This week, we look at how AI agents are moving into enterprise workflows, why software is starting to behave differently, and how connectors and infrastructure determine whether these systems can actually work inside companies.

AI in Action

The Next Enterprise Software Battle [Enterprise/Agentic AI]

Anthropic recently launched finance-focused AI agents for banks and insurers, with use cases familiar to anyone working in financial services: drafting pitchbooks, summarizing earnings calls, preparing credit memos, reviewing audits, flagging suspicious activity, and supporting compliance workflows. Major firms including Goldman Sachs, Citi, Visa, and AIG are already using Anthropic’s technology in some capacity. Enterprise AI is starting to move beyond answering questions and into coordinating work itself.

Traditional SaaS software still depends heavily on employees moving information between systems. An analyst pulls data, copies it into PowerPoint, drafts commentary, checks compliance language, and routes materials for approval. AI agents change how that work gets coordinated. A user can ask the system to identify at-risk clients, summarize recent fund flows, draft talking points, and prepare a first-pass presentation using information pulled across multiple tools and databases. The software starts handling more of the movement between steps, while employees shift toward reviewing, directing, and approving the work. Enterprise software itself may gradually shift toward systems employees manage and review rather than directly operate step by step.

The operational reality is far more complicated than the demos. Enterprise workflows involve different data structures, approval chains, permissions, legacy systems, compliance rules, and institutional knowledge that often sits outside formal documentation. One firm’s process for preparing a PM deck or underwriting memo can look completely different from another’s. Agents rely on connectors, APIs, retrieval systems, and orchestration layers to access enterprise tools while respecting permissions and governance policies. Without those connections, the system has very little context about the organization it operates inside. Most firms also do not have the infrastructure or operational capability to host large-scale AI systems independently, which keeps cloud ecosystems like Microsoft Azure, AWS, and Google Cloud deeply embedded in the enterprise AI stack. Many companies are also discovering that deploying AI agents means confronting years of fragmented systems and undocumented workflows.

Reliability and governance are becoming just as important as model capability. A finance agent can draft a credit memo while still missing a risk buried inside an attachment or outdated spreadsheet. A compliance workflow can flag suspicious activity while still generating false positives that require human judgment. Enterprises increasingly need to know what data the agent accessed, how decisions are traced, who reviews the output, and where accountability sits when something goes wrong. Enterprise software is starting to evolve from something employees operate into something that increasingly operates alongside them.

Bits of Brilliance

AI Agents are only as useful as the systems they can reach [Agentic AI/Enterprise]

AI agents often look far more capable in demos than they do inside real companies. Real enterprise environments are messy. Client data might live in Salesforce, documents in SharePoint, approvals inside email chains, workflows in ServiceNow, and important context buried in spreadsheets someone has maintained for years. An AI agent asked to “summarize our biggest client risks” can only work with the systems and information it is actually connected to.

That is where connectors come in. Connectors give AI systems a way to move through the same tools, documents, and workflows employees already use every day. They allow agents to retrieve documents, search databases, trigger workflows, and coordinate actions across systems. In enterprise environments, many connectors also inherit existing user permissions, which means the agent can only access the information an employee is already authorized to see. Once those connections are in place, the AI starts operating inside the company’s actual workflow environment rather than responding to isolated prompts. This is one reason companies like Microsoft, Google, and Amazon sit at the center of the enterprise AI race. Many organizations already run their workflows, permissions, and infrastructure there.

Every organization also operates differently. One firm’s process for preparing a PM deck, approving marketing material, or reviewing a compliance issue can look completely different from another’s. Data structures, naming conventions, approval chains, and legacy systems vary everywhere. In many cases, the AI is revealing operational problems that already existed inside the organization. The work quickly becomes integration, governance, workflow design, and permissions management alongside the model itself. Deploying enterprise AI increasingly involves redesigning how information and decisions move through the organization.

Even highly connected agents still operate with constraints. They can retrieve outdated files, miss context buried inside attachments, surface conflicting information, or generate outputs that still require human judgment. Governance, auditability, and review remain central because enterprise AI operates inside real accountability structures. The model provides the intelligence. The connections determine whether the system can operate inside the organization at all.

_{Byte-Sized Intelligence is a personal newsletter created for educational and informational purposes only. The content reflects the personal views of the author and does not represent the opinions of any employer or affiliated organization. This publication does not offer financial, investment, legal, or professional advice. Any references to tools, technologies, or companies are for illustrative purposes only and do not constitute endorsements. Readers should independently verify any information before acting on it. All AI-generated content or tool usage should be approached critically. Always apply human judgment and discretion when using or interpreting AI outputs.}