Organizations rarely struggle to build AI models. They struggle to scale them.

Across industries, organizations continue to invest heavily in data and AI, yet many initiatives never move beyond pilot projects. The challenge isn't creating better models. It's maintaining the infrastructure that surrounds them.

Data pipelines, governance, business rules and user interfaces are deeply interconnected. A change to one layer often requires teams to revisit everything else, slowing deployment and increasing operational complexity.

During the SAS Models roadmap session at SAS Innovate, one theme emerged repeatedly: scaling requires more than better models. It requires rebuilding the surrounding AI stack with modular, industry-focused components that reduce complexity and make AI easier to operationalize.

The real barrier to scaling AI

Across industries, organizations continue to invest heavily in data and AI, but many still struggle to move beyond pilot projects. The issue isn’t the quality of the models, it’s the complexity of the systems around them.

Data pipelines, business rules, governance and user interfaces are tightly linked. When one layer changes, whether from new technology, regulatory shifts, or organizational updates, everything else must be revalidated and rebuilt. This cycle slows progress and keeps teams from focusing on delivering outcomes.

Instead of patching up the same brittle stack, SAS is transforming the approach by isolating problems and scaling AI one focused workload at a time.

Through the SAS Models program, organizations can deploy preconfigured, industry-focused models and agents that integrate with existing environments, rather than rebuilding solutions from scratch. By treating AI as a collection of modular components, teams can adapt individual capabilities as needs evolve without disrupting the broader system.

Why AI initiatives stall

AI projects rarely stall because organizations can’t build models. More often, they stall because the surrounding architecture becomes difficult to maintain and scale.

In regulated industries in particular, every change triggers additional layers of validation and approval. This creates a cycle where:

• Teams spend more time maintaining systems than innovating.
• Experiments rarely make it into production.
• Time-to-value stretches from weeks into months, or worse.

The result is a gap between what’s technically possible and what organizations can realistically operationalize.

Why composability matters

One of the biggest challenges in enterprise AI is maintaining new capabilities over time. As business needs evolve, organizations find themselves having to revisit entire AI systems just to update a single component.

This is why composability is becoming an increasingly important design principle.

Rather than rebuilding an entire system when something changes, organizations can adjust individual components without disrupting everything else.

The SAS Models roadmap reflects this shift. Instead of requiring organizations to build everything from the ground up, SAS is delivering prebuilt, industry-focused components that integrate with existing workflows. This modular approach helps organizations to update individual capabilities as requirements change, making AI easier to adapt, govern and scale over time.

Redefining what AI agents can do

One of the most compelling ideas from the session was about how AI agents are defined.

Rather than positioning them as simple assistants or chatbot interfaces, AI agents were framed as actors that can take action and collaborate with humans.

The session captured this idea directly:

Agents should be thought of as accountable coworkers designed to act, explain themselves, and operate within governance while keeping humans in the loop and compressing cycle time. Jim Georges, Principal Data Scientist, SAS

This shift reflects an important distinction for enterprise AI. These agents don't simply answer questions—they execute tasks, remain auditable and collaborate with people within clearly defined governance and access controls.

That level of accountability is essential in regulated industries where transparency and trust cannot be compromised.

Removing one of AI’s biggest bottlenecks

Data integration has historically been one of the most time-consuming parts of any analytics or AI initiative. Tasks like preparing data, mapping schemas and aligning datasets to specific model requirements often require significant manual effort and specialized expertise.

In many organizations, that work can take weeks or even months before teams are ready to generate insights.

The SAS Data Mapper agent reframes that process entirely. Instead of relying on manual transformation and mapping, the agent interprets source data structures, aligns them with the required target schema, and automatically generates the necessary code and documentation. As demonstrated in the session, work that traditionally took months can now be completed in minutes.

This kind of acceleration has a measurable impact on how organizations operate:

• Teams can move from raw data to usable outputs much faster.
• Iterations become easier and more frequent.
• New data sources or structural changes can be incorporated without restarting the process.

More broadly, it shifts data integration from a slow, sequential step to a dynamic, adaptable process, removing one of the biggest bottlenecks between data and decision-making.

A practical example in supply chain planning

The session provided a concrete example of how these concepts come together in the SAS Supply Chain Agent.

In traditional supply chain planning, teams work through a series of interconnected decisions: what to produce, where to allocate inventory and how to fulfill demand. This process often relies on manual coordination across multiple roles and typically runs in cycles that can take several weeks. [

With an agent-driven approach:

• Different aspects of the workflow (demand, supply, inventory) can be handled in parallel.
• Teams can interact with agents in real time.
• Scenarios and trade-offs can be evaluated dynamically.

Instead of waiting weeks between planning cycles, teams can revisit decisions in days, improving both responsiveness and business outcomes.

The takeaway for leaders

This particular session points to a broader shift in how AI is built and used in enterprises.

Success is becoming less about building better models and more about creating systems that are modular, governed and designed for operational use. Organizations are increasingly looking to embed trusted intelligence directly into workflows rather than treating AI as a standalone capability.

By combining composable models with accountable AI agents, SAS is helping organizations reduce the complexity of the gap between insight and execution. The organizations that scale AI successfully won't necessarily build the most models – they'll build the infrastructure that allows those models to evolve without having to rebuild everything around them.

Where to go next

To explore these ideas further:

• The SAS Models Roadmap session provides a detailed view of the strategy, architecture, and future direction.
• The Supply Chain Agent demonstrations offer a closer look at how agent-driven workflows operate in practice.