Lightfield Wrote Its Own CRM. Process Takeaways From the $47M Round

Lightfield announced a $47 million Series A on September 9, 2026, led by Andreessen Horowitz. That is the CRM company. Lightspeed Venture Partners is an investor in the round, not the name on the product. Founders Keith Peiris (CEO) and Henri Liriani had already killed Tome, an AI presentation tool that reached about 25 million users, and written a CRM from customer context instead of fields reps type after the call.

The raise is the peg. The process is the story. They shut a working product, cut a 70-person company to a conference table, traded office space for the first five customers, and built the data layer before the agent demo.

The company wire lists participation from Maverick Capital, Coatue, Audacious, Alumni Ventures, Greylock, and Lightspeed. a16z's Joe Schmidt, Alex Rampell, and Fabrizio Serafini wrote the lead memo the same day.

Kill the product that cannot hold context

Tome was an AI slide tool. a16z says it grew to 25 million users. Upstarts and Forbes put the earlier raise at about $80 million; Peiris told Open Source CEO it was roughly $81 million. Lightspeed, Coatue, and Greylock were already on the cap table. The product still could not write a serious deck.

Peiris's line, in that July interview: they had the growth, the capital, and the team, and still could not make good presentations for people doing serious work. "It's very hard to overcome having the wrong idea." Tome could start a slide. It did not know the customer, the deal, or the last call. That is the same hole every bolted-on AI sales tool falls into.

In March 2025 he posted the pivot on LinkedIn: LLMs are good at starting work and lack the context required to finish it. Tome became Lightfield. The company blog later put the architecture in one sentence: capture the raw traces first. Summaries and fields are derived views, not the source of truth.

"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."

Keith Peiris, Lightfield Series A wire

Shrink until the old story leaves the room

Upstarts says they cut headcount from 70 to 7. Open Source CEO has Peiris at 70 to six. Forbes, writing the July pivot feature, has about 70 to six. The range is the fact. The remaining people were staring at negative momentum.

Peiris told Open Source CEO he called Slack's CEO. The rule he brought back: a team small enough to fit around a conference table, customers every day, talk about those customers until people stop talking about the old thing. Forbes independently reported that he asked Stewart Butterfield for pivot advice and got a two-pizza-team version of the same instruction. Both are Peiris recounting the call. The operational move is not in dispute: they shrank, then talked to customers daily.

Buy the first five customers with the office

A three-month-old CRM is a hard sell. Peiris told Open Source CEO they used the giant office as the trick. Teams that would use the CRM and give them training data could work out of the space. That produced the first five customers. The signal was not NPS. It was when those five stopped complaining.

The next hundred came from proximity. Same interview: he cold-emailed Y Combinator founders, hosted events, and went zero to one with technical founders who were still on a spreadsheet and a call recorder. a16z's memo is the strategy version of that bet. Siebel won client-server. Salesforce won cloud. HubSpot won inbound SMB. Each took a greenfield generation the incumbent could not see. Lightfield is aiming at companies choosing a system of record for the first time, not a 20-year Salesforce rip-and-replace on day one.

Data layer first. Agents after.

The wire and the a16z memo agree on the stack. Connect email, calendar, calls, Slack, LinkedIn. Structure the record as it happens. Build a temporal context graph so agents can see what is true and what changed. Then let agents write follow-ups, score deals, and draft sequences. The founders' November 2025 essay is blunter: automation is a consequence of understanding, not the starting point.

Four pieces, from the wire:

• A record that updates itself. Interactions write the CRM. Reps do not.
• A business world model. Interactions map to people, deals, and accounts, with history, so an agent can reason instead of filling a stage picker.
• An agent harness. SDK, code sandbox, evals. Same request, same quality of work.
• An open interface. Readable and writable over API, MCP, and CLI.

That is the opposite of an "AI CRM" that sells the use case and hopes Salesforce is clean. If the model cannot read the customer, the system of record is wrong.

Forward-deployed engineers, then price the work

Lightfield is not self-serve, Peiris told Upstarts. Implementation is sales-led and consultative. Every engineer rotates as a forward-deployed engineer, paired with someone who has done the consulting work. Target: in and out in 30 days. Learn what the company wants to track, wire the sources, leave a world model they will not rip out.

a16z's pricing line is the one on paper: it "follows the work done and value delivered instead of static seats you buy." That is as specific as the primaries get. Treat seats-plus-credits talk as color unless Lightfield publishes the SKU.

What they can claim on the raise

The wire: more than 5,000 companies signed up since launch in November 2025, from early-stage startups to accounts with hundreds of users. Upstarts, citing Peiris, says about 5,000 companies use the software today, some with more than 100 people on it, and that net dollar retention is 400 percent. That NDR number is the company's. Attribute it.

Upstarts puts the current team at about 40 and says the raise is meant to scale sales and engineering back toward Tome's prior peak. Open Source CEO, in July, had Peiris at a little under 40. a16z says teams are ripping out Salesforce, HubSpot, and Attio. The wire's "signed up" and Peiris's "use" are not the same sentence. Keep them separate.

"Companies that love us the most treat us like their business world model. We are becoming the oracle to the executive team."

Keith Peiris, to Upstarts

The takeaways that travel

1. Wrong idea beats momentum. 25 million users and $80 million did not fix a product that could not hold customer context. More growth made the hole deeper.
2. Context is the product. Horizontal AI tools die on the second use. Vertical systems of record win because they sit on email, calls, tickets, and Slack.
3. Shrink until the old product leaves the room. Peiris used Butterfield's rule as an operating constraint, not a slogan.
4. Trick the first customers into teaching you. Office space for usage and training data. YC cold email for the next hundred. You cannot write a CRM in private.
5. Greenfield first. Win companies that have not industrialized GTM yet. Brownfield comes later.
6. Build the lake, then the agents. Lightfield's own essay: you cannot automate what you do not understand.
7. Engineers sit with the customer. FDE rotation is how a three-month-old system of record earns trust.
8. Price the work if the work is agents. a16z is already writing the post-seat sentence. The SKU is unpublished.

Most teams should not write a CRM. Salesforce is still a $200 billion incumbent, per Upstarts. The graveyard of Salesforce killers is crowded, and Lightfield is walking into it against Clay, Attio, Apollo, Unify, Reevo, and Salesforce's own AI layer.

The teams that should build are the ones whose agents already bounce off the current record. If the model cannot see the customer, and the reps will not type the customer in, you do not have a CRM problem you can buy your way out of. You have a data-model problem. That is what this team bet two years of pain on.

Lightspeed's role is easy to miss if you only watch the name collision. They funded Tome. They stayed in the Series A when the founders cut the company to a conference table and wrote the system they wished they had when Tome was failing. Lightfield is the company. Lightspeed is the firm.

Do not bolt an agent onto a strainer. Write the record the agent can think in. Or do not bother.

Company: lightfield.app. Series A wire: PR Newswire. Lead memo: a16z.