Lightfield: The $300M Startup That Killed Its Own 25M-User Product to Build a CRM

Most founders would kill for 25 million users. Keith Peiris and Henri Liriani already had them — and walked away.

Their company Tome, the AI-powered presentation tool backed by $81 million from Coatue, Greylock, Lightspeed, and GV at a $300 million valuation, was one of the fastest productivity tools to reach a million users. It launched publicly in March 2022 and hit 25 million users by mid-2024. Eric Schmidt, the former Google CEO, was among the angel investors. By any startup metric, Tome was a success.

Then the founders shut it down and started building a CRM.

Why Abandon a Viral Product?

Peiris spelled it out in a public post: “We started to build Tome in 2020. 5 years and 25M users later, it’s become clear to us that LLMs are good at starting work, but lack the context required to finish it.”

The realization came through customer pilots with marketing and sales teams. The real pain point wasn’t generating presentations — it was the mess underneath. Sales reps were drowning in fragmented data spread across emails, Slack, calendars, meeting transcripts, and support tickets. No tool was stitching it all together.

Peiris and Liriani — both ex-Meta, where they built products used by billions (Peiris managed products across Facebook, Instagram, and Oculus; Liriani led Project Lightspeed, which cut Messenger’s codebase by 84%) — saw an opening. They cut the team from 70 to 28 engineers, spent a year in stealth, and built Lightfield from scratch.

Dan Rose at Coatue, who led the original Tome investment, backed the pivot.

What Lightfield Actually Does

Peiris describes the CRM category as “perhaps the most complex and lowest satisfaction piece of software on Earth.” Lightfield’s bet is that the entire concept of manual CRM data entry is obsolete.

Here’s how it works: connect your inbox, calendar, Slack, meeting tools, and support platforms. Lightfield ingests everything — emails, call transcripts, meeting notes, tickets — and builds a living record of every customer relationship. No dropdown menus. No custom fields to configure. No logging calls after they happen.

The key architectural difference from legacy CRMs: Lightfield stores the complete, unstructured record of what customers actually say and do. Traditional CRMs like Salesforce and HubSpot force interactions through predefined schemas — you pick from dropdowns, fill in fields, check boxes. That means most of the actual context gets lost the moment someone closes the “log activity” dialog.

Lightfield calls this “lossless customer memory.” When you change your data model — say you want to start tracking a new field — it backfills historical data up to two years, because the raw conversations were always there.

On the practical side, the system preps you before meetings (surfacing relevant history from past interactions), summarizes conversations, drafts follow-ups, suggests pipeline updates through human-in-the-loop analysis, and answers natural language questions about your business with citations back to the original conversations.

Code Execution: The Feature That Sets It Apart

In February 2026, Lightfield shipped something unusual for a CRM: code execution. The AI agent can now write and execute Python programs with full access to your CRM data.

This isn’t a gimmick. The agent plans an approach, writes a Python script, runs it in a sandbox, and evaluates the results. When queries need deeper analysis, it dispatches sub-agents to read individual account histories and interpret context.

Why Python instead of just LLM outputs? Three reasons:

Scale: Python’s data libraries process thousands of records quickly, getting around LLM context limitations
Determinism: Code produces identical outputs every time, unlike variable LLM interpretations
Output quality: Structured reports, visualizations, and dashboards — not just text tables

Real use cases include competitive battle cards generated from competitor mentions across all customer interactions, sales scorecards that grade reps on discovery quality with specific conversation examples, and feature loop closure that tracks requests through Linear and Notion and auto-drafts follow-ups when engineering ships the feature.

How Lightfield Compares to the Competition

The CRM market hit $73.4 billion in 2024 and is projected to reach $163.2 billion by 2030. Lightfield isn’t just competing with other startups — it’s going after Salesforce ($34.9 billion in revenue) and HubSpot.

Lightfield vs. Salesforce/HubSpot: The legacy players require rigid schemas defined upfront. They’re retrofitting AI onto decades-old architecture. Lightfield argues that starting with unstructured data and applying structure on demand is fundamentally better for LLM-powered automation. The pitch to customers: replace 10+ enterprise tools costing $3,000–$5,000 per seat with one platform.

Lightfield vs. Attio: Attio is the closest competitor in philosophy — an AI-native CRM with a customizable data model targeting startup founders. But Attio optimizes for control and customization (Plus tier at ~$36/user/month, Pro at ~$86/user/month), while Lightfield optimizes for automation and delegation. Different bets on what users actually want.

Lightfield vs. Clay: Clay focuses on prospecting — aggregating data from 100+ external sources to identify and qualify leads (Pro plan at $800/month for 50,000 credits). Lightfield focuses on maintaining context across the entire customer lifecycle from your own interactions. They solve different problems and could theoretically coexist in the same stack.

Lightfield vs. People.ai: People.ai ($200M raised) targets enterprise sales analytics. It’s more established but narrower in scope, focused on activity capture and revenue intelligence rather than being a full CRM replacement.

Traction and the Product Hunt Moment

On March 18, 2026, Lightfield hit #2 on Product Hunt with 497 upvotes. The same week, SaaStr named it AI App of the Week, and VentureBeat covered the pivot story prominently.

The early numbers are interesting: 100+ companies using it daily, with over half spending more than an hour per day in the system. For a CRM — a category where low adoption has been the norm — that engagement metric stands out.

Pricing follows a usage-based model starting at $40 per seller per month, with additional charges for AI inference time on automations, email generation, and analytics queries. Peiris has emphasized broad data access across the organization with consumption-based charges for heavy AI users — a deliberate contrast to per-seat models that discourage adoption.

Tens of thousands are reportedly on the waitlist.

The Bigger Question

Lightfield’s pivot is a case study in a bet that keeps getting made across the AI industry: that the LLM era will produce entirely new categories of software, not just AI features bolted onto existing products.

The risk is real. Salesforce isn’t standing still — it’s pouring resources into AI features. HubSpot is doing the same. And the AI-native CRM space is getting crowded fast. Whether the market ultimately wants a unified AI platform or specialized tools remains an open question.

But the founders’ willingness to kill a 25-million-user product suggests they’re not optimizing for comfort. Whether that conviction translates into a product that cracks the CRM market wide open is the question worth watching.

FAQ

How much does Lightfield cost?
Lightfield uses usage-based pricing starting at $40 per seller per month. Additional charges apply for AI inference time spent on automations, email generation, and analytics queries. This model is designed to encourage broad adoption within organizations rather than limiting seats.

What integrations does Lightfield support?
Lightfield connects to email, calendar, Slack, meeting platforms (for transcripts), support ticket systems, and analytics tools. It also integrates with project management tools like Linear and Notion for feature tracking workflows.

How is Lightfield different from Salesforce or HubSpot?
The fundamental difference is architectural. Salesforce and HubSpot require you to define data schemas upfront and manually log interactions. Lightfield ingests raw, unstructured data from all customer touchpoints and applies structure on demand. This means changing your data model doesn’t require re-entering historical data — it backfills automatically from existing conversations.

Who is behind Lightfield?
CEO Keith Peiris and CPO Henri Liriani, both ex-Meta product leaders who previously built Tome (25M users, $81M raised, $300M valuation). The company has 28 employees and is backed by Coatue, Greylock, Lightspeed, 8VC, and GV.

Can Lightfield replace my existing CRM?
That’s the intent. Lightfield is specifically targeting companies before they adopt traditional CRMs, as well as teams frustrated with legacy systems. You can import data from existing CRMs via CSV upload and have everything recreated in under five minutes, according to the company.

Top AI Product

Leave a comment Cancel reply