What This Session Is About
Patrick Thompson walked through Clarify.ai’s evolution from a customer data layer into an autonomous CRM. The goal is to reduce the work of maintaining a sales system: gathering customer context, preparing for calls, updating records, and drafting follow-ups. Email, calls, Slack, and other connected sources supply the context agents need to act.
The session connected product design with engineering and distribution. Patrick explained the move from reactive chat to proactive agents, the work of exposing CRM capabilities through Model Context Protocol (MCP), and why a connector should lead directly into a usable workspace. He also demonstrated meeting preparation, customer scoring, artifacts, and activity histories inside Clarify.
How Clarify Is Rethinking CRM
Patrick described a shift from waiting for a user to ask a question toward agents that observe customer activity and suggest the next action. Meeting preparation, drafted follow-ups, and timely nudges bring the CRM into the work a salesperson is already doing. The challenge is making those suggestions useful enough to avoid becoming another source of notifications.
Clarify exposes CRM capabilities through MCP so external agents can query data, update records, and carry out workflows. Patrick emphasized purpose-built tools, self-describing schemas, and skills that explain how to complete multi-step tasks. A feedback tool captures agent intent and errors that would otherwise be difficult for the team to observe.
Patrick described an onboarding gap: a user could authenticate a connector without reaching a working product. Automatically creating a workspace and sending a welcome email turned that interaction into an entry point. His recommendation was to measure agent reads, writes, and updates as part of the acquisition funnel, alongside conventional website traffic.
Patrick described moving away from per-seat fees toward usage tied to inference credits. He also discussed outcome-based pricing as a longer-term possibility. On adoption, he explained how internal tooling assesses legacy data and uses AI to shorten migrations from existing CRMs, reducing the effort required to switch.
Key Insights
- 01Customer context determines how useful an agent can be. Patrick linked improvements in autonomous actions to connecting more sources of customer information. The product work includes bringing email, calls, support interactions, and structured records together so the agent can reason about the relationship.
- 02Proactive agents need feedback from everyday work. Patrick described using edits to drafted emails as a signal for improving future suggestions. The interaction gives the team evidence about what users accept, change, or find unhelpful when an agent takes the initiative.
- 03Tool design and instructions belong together. An API alone does not explain how to complete a sales workflow. Patrick described combining MCP tools with skills that guide external agents through complex tasks, while guardrails and descriptive schemas constrain how those tools are used.
- 04Agent errors need an explicit feedback channel. When an external agent cannot complete an action, the application may see only a failed call. Clarify’s feedback tool is intended to capture what the agent was trying to do, giving engineers more context for debugging and improving the integration.
- 05Connector onboarding is part of the growth funnel. Patrick described marketplace integrations as a way to reach users inside their existing AI tools. Creating a workspace on authentication and measuring subsequent tool activity helps connect discovery to actual product use.
- 06AI changes both switching costs and pricing assumptions. Patrick discussed using AI-assisted migration to reduce setup work, usage-based billing to reflect inference costs, and outcome-based pricing as a possible future model. These were presented as product and business-model choices alongside the technical architecture.
From the Q&A
How do you keep proactive nudges from becoming noise?
Patrick acknowledged that unwanted suggestions can undermine the experience. He described learning from user behavior, including edits to pre-drafted emails, to improve what the agent suggests over time.
How do you understand what went wrong inside an external agent?
Patrick described adding a feedback tool to the MCP integration. It lets an agent report its intent and reasoning around an error, giving the team visibility beyond the raw tool call.
What changed when connectors started driving growth?
Patrick pointed to an onboarding fix: automatically creating a workspace and sending a welcome email when someone authenticated through a connector. That gave new users a route into the product instead of ending the experience at authorization.
How much work is involved in moving from another CRM?
Patrick said internal tooling assesses the quality of legacy data and helps migrate records from systems such as Salesforce and HubSpot. He described examples of reducing migrations from weeks to hours, while noting that some customers choose to start fresh.
Where does pricing go when agents do more of the work?
Patrick described Clarify’s usage-based approach around inference credits and discussed outcome-based pricing as a longer-term direction. Charging a share of a completed business outcome was an example of that future model, rather than a claim that every workflow is priced that way today.
Can AI replace the best salespeople?
Patrick emphasized the human side of relationship selling. His framing was that agents should handle administrative work, follow-ups, and coaching so people can spend more time on the relationships and judgment that drive a sale.