DEPLOY STAGE

The video discusses how to enhance team productivity and growth through connected AI agents, particularly within the Salesforce ecosystem and Slack. Christina Hastings from Salesforce highlights the concept of operating like a team ten times your size by leveraging AI agents across the sales cycle. She emphasizes that top performers often possess context and understanding of customers and products, which AI agents can replicate and scale. The core idea is to implement agents at each stage of the sales cycle, creating a multiplier effect when their functionalities are connected. For prospecting, an agent can identify high-value leads from the same data sources reps use, serving up curated lists weekly. This frees up reps to focus on customer engagement. The crucial multiplier effect occurs when the context gathered by the prospecting agent is shared with the engagement agent, enabling highly contextual and non-intrusive customer interactions. Similarly, account agents can provide contextual briefs by pulling data from CRMs and other tools, going beyond generic AI-generated summaries. A key aspect of this strategy is that these agents must work where reps work. While "headless" AI is a trend, the real value comes from integrating agents into existing workflows. Salesforce has integrated its CRM into Slack, which is presented as the "system of engagement" for sales teams. This allows reps to access crucial information and drive deals forward in a single, familiar platform. A live demo showcases this integration within Slack. A rep's "today view" displays high-signal leads identified by a prospecting agent, highlights, and schedule summaries. To engage these leads, the rep can ask Slackbot to deploy the engagement agent, which will nurture, qualify, answer basic questions, and book meetings. For an upcoming call with Omega, the account team channel in Slack provides a full picture of the deal, including insights from sales agents. The demo then illustrates prepping for the call. By asking Slackbot for a meeting brief, the rep receives a comprehensive account overview, including attendees, discussion points, and relevant context like SOC 2 compliance and data residency when legal or financial stakeholders are present. This pre-meeting preparation, which traditionally involves checking multiple systems, is streamlined to take seconds. Following the call, the demo moves to pipeline management. The rep can update opportunities directly within Slack, including stage, close date, and next steps. However, to address the often tedious task of CRM updates and ensure data richness, Slackbot can engage a pipeline management agent. This agent suggests changes to fields, stages, and next steps, which the rep can review and approve with a single click. These updates then sync back to Salesforce automatically, ensuring an accurate and trustworthy pipeline and forecast. The presentation also touches on sales coaching. Slackbot and Salesforce can provide reps with a clear view of their quota attainment and pipeline. If performance gaps exist, the agent can suggest adjustments and engage in a conversational coaching session. The demo shows a rep on track to meet their quarterly goals, with all data sourced from Salesforce and the CRM for accuracy. The speaker notes that this integration is not just a vision but is actively used by companies like Salesforce itself and customers such as Momentum, where reps haven't logged into Salesforce in months, operating successfully through Slack. Anthropic is cited as another customer experiencing significant success, including 60% faster deal cycles and substantial cost savings, by embracing Slack as their system of engagement. The core message is that the CRM no longer needs to reside solely in Salesforce; bringing it to Slack represents the future of sales, enabling automated funnels, centralized information, and enhanced context through connected agents. The latter part of the transcript introduces Simon from Assistant UI, who discusses building internal AI assistants. He notes the fragmentation caused by each SaaS app having its own AI assistant and advocates for a consolidated, in-house solution. Simon argues for owning the tech stack to gain control, flexibility, and leverage. He explains that proprietary solutions like ChatGPT or Cloud lock users into specific model providers and make data extraction difficult. Building one's own assistant allows for switching model providers, integrating custom workflows, and negotiating better terms. Simon outlines a typical architecture: a UI layer, an agent harness, and connections to AI servers and skills. He demonstrates building an AI assistant using Assistant UI for the frontend, Verse ACK as the harness, and XMCP for building the AI server. The live demo shows bootstrapping a frontend with Assistant UI, adding an XMCP server with a weather widget, and then developing custom tools, such as one to list open GitHub issues. He highlights the ease of debugging and iterating on the server with a self-restarting development server. Assistant UI also offers templates with features like Google Drive integration, attachment libraries, and multiple chat conversations. During the Q&A, Simon addresses querying large wikis or codebases. He explains that for extensive knowledge bases, the approach involves placing data in a file system and using tools like `grep` for searching and navigating, similar to how CodeWhisperer works. He asserts that costs for inference remain low, even with tens of thousands of users, as the system efficiently uses tools like file search and directory listing rather than reading every file. This minimizes the context window usage.