Lightfield banks $47M on rebuilding the CRM record so agents can read it

A record that learns only from the interactions it sits inside begins the day it is switched on ^[5]. The half-filled fields and stale close dates already sitting in the incumbent do not travel. They get migrated as the same rows an agent could not parse before, or they get left behind, and either way the world model's history starts at install. That sets the honest rollout sequence for anyone buying this: new segments and new deals first, and no expectation that last year's pipeline becomes legible to an agent because the schema underneath it changed.

The traction figure needs the same care. More than 5,000 companies have signed up since the November launch ^[8], and the company's own count of Salesforce replacements is "dozens" ^[9]. Read dozens at a generous top end of 99 and that is under 2 percent of the sign-up base ^[13]. Sign-up is also the softest number a vendor has: it is not paid seats, not deployed teams, and not teams still running their pipeline there two quarters later. SiliconANGLE's report carries no pricing, revenue, seat, retention or usage-depth figures ^[10], and both available write-ups of the round are versions of the same article ^[11]. The claim that schema is what breaks agents is testable, but not from this evidence.

The fourth design choice cuts both ways. Lightfield is readable and writable through APIs, command line interfaces and the Model Context Protocol ^[7]. MCP is also the layer that lets a buyer leave Salesforce where it is and expose it to agents through a translation tier, which is the cheaper answer for any company whose forecasts and integrations already point at the incumbent. The round funds one side of the rewrite-or-wrap question. It does not resolve it.

Two axes decide it for a given team. First, whether the agent work you want needs history nobody ever wrote down, or only what happens from installation forward. Second, whether standing ingestion of mailboxes, calendars and Slack clears your legal and IT review. If you need the old history and cannot clear ingestion, you wrap: your ceiling is whatever the incumbent's fields already hold. If you need only forward-looking work and can clear ingestion, a rewrite is testable on one segment inside a quarter, with the incumbent still running everything else. Where you need the history and can also clear ingestion, run both and accept you are paying twice while the new record fills. And if you need neither, the problem you have is not a CRM problem at all.

For the team running that one-segment test, the measure at the end of the quarter is how often an agent's proposed next step survives a rep's review without correction, compared against the same reps working from the old record. That is the specific promise the new schema makes. Checking it is cheap.