Lightfield's $47M: your CRM record is the agent ceiling

Lightfield announced a $47 million Series A led by Andreessen Horowitz on September 9, and the pitch is not "CRM with AI bolted on." It's that your CRM record is the thing holding your agents back. We agree with the diagnosis more than we agree with the remedy, and the distinction matters if you're about to spend money on either.

What actually happened

Per the company's announcement, a16z led with Maverick Capital, Coatue, Greylock, Lightspeed, Audacious, and Alumni Ventures participating. More than 5,000 companies have signed up since the product launched in November 2025.

The architecture is the interesting part. Lightfield ingests email, calls, calendar, Slack, and LinkedIn, then assembles a self-updating record and connects those interactions to people, deals, and accounts over time — what it calls a business world model. On top of that sits a standardized agent SDK with quality monitoring, and the platform is reachable by API, MCP, and CLI. CEO Keith Peiris frames the problem bluntly: agents don't fail because models are weak, they fail because the data is incomplete, inaccurate, and unstructured.

Worth knowing who's building it. Peiris and cofounder Henri Liriani previously built Tome, the presentation app VentureBeat reports reached roughly 20 million users before they walked away from it to start over. Valuation figures circulating in secondary-market coverage aren't confirmed by the company, so we're leaving them out.

Why your CRM record is the real constraint

Run this test on your own business. Pick a customer who bought something in the last 90 days. Can you reconstruct — without asking a person — the first touch, every objection raised, what was promised on the call, and which support ticket came before the upgrade?

If the answer is no, no agent is going to fix it. It will confidently make things up in the gaps, because that's what a language model does with missing context. The people selling you agents rarely lead with this.

Three things follow:

Structure beats volume. Dumping transcripts into a vector store is not a customer record. A record has entities and relationships — this person, at this account, on this deal, said this thing on this date. Retrieval over a pile of text gives you plausible answers. A structured record gives you correct ones.

MCP and CLI access is the part to copy. Lightfield exposing the record over MCP is the right instinct regardless of which vendor you pick. If your customer data can only be reached through one company's UI, every agent you ever build is downstream of that company's roadmap. Ask any CRM vendor for programmatic read access to the full record before you sign.

Migration cost is the thing you're actually buying. Five thousand signups in ten months is real traction and also a reminder: agent-native CRMs are a new category, and new categories consolidate. Get your export path in writing on day one.

We'd rather see a small company fix its record-keeping in the system it already owns than migrate to a fifth CRM in six years. The ceiling on agent quality is the data underneath, and that ceiling moves whether or not you change vendors.

Sources: Lightfield announcement, VentureBeat.