a16z is putting $47m into a CRM that builds itself from your email

Every sales team keeps a database that nobody fills in. That is most of the history of customer relationship management software, and it is why the oldest category in business software is also the most resented one.

Andreessen Horowitz is betting $47 million that agents change that. On 9 September 2026 the firm said it was leading a Series A in Lightfield, an AI-native CRM built by Keith Peiris and Henri Liriani, the two founders of the presentations app Tome. Alex Konrad reported the round first, and a16z's Joe Schmidt, Alex Rampell and Fabri Sera confirmed it about an hour later.

The product claim is a single sentence in the announcement. Lightfield offers "a data model that needs no configuration, builds itself from your emails and calls in minutes, and fits your business rather than the other way around." Agents then do the work that piles up around the record, which a16z lists as the updates, the follow-ups, the handoffs and the reporting. Every record is reachable through an API, through MCP (the open protocol an agent uses to call a tool) and through a command line.

The category is boring, and that is the point

a16z's own framing is the honest one. "Every company runs on a CRM, and almost no one likes the one they have." The reason is mechanical rather than mysterious. A CRM is a database whose value depends on a person typing into it after the call.

The person does not type into it after the call.

So agents move that constraint from both ends at once. They can do the typing, and they cannot do anything useful without a record that lives somewhere other than the account owner's head. We made a version of this argument about a different product in OpenAI's Data agent borrows the one thing that makes it work.

Rohan Paul wrote the test he cares about, and it is probably the right one: "can someone take over an account without asking the previous owner to explain everything? I'd ask the same question about an AI agent."

So which part of this is actually new?

Not the ambition. Zero configuration is the promise every buyer in this category has been sold before, and we'd treat "builds itself" as the claim to be tested rather than the feature that was shipped. What has genuinely changed is the reader. Turning a year of call transcripts into a schema seems plausible now in a way it was not three years ago, because the thing doing the reading got good.

The strongest case against the round is that the record was never the hard half. Automate the updates, leave the handoff untouched, and you have the same handoff problem you had before, running faster. But exposing every record over MCP and a CLI suggests Peiris knows this, and is designing for a customer that is not a person.

Then there is the adoption number, which is doing a lot of work for a company this young. Peiris told Konrad that 5,000 businesses have tried Lightfield. a16z's wording is softer, that "thousands of companies have adopted" it since a launch late last year, and a trial is roughly the cheapest thing a software company can count.

Neither the valuation nor the total raised was disclosed. Tome took venture money as a slides company before the pivot, and the a16z announcement does not mention Tome anywhere. (Founders are entitled to bury a previous product. It is still worth noticing which ones a new investor leaves out of the press release.)

Who should be uncomfortable? HubSpot at the low end, and everyone selling an "AI features" upgrade to tenants of an existing CRM, because a record that assembles itself is a different product from a record with a chatbot bolted to the side. We'd expect Lightfield to raise again above a billion-dollar valuation before the end of 2027, and Salesforce to ship a zero-configuration object mode first.

The figure that settles this is not 5,000. It is how many of those records an agent is still writing to next September, with nobody checking its work.