a16z just bet $47M that Salesforce is the wrong shape...
a16z just bet $47M that Salesforce is the wrong shape for AI agents
Last updated 20h ago•9sources
Lightfield closed a $47M Series A led by a16z, and the framing around it is more interesting than the funding number.
The core argument: every AI agent workflow built on top of Salesforce or HubSpot is sitting on a foundation that was never designed for agents. Those systems were built for humans to manually log things, search things, and hand things off by talking to each other. When you drop an AI agent into that stack, it inherits all the same context gaps that made human handoffs painful. The meeting prep agent doesn’t know what the follow-up agent promised yesterday. You’ve automated the work and kept the problem.
Lightfield’s answer is to rebuild the record layer from scratch. Every customer interaction — calls, emails, meetings, Slack threads — gets pulled into a continuously updating model of that customer. The key detail is the interface: every record is readable and writable through API, MCP, and CLI. That’s not a CRM with an API bolted on. It’s a data layer that agents can actually work from without a human in the middle translating.
The reaction in developer circles has been mostly positive, with a few people flagging that this is a bigger idea than it looks. One take that’s circulating: “If AI agents are going to make business decisions, they need an accurate model of the business first.” That’s the real pitch — not better CRM features, but a world model for the business that agents can read and write against.
The skeptic’s version of this story is that it’s a well-funded bet on a specific architecture (MCP, agent-native records) that hasn’t proven out at scale yet. The optimist’s version is that Salesforce’s $300/seat pricing and $100k+ admin overhead is a sitting target, and the first CRM that agents can actually use without human babysitting wins the whole market.
Co-founder Keith Peiris is positioning CRM as the beachhead, with the obvious implication that a continuously-updating business world model doesn’t have to stop at customer records. Whether that expansion happens or whether this stays a niche tool for agent-heavy sales teams is the open question.
Community take
8
The discussion is waiting for a spark. Share your take and get it started.
Deep dive
Questions this post answers
What is Lightfield and how is it different from Salesforce Agentforce or HubSpot Breeze?
Lightfield is a CRM built from scratch for AI agents rather than humans, having raised a $47M Series A led by a16z. Unlike Salesforce's Agentforce or HubSpot's Breeze, which layer agents on top of human-designed data models, Lightfield continuously builds a customer context model from calls, emails, and meetings, and exposes every record via API, MCP, and CLI for agent access. Developers weighing agent-native tools versus retrofitted incumbents can follow CRM and AI agent shifts on daily.dev.
Why do AI agents underperform when added on top of existing CRMs like Salesforce?
Agents underperform because they are fighting an underlying data model designed for humans, not machines, so they lack enough structured context to reason over a full customer relationship. Lightfield's pitch is that context itself is the product: without a reasoning-ready history of interactions, an agent is effectively just autocomplete wrapped around a CRM login. Teams debating whether to bolt AI onto legacy systems or adopt agent-native tools can track this reasoning on daily.dev.