In November 2025, the two founders who built the presentation tool Tome — the deck-maker that hit 25 million users and raised $81M — quietly killed the product, gave back what was left of the money, and started over.
Six months later, Lightfield, the AI-native CRM they pivoted into, has nearly 2,500 paying companies, including more than 100 startups from recent Y Combinator cohorts. And by the look of SaaStr's recent write-up, it's the first credible attempt to answer the question every revenue leader has been asking since GPT-4: what does CRM look like if you build it for agents instead of salespeople?
The Pitch: Sales Reps Don't Update CRMs, Agents Do
Every traditional CRM — Salesforce, HubSpot, Pipedrive — is built on a brutal assumption: the rep will log it. Calls, notes, next steps, deal stages, all entered by hand. Industry surveys consistently show reps spend roughly 70% of their week on non-selling work, and the bulk of that is data entry into the system that exists to help them sell.
Lightfield's answer is to invert the model. The CRM doesn't wait to be told what happened. It watches.
Connect Gmail or Outlook, your calendar, Slack, your meeting transcription tool, your support inbox, and your product analytics — Lightfield ingests all of it, then autonomously builds your accounts, contacts, opportunities, and relationship histories. No manual entry. No "log this call" guilt loop.
The hook for founders: you connect an inbox, optionally drop in a CSV from your old CRM, and Lightfield reconstructs your pipeline in under five minutes, per the company's product page.
What "Agents Do the Work" Actually Means
The reason Lightfield is interesting — and the reason 2,500 companies paid for it inside six months — isn't the data ingestion. It's what sits on top.
Inside Lightfield you can spin up agents with a single prompt. The current shipping agents include:
- A prospecting agent that finds net-new accounts matching an ICP definition and enriches them
- A follow-up agent that drafts and queues outreach after every meeting
- A deal coach that flags risk signals — silence on a thread, missing stakeholders, slipped close dates — and tells you what to do
- A meeting prep agent that surfaces relevant context the morning of a call: account history, recent emails, related deals, news
These aren't workflows. They're persistent processes that run against the CRM data and act on your behalf, with the work showing up in your queue rather than as an inbox of suggestions to triage.
The architectural bet — and it's the same one Microsoft made with Agent 365 and Salesforce is racing to match — is that the system of record is becoming the system of action. Whoever owns the customer data graph owns the agents that operate on it.
The Founders Already Did the Hard Thing Once
The most interesting fact about Lightfield isn't the product. It's the decision to ship it.
Keith Peiris and Henri Liriani built Tome at Meta, then spun it out in 2020. Tome raised $81M across rounds led by Coatue, Greylock, Lightspeed, 8VC and GV, and grew to 25 million users as an AI-powered presentation tool. By traditional venture metrics it was a success.
By the founders' own metrics, it wasn't. In a candid interview with VentureBeat, Peiris explained that Tome's user growth never converted into the kind of business they wanted to be running long-term. So they wound it down — returning unspent capital to investors — and convinced Dan Rose at Coatue, the original Tome backer, to back the pivot.
That kind of cold-blooded restart is rare. It also signals something investors clearly noticed: founders who will kill 25 million users to build the better product are exactly the founders you want running a category bet against Salesforce.
The Competitive Frame
Lightfield isn't alone in the AI-native CRM lane. Clarify, Attio, folk, and a half-dozen others are all racing toward roughly the same vision — autonomous data capture, agent-driven workflows, natural-language queries instead of saved reports. Salesforce is sprinting to retrofit Agentforce into Sales Cloud. HubSpot launched Breeze.
What separates Lightfield in May 2026 is the velocity of agent adoption. SaaStr called it "the AI app of the week" not because of the data model, but because customers are reporting that the agents are actually moving deals — not just summarizing them.
The skeptic's read: an AI-native CRM in a six-month-old product is one model regression away from breaking a sales team's revenue motion. The optimist's read: the same is true of Salesforce on a bad release day, and Lightfield ships fixes in hours where Salesforce ships them in quarters.
The Bottom Line
The CRM category has been one of enterprise software's most boring incumbents for two decades. Lightfield is the clearest signal yet that AI doesn't just augment CRM — it rewrites the contract between the sales rep and the system. The rep stops logging. The system starts acting. The agents close the loop.
Whether Lightfield specifically wins the category, or whether Salesforce manages the retrofit fast enough to defend it, the architecture is now visible: a CRM is a graph of customer interactions, with agents running on top of it, and humans approving the work.
The 2,500 companies that have already switched are the leading indicator. The Tome founders killed their last product to chase this one. That's worth paying attention to.
