Sales teams have been stuck using digital stone tools while engineers got lightsabers, and that gap just became a business problem worth solving.
The Summary
- Von launched as an AI platform that automates model selection and orchestration for sales intelligence, building a "context graph" from CRMs, call recordings, and emails to understand business-specific language
- Built by the team behind process automation startup Rattle, targeting the gap between AI-powered dev tools (Claude, Cursor) and the fragmented "revenue stack" most companies still use
- Positions itself as an intelligence layer rather than point solution, automatically routing queries to the best-fit model (GPT, Claude, Gemini) based on task requirements
The Signal
Engineering workflows changed overnight. Sales workflows barely changed at all. That divergence is the real story here, and Von's approach to multi-model orchestration is a bet that the problem isn't the models themselves, but the context they're missing.
Most enterprise AI tools treat sales data like a search problem: you ask a question, the LLM retrieves an answer. Von CEO Sahil Aggarwal says that approach fails because sales organizations speak in dialects. Your "Stage 3" deal means something different than mine. Your territory definitions, handoff protocols, and deal review criteria are institutional knowledge that generic models can't access.
"Once Von builds this context graph, it will understand your business better than anyone else in the company."
Von starts by building what they call a context graph: structured CRM data from Salesforce or HubSpot combined with unstructured sources like Gong call recordings, Zoom transcripts, email threads, and internal docs. The system learns the company's ontology, the actual language of how deals move and revenue gets generated. This matters because the best AI model for summarizing a sales call is different from the best one for forecasting pipeline risk or drafting a follow-up email based on six months of negotiation history.
The platform doesn't lock you into one model. It automatically routes tasks:
- GPT-4 for complex reasoning about deal structure
- Claude for long-context analysis of email threads and call transcripts
- Gemini for multimodal tasks that span documents and conversations
This is what orchestration looks like when it's built for a specific job rather than sold as general-purpose infrastructure. The question isn't "which model is best" but "which model is best for this task, given this company's specific context." That shift from model selection to automated task routing is the actual product innovation.
The timing matters. Developer tools got smart fast because code has structure. It compiles or it doesn't. Sales data is messy, political, and wrapped in jargon that changes company to company. The gap between what AI can do for engineers and what it can do for sales teams isn't a technical problem anymore. It's a context problem. Von is betting that solving context is worth more than adding another chatbot to the revenue stack.
The Implication
If Von's approach works, the playbook for enterprise AI shifts from "buy the best model" to "build the best context graph and let the models compete for each task." That's a different kind of moat. It also means sales organizations that feed their specific ontology into these systems early will have an advantage that compounds over time.
Watch for how quickly Von can onboard complex enterprise customers and whether the context graph actually improves forecast accuracy or just produces better-sounding summaries. The difference between those two outcomes determines whether this is infrastructure or just another dashboard.