$47M · Lightfield + 2 more

What they're building

Lightfield is a CRM designed so that the primary user is an agent, not a sales rep. The core idea is what Keith Peiris calls a business world model: instead of storing outcomes in fields that humans forget to fill in, the record updates itself from every call, email, meeting, Slack thread and calendar invite, and holds enough context that a model can reason from it rather than just query it. Salesforce stores what happened. Lightfield tries to store why. That architectural choice is the whole pitch. Peiris's argument is that agents can already do most of the go-to-market work you used to hire for, but the ceiling on their output is what they understand about your business, and a twenty-year-old schema built for humans to type into gives them almost nothing. So the agent isn't a copilot bolted onto a CRM; the record itself was rebuilt for the agent to operate. In practice that means delegating real work: generating pipeline by pattern-matching against closed-won deals, flagging deals at risk, responding to customers autonomously, ramping new reps on how the best deals were actually won. Peiris runs it on himself, simulating what happens if Lightfield hires 30 early-career reps instead of 10 and which customers that wins or loses. It is deliberately not self-serve. Every engineer rotates onto forward-deployed work with customers, paired with an ex-consultant, with a 30-day in-and-out target. That is a hiring signal as much as a product one.

Why this matters

5,000 companies signed up since the November 2025 launch, and net dollar retention sits at 400% per the company. Customers include Macroscope, where CEO Kayvon Beykpour (sold Periscope to Twitter) replaced Clay with it, and Weaver, whose CEO Peter Grant ran Salesforce UK through its IPO and describes himself as someone who 'used to bleed Salesforce.' Those are unusually credentialed people to switch on. The hard part is that this is the most contested category in software. Salesforce is still a $200B+ market cap, and the startup lane is crowded with Attio, Reevo, Apollo, Unify, Day AI, Monaco, and Clay, which just raised at a $7.1B valuation. Peiris named exactly one company he isn't going after: Sierra. Everything else customer-facing he considers fair game. That is either the correct level of ambition or the thing that kills the focus, and the next eighteen months decide which. Worth noting what Lightfield is not: it does not train its own models. Peiris is blunt that '$47 million is not enough for that.' This is infrastructure that helps a frontier model understand and simulate a business.