Agentic AI Weekly Roundup: Enterprise Agents Move Into Systems of Record

This week in agentic AI, the strongest signal came from enterprise software rather than model labs. SAP, Coupa, Notion, SailPoint and specialist security startups all pointed in the same direction: AI agents are being designed for the systems where work, data, approvals and accountability already live.

For buyers, that means the evaluation question is changing. It is no longer enough to ask whether an agent can complete a task in a demo. The harder questions are whether it understands business context, respects permissions, leaves an audit trail, works across existing systems and can be governed when it acts at scale.

1. SAP put agents at the centre of the autonomous enterprise

At SAP Sapphire, SAP introduced its Autonomous Enterprise strategy, including SAP Business AI Platform, SAP Autonomous Suite and Joule Work. SAP said the platform will unify SAP Business Technology Platform, SAP Business Data Cloud and SAP Business AI into a governed environment for building, contextualising and governing agents.

The company also said SAP Autonomous Suite will deploy more than 50 domain-specific Joule Assistants across finance, supply chain, procurement, human capital management and customer experience, orchestrating a subset of more than 200 specialised agents. SAP highlighted examples such as an Autonomous Close Assistant for finance and an Industry AI scenario with RWE for offshore wind asset management.

Why it matters: this is a clear systems-of-record signal. Agents are being positioned inside core enterprise processes, not just around them. Buyers should ask how agent decisions map to process ownership, compliance duties and data context. Suppliers need to show where their agents fit into existing ERP, workflow and governance layers.

2. Spend management and workspace tools are becoming agent platforms

Coupa launched Coupa Compose, an Agentic-as-a-Service bundle for building, managing and orchestrating agents across procurement, finance and supply chain. The announcement includes Navi Agent Studio for no-code custom agents, Smart Intake & Orchestration for request handling, and Navi Connect for agent-to-agent and third-party system connectivity.

Notion also launched its Developer Platform, with Workers, database sync, an External Agent API and a CLI. Notion says Workers let teams deploy custom code to a hosted runtime, sync data from systems such as CRMs, support platforms and production databases, and build deterministic tools for agents. It also says governance, permissions and sandboxing are part of the platform from the first deploy.

Why it matters: enterprise agents are not arriving through one channel. They are appearing inside finance suites, productivity platforms, developer tools and departmental applications. Buyers will need a joined-up view of agent sprawl. Suppliers should assume their agents will be expected to interoperate with workplace data, APIs and workflow hubs.

3. Identity, runtime controls and AI policy enforcement moved up the agenda

SailPoint launched Agentic Fabric to help enterprises discover, govern and protect AI agents and other non-human identities. Its announcement focuses on mapping agents to human owners, systems, data and access policies, with discovery, governance and real-time protection across cloud environments, applications and endpoints.

SAP and NVIDIA also expanded their collaboration, with NVIDIA saying SAP is embedding NVIDIA OpenShell into SAP Business AI Platform as a runtime security layer for SAP AI agents and custom agents built in Joule Studio. NVIDIA describes OpenShell as providing isolated execution environments, policy enforcement at filesystem and network layers, and infrastructure-level containment.

Funding activity echoed the same concern. Fortune reported that White Circle, a Paris-based AI control platform, raised $11 million to build real-time policy enforcement between company users and AI models, checking inputs and outputs against organisation-specific rules.

Why it matters: agent governance is becoming a product category. Buyers should expect to inventory agents like identities, assign ownership, limit privileges, monitor actions and enforce policy at runtime. Suppliers that cannot explain containment, escalation, logging and accountability will face longer procurement cycles.

4. Regulation is tightening around transparency, documentation and oversight

The European Commission's AI Act guidance remains an important backdrop for agent deployment. Its official AI Act page reiterates the risk-based structure, including high-risk obligations around risk management, data quality, logging, documentation, information to deployers, human oversight, robustness, cybersecurity and accuracy. The transparency rules are due to come into effect in August 2026.

That matters for agentic AI because autonomous systems can cross boundaries between software categories. A customer-facing assistant, an internal workflow agent and a decision-support tool may create different compliance obligations depending on what they do, what data they use and whether humans can understand or contest the output.

Why it matters: enterprise buyers need an AI inventory and governance model before agents proliferate across teams. Suppliers should prepare evidence packs covering system purpose, data handling, human oversight, logging, security controls and deployment boundaries. Regulation will reward products that can be explained, monitored and controlled.

5. The market is shifting from capability claims to deployment evidence

The most credible announcements this week shared one pattern: they tied agents to defined workflows. SAP spoke about financial close and asset management. Coupa focused on sourcing, bid comparison, risk and sanctions workflows. Notion focused on synced workspace context and custom tools. SailPoint focused on agent identity and access governance.

That is the right direction. Generic agent claims are becoming less useful to enterprise buyers. The market is starting to separate products that can show operational fit from those that only show model fluency.

Why it matters: buyers should ask for reference workflows, success metrics, failure handling and governance controls. Suppliers should lead with use cases, integration depth and measurable outcomes rather than broad claims about autonomy.

The Agentic Expo takeaway

This was a week of enterprise grounding. Agents are moving into ERP, spend management, collaborative workspaces, identity security and runtime policy layers. The opportunity is significant, but the buying criteria are becoming more serious.

For enterprise buyers, the next phase is about choosing agents that can be trusted inside real operating models. For suppliers, the winning story is not simply that an agent can act. It is that it can act with context, controls, accountability and clear commercial value.