic propagation, where an agent acts on behalf of a user, inherits that user’s permissions and scope.
System-activated agents, where the agent operates under their own identity and role-defined privileges, act more like an automated HR role than a personal assistant.
Both models require the same basic infrastructure: an agent hub where operators can see what agents are available, what APIs they can access, and what they’re allowed to do. Observability must be properly harnessed for AI, along with both technical and business assessments.
"Transparency in production is very important," Ameling says. "We use OpenTelemetry as a framework, so we can integrate with other solutions for comprehensive monitoring of the tool, third-party agents, and more."
Moreover, standard technical evaluations that check whether an agent produces consistent results are necessary but not sufficient. Business evaluations evaluate whether an agent is actually moving the performance indicators it has set out to improve, but it must work from end to end.
Equally important is where the test takes place. The traditional software development cycle across development, test, and production environments breaks down when a model produces different results depending on whether it runs against test data or live data. Getting reliable AI into production means accepting that validation looks fundamentally different from what engineering teams have practiced for decades, live environment testing, even A/B/C testing to ensure the results are valid.
How AI-powered code generation is changing software engineering roles
The role of the developer does not disappear in this environment, but its center of gravity changes. The productivity multiplier is significant when developers can run multiple coding agents in parallel between open terminals, each working on a separate problem and each taking minutes to complete. But this introduces a new cognitive demand because people need to stay in the loop. This means keeping track of context between parallel workflows, evaluating variable results across large codebases, and making architectural judgments that no single agent can rely on alone.
"The more specific and complete, the less intervention is required, and developers are learning that bringing more context up front pays dividends in reduced rollback." Ameling says. "But the output still needs to be understood, not accepted."
Competitive advantage will remain intellectual property, not tools. The companies that thrive will be the ones that most effectively code domain knowledge into the systems they build.
"A manufacturer’s process expertise, a financial institution’s risk logic, a logistics firm’s routing intelligence, these are assets that AI can accelerate, but only if the organizations that hold them do the work to make them accessible and usable." Ameling says. "Protect it and apply AI to accelerate your differentiation."
Sponsored articles are content produced by a company that paid for the post or has a business relationship with VentureBeat and is always clearly marked. Contact for more information sales@venturebeat.com.
