Attio launches hybrid agentic architecture to let AI agents work inside CRM data with full guardrails

Attio is rebuilding how AI agents interact with CRM data, betting that the next wave of enterprise go-to-market won't look much like the last one.

Nicolas Sharp, Attio's founder, describes the launch as a new hybrid architecture that lets AI agents operate directly on CRM data while enforcing the governance and policy controls that enterprise buyers expect. The pitch is that raw access to a database is easy to achieve but commercially unusable without guardrails around customer data. Attio's architecture tries to deliver both: unrestricted querying and analysis on one side, policy enforcement and audit controls on the other.

The system builds on universal context, a feature Attio shipped earlier this year that creates a full 360-degree view of unstructured data in a vector store, giving agents a single surface to search across fragmented customer information. Sharp says agents running on top of this can generate sophisticated queries that would previously have required manual analyst work.

“We have launched our new architecture, hybrid architecture, for agentic revenue. It's all about allowing agents to be unleashed on your data while maintaining all of the guardrails you'd expect from business applications... The best go to market teams are not just simply replicating the SDR model. They're rethinking the customer flows entirely.”

What the best GTM teams are doing

Sharp's read on leading go-to-market teams is that they are not rebuilding old workflows with AI grafted on. They are rethinking customer flows from scratch, rather than producing what he calls a "skeuomorphic" version of the SDR-to-AE handoff model. In practice, that means doing substantially more research per customer, eliminating the handoffs where context typically gets lost, and delivering a high-touch experience at a scale that was previously reserved for the highest-value accounts only. For product-led growth companies, it means arriving at every customer call with a complete picture rather than a cold open.

Where model progress matters most

Sharp is more interested in decision models than reasoning models for this work. Much of the pipeline, he argues, involves making large volumes of discrete decisions before any unstructured output is needed. Access to probabilities from those models opens up workflow designs that aren't possible with standard LLMs, and he sees the next order-of-magnitude efficiency gains coming from that layer rather than from frontier reasoning models.

Attio uses its own product as the primary test environment. Sharp describes the team as "patient zero," running day-zero releases internally before customers see them, with the result that the internal version typically runs about a month ahead of what's in production for paying users.