Architecting the Agentic Enterprise
At the Sierra Ventures 20th Annual CXO Summit, Salesforce COO Vivienne Wei and Glean Co-Founder Tony Gentilcore explored what it really takes for large organizations to move from basic AI tools to fully agentic systems that handle meaningful work across the enterprise.
The Enterprise Gap
The contrast between consumer AI and enterprise AI is growing. In our daily lives, voice assistance and content generation feel effortless. Inside companies, leaders face a maze of copilots, bolt-on features, and standalone tools that do not connect to core workflows. There is a long list of processes where agents could add significant value. Still, progress is often slowed by one foundational issue: accessing and governing data in a way that is secure, reliable, and usable.
From Productivity Gains to Workflow Automation
Vivienne described how the first wave of enterprise AI focused on individual productivity. That shifted when teams began applying agents to full workflows rather than isolated tasks. The push to adopt agentic systems internally created urgency, collaboration between engineering and support, and a willingness to remove friction. The lesson was not about the tool itself but about the speed that is possible when teams align around a clear outcome.
Data, Context, and the Hard Part of Enterprise AI
Glean tackled the other side of the challenge. For agents to be helpful, they need an accurate view of how a company works. Most organizations lack a unified understanding of their own knowledge. Documents are scattered, permissions are inconsistent, and outdated content sits alongside mission-critical material.
Tony noted that companies make progress when they connect their structured and unstructured data, understand what is current versus deprecated, and put the right permissioning and governance controls in place. Context matters as much as content. Without it, agents produce generic or incorrect answers even when the underlying model is strong.
Making Enterprise Data Useful
Enterprises can build a more coherent data foundation by pulling together data from systems such as CRM, chat, productivity suites, and analytics tools. A semantic layer helps interpret that data within the company’s own language and definitions. Once the foundation is established, activation becomes the unlock. Natural language queries, visualizations, and embedded agent actions help AI fit into the existing workflow rather than forcing people to change their habits.
Driving Adoption Without Disruption
Both speakers emphasized that the most challenging part of deploying agents is not the technology. It is getting people to try new ways of working. Teams respond best when the experience feels familiar and when early examples come from peers rather than mandates.
New Roles for an Agentic World
Agentic systems require small cross-functional teams that blend technical skill with domain expertise. These teams define agents' behavior, determine which data sources matter, and partner closely with business owners who understand the desired outcomes. This collaborative model helps enterprises move quickly while staying grounded in real workflows.
Building for Flexibility
Enterprises should expect a mix of approaches. Some agents will support critical processes and require tight oversight. Others will be created by business users experimenting with new ideas. What matters is that companies maintain flexibility, avoid lock-in, and build a data layer that can support multiple models and tools as the landscape evolves.
The Road Ahead
Agentic systems are beginning to reshape how work is done within large organizations. The shift will accelerate as enterprises combine strong data foundations with thoughtful change management. The leaders who make the most progress will be the ones who focus on context, culture, and clear business outcomes rather than on individual tools.
The agentic enterprise is no longer a concept. It is taking shape now, and the organizations that prepare their data, workflows, and teams will be the ones best positioned to capture its value.
