Lightfield Raises $47M Series A Led by a16z for an AI-Native Customer System

The team that built Tome, the AI presentation tool that racked up more than 20 million users and then quietly stopped being the point, has resurfaced with a far less glamorous ambition: replacing the CRM.

Lightfield announced on September 9 that it has raised a $47 million Series A led by Andreessen Horowitz, with participation from Maverick Capital, Coatue, Greylock, Lightspeed Venture Partners, Audacious and Alumni Ventures. The San Francisco company — still legally incorporated as Magical Tome Inc. — is building what it calls an AI-native customer system: a record of every customer relationship that assembles itself from the raw exhaust of business communication rather than waiting for a salesperson to type it in.

The pitch is blunt about what it is attacking. Salesforce and HubSpot, Lightfield argues, encoded four decades of assumptions about how humans log work: fields, stages, close dates, notes written from memory an hour after the call. Bolt an AI agent onto that substrate and the agent inherits every gap in it.

“Agents don’t fail because the models aren’t capable. They fail because the data they work with is incomplete, inaccurate, and missing the structure needed for comprehension,” said Keith Peiris, cofounder and chief executive of Lightfield. “Lightfield builds a world model of the business from every customer interaction, so every person and every agent works from the same understanding of what’s true and what happens next.”

From presentations to pipeline

Peiris and cofounder Henri Liriani, both former Meta product leaders, started the company in 2020. Its first act was Tome, an AI-native storytelling and presentation product that grew fast — north of 20 million users by most counts — and raised an $81 million Series B in 2023 at a reported $300 million valuation from Lightspeed, Coatue and Greylock, several of whom returned for this round.

Tome was a consumer-velocity product in an enterprise-money market. The pivot to CRM, which the company shipped in November 2025, traded viral signups for something harder to win and much harder to lose: the system of record. Since launch, Lightfield says more than 5,000 companies have signed up, ranging from seed-stage startups to organizations running hundreds of seats. Dealroom, citing company figures, has reported net dollar retention of roughly 400 percent — the signature of a product that lands in one team and metastasizes across a company. Peiris has said dozens of customers ripped out Salesforce to make the switch.

Total capital raised across the company’s life now exceeds $100 million.

What is actually different

Lightfield describes four architectural bets. The first is a record that updates itself: the system ingests email, calendar events, calls, Slack messages and LinkedIn activity as they happen and writes them into the record without human intervention. The second is the world model — every interaction is bound to the people, deals and accounts it touches, and the history of how each changed is preserved, so an agent can reason about trajectory rather than reading a snapshot of fields.

The third is an agent harness. Agents at Lightfield do not free-associate against a database; they operate through a standardized SDK, run analysis inside a code sandbox, and are graded by evals meant to hold output quality steady run over run. The fourth is openness: every record is readable and writable through API, CLI and the Model Context Protocol, and the company says anything a person can do in the product, an agent can do too.

Why this matters

The AI-native SaaS thesis has spent two years being asserted and rarely proven. Most of what shipped in 2024 and 2025 was AI features grafted onto existing systems — a summarize button here, a draft-email panel there — because incumbents had distribution and could afford to move slowly. Lightfield is the sharpest test yet of the opposite claim: that the underlying data model, not the interface, is what agents are actually blocked on, and that fixing it requires starting over.

Andreessen Horowitz is betting the platform-shift analogy holds. “Every platform shift produces a new system of record. Salesforce defined it for the cloud era, and Lightfield is defining it for the agent era,” said Alex Rampell, a general partner at the firm. “Companies building with agents need more than a CRM with AI features. They need a system designed from the ground up for agents to work from.”

His colleague Joe Schmidt IV, a partner at a16z, framed the traction rather than the architecture as the hard part. “Building a system of record is one of the hardest things to do in software, because you have to earn a customer’s trust before they’ll hand you the data that runs their business,” he said. “This team earned it with thousands of companies in under a year.”

The counterargument is familiar and not weak. CRM displacement has been declared imminent roughly once per decade, and Salesforce, with more than $37 billion in annual revenue and its own Agentforce push, has enormous latitude to re-platform under its customers rather than lose them. Continuous ingestion of email, calls and Slack also drags a startup straight into procurement conversations about data residency, retention and consent that a presentation tool never had to survive.

What to watch

Three things. Whether that 400 percent net dollar retention survives contact with contract renewals in a full annual cycle, rather than reflecting a young cohort expanding off a small base. Whether Lightfield can move upmarket past the scaling-startup segment into companies with compliance teams and Salesforce contracts that renew in three-year increments. And whether the incumbents respond by opening their own data models to MCP and agent SDKs — because if Salesforce makes its record fully agent-legible, Lightfield’s clearest differentiator narrows to speed of ingestion rather than kind of system.

The $47 million buys about a year to find out.