Operationalizing AI in workflows: Lee Spacagna, Solutions Engineer, OpenAI

Key Concepts

• Workspace Agents: AI systems designed to perform delegated, multi-step tasks by interacting with existing productivity tools (email, calendar, etc.) rather than just answering queries.
• Agent Builder: A platform feature allowing users to create, customize, and deploy agents using natural language without requiring coding skills.
• Skills: Reusable snippets of instructions that capture "tribal knowledge" and specific workflows, turning them into repeatable processes.
• Frontier: An enterprise-grade platform for managing, governing, and scaling thousands of AI agents across an organization.
• Contextual Integration: The ability for agents to pull data from disparate silos like SharePoint, Salesforce, Outlook, and Teams to inform decision-making.

---

1. The AI Adoption Framework

The speaker, Lee, identifies three distinct layers of AI adoption within financial services:

• Bottom-Up (Chatbots/Codex): Individual employees using AI to assist with daily tasks.
• Top-Down (AI Systems): Major, transformational initiatives for new products and operational support.
• The Middle Layer (Workspace Agents): The focus of the presentation, designed to automate workflows at the team and department level, closing the gap between individual productivity and enterprise-wide transformation.

2. Step-by-Step: Building a "Chief of Staff" Agent

The demo illustrates how a business user can build a sophisticated agent without technical expertise:

1. Initialization: Select a template (e.g., "Chief of Staff") from the Agent interface.
2. Tool Integration: Connect existing enterprise tools (Outlook, Teams, SharePoint, Salesforce) to provide the agent with necessary data access.
3. Natural Language Configuration: Use an "agent-to-build-an-agent" approach where the user describes requirements (e.g., "Run at 9:00 a.m. daily") in plain English.
4. Skill Assignment: Attach specific "Skills" (e.g., meeting prep formatting) to define how the agent should structure outputs.
5. Deployment & Testing: Validate the agent via starter prompts and deploy it to team channels (e.g., Microsoft Teams).

3. Real-World Applications in Financial Services

Beyond the "Chief of Staff" example, the speaker highlights several high-impact use cases for agents:

• KYC (Know Your Customer) Onboarding: Automating the collection and verification of client data.
• AML (Anti-Money Laundering) Investigations: Streamlining the research and reporting process for suspicious activity.
• Relationship Management: Proactively gathering CRM and communication context to prepare advisors for client meetings.

4. Key Arguments and Perspectives

• Delegation vs. Interaction: The core value proposition is moving from "asking questions" to "delegating meaningful work." Agents are designed to complete tasks from start to finish.
• Democratization of Automation: By using natural language to build agents, the barrier to entry is lowered, allowing non-technical staff to automate their own departmental workflows.
• Capturing Tribal Knowledge: "Skills" serve as a repository for institutional knowledge that is otherwise trapped in individual employees' heads, ensuring consistency across the team.
• Governance at Scale: The introduction of the Frontier platform addresses the challenge of managing thousands of agents, ensuring they operate within a secure, governed environment while continuously learning from performance data.

5. Notable Quotes

• "When we say agents, we mean AI systems that we can delegate meaningful tasks to, not just ask questions of."
• "Skills are a way of capturing snippets of information instructions to perform critical tasks... turning those into repeatable workflows."
• "The opportunity here isn't about one single automation project. It's actually about a brand new operating model."

6. Synthesis and Conclusion

The presentation marks a shift in AI strategy from simple conversational interfaces to autonomous, integrated "co-workers." By leveraging the Agent Builder and Frontier platform, financial institutions can bridge the gap between individual AI usage and enterprise-wide automation. The ultimate goal is to reclaim employee time—as demonstrated by the speaker’s own experience of saving an hour each morning—by allowing agents to handle the synthesis of emails, meeting prep, and data cross-referencing, thereby enabling staff to focus on higher-value strategic work.