The AI-Native CRM

1. Key Themes

Intelligence Is Greater Than Schema

Lightfield's core architectural bet is that rigid, pre-defined data structures are the wrong foundation for a modern CRM. Instead of forcing companies to define fields and stages upfront, Lightfield built a chronological "activity log" of every interaction and lets intelligence infer structure after the fact. Keith Peiris explained the design philosophy directly: "We had this idea that's, well, it's funny, we spoke to a lot of CRM consultants... we found that the biggest, most consequential decision they help you with is your data model... If you get the wrong stages, the wrong fields, you can't get the reps to go back in time and fill it out, it's over" 00:14:59. This led to a "schema-less setup where you log in, you connect your email and more... it just assembles for you in real time. Basically, like, intelligence is greater than schema" 00:15:45.

The Pivot Discipline: Kill the Product You Don't Love, Even at 25 Million Users

Peiris walked away from Tome despite explosive growth (2 million monthly users, inference-constrained demand) because the team lacked conviction in the product's ceiling. "Deep down, I think at an instinctual level, none of us like the product... I think at the end of the day, if you're a founder, you have to love the product that you're building" 00:03:19. The technical reasoning: "the model just didn't have enough context to really understand the presenter, the audience, the relationship between the presenter and the audience. And no amount of general reasoning gets you past that" 00:04:43.

Follow the Heat, Not the Roadmap

Rather than pre-planning the pivot, the Tome team let customer behavior guide them from presentations to research to CRM. "We're like, sure, let's just follow the heat, see what happens. I think what makes early stage founders good at this is that you sort of have no priors" 00:06:24. This same instinct—minimal upfront theory, maximal responsiveness—shows up again in how Lightfield discovered its wedge: "It took us, honestly, like six months of having customers... staring at them to sort of, for that to emerge" 00:22:18.

Negative Pricing and Willingness to Be Uncomfortably Early

To get anyone to touch a four-month-old CRM, the team gave away free office space in exchange for usage. "We sort of looked at the only asset we had left, which was this giant office space we couldn't get rid of... I'm going to post on X on LinkedIn. You can sit in our office space if you use our CRM" 00:08:22. The signal they were chasing wasn't revenue but engagement intensity: "they were mad about everything missing... they were in it every day and they were giving us Slack feedback about it every two hours" 00:08:22.

Greenfield-to-Brownfield as a Deliberate Sequencing Strategy

Alex Rampell frames the company's go-to-market motion as intentionally starting in greenfield (new companies with no incumbent CRM) to build proof points before assaulting entrenched Brownfield accounts. "We had this perspective that... building a revenue team is a very rich problem set... we like, we believe there's something in here, but we weren't sure exactly what it was. So we sort of figured, let's just try to win a new company first" 00:20:17. The eventual Brownfield wedge wasn't task automation but a category the team believes incumbents can't easily copy: "our team always had this perspective that that wasn't really the wedge to do Brownfield... I think the wedge in Brownfield has to do with like, you know, better understanding your company" 00:21:38.

Pricing as Empirical Discovery, Not Design

Lightfield tested both pricing extremes and let usage data reveal the right hybrid model. "We started with pure seat pricing... but then the head just, the head was using 10,000x more than the tail... So then we tried purest consumption pricing... we found that nobody touched anything... It was the worst three weeks of the company's life" 00:32:19. They landed on a segmented model: core CRM work is covered by a platform/seat fee, while pipeline generation and workflow automation are consumption-priced, because "there's just an expectation that you pay for those" 00:34:18 when there's clear ROI.

Company-Wide Distribution as a Retention Moat

Lightfield gives the product away free to every employee at a customer company—not just sales—deliberately building organizational lock-in. "It'll create sort of real company network effects that make it harder to rip and replace us" 00:24:23. This also solves the classic Greenfield/Brownfield tension where a newly-hired VP of Sales insists on the tool they already know: "the rest of the company is like, well, hold on... this is how customer success does account scoring. Can you try harder to figure this out?" 00:24:52.

Organizational Design for the AI-Native Era: No Swim Lanes

Lightfield deliberately avoids functional silos, treating the Tome experience as a cautionary tale. "We had a lot of people playing house... they all have their swim lanes and they get really mad if someone gives them feedback... it was impossible to pivot, right? Because you had all of these like appendages that weren't talking to the brains" 00:36:45. The replacement system: "everyone shows up to the same stand up every morning. We stack rank the most important problems... and whoever's free just takes them" with "continuous planning" where "every day the list can change" 00:37:37.

Silicon Valley Overhype vs. Rest-of-World Underhype

Alex Rampell articulates a broader market thesis: AI adoption looks saturated inside the Bay Area bubble but is barely started everywhere else, and reference logos from Silicon Valley function purely as marketing assets rather than a real revenue base. Keith confirms: "for most of these record companies, you end up getting most of your revenue scale from actually the 50 miles out of here and more... we just need to do amazing work for [SV customers] so we can take their logos when we go to the rest of the world" 00:44:03.

Referenceability Trumps Objective Product Quality in Systems of Record

Buyers of infrastructure-level software (CRM, ERP, banking) make decisions based on social proof from perceived-similar companies rather than pure evaluation, because the switching cost of being wrong is enormous. Keith states, "I think, in many ways, your CRM is maybe harder to move off of than your bank" 00:46:11, and Alex adds the mental model buyers use: "I don't want to choose the wrong ERP. So I probably want to choose the ERP that companies who look like me chose... so that I never have to think about this after I buy it" 00:47:10.

2. Contrarian Perspectives

A Fast-Growing, Widely-Used Product Can Still Be the Wrong Company

Most founders treat 25 million users and viral growth as unambiguous success signals. Peiris argues the opposite: growth without founder conviction is a trap. "We just couldn't, for the life of us, make good presentations... We just thought the technology constrained us to being a tool for individuals and students" 00:03:19 — despite having "two million users a month" and demand outstripping their inference capacity 00:02:59.

AI CRM Billboards Are Solving the Wrong Problem

The dominant AI-CRM marketing narrative — "do the work, do the work, do the work" (lead scoring, automated outbound) — is, according to Peiris, not actually a durable wedge against incumbents. "If you like, looked at all of the AI CRM billboards around Silicon Valley in the past couple of years, they've always been about like, do the work... Like Salesforce is going to send emails. I mean, I guess they have as of today, right? And we were like, I think the wedge in Brownfield has to do with... better understanding your company" 00:21:38. This is contrarian because most of the well-funded AI-CRM startups are pursuing exactly the "do the work" strategy he dismisses.

DIY/"Build It Yourself With AI" Fear Is Overrated for Serious Buyers

Despite a Silicon Valley narrative that AI lets anyone build their own software (the "SaaSpocalypse"), Peiris finds this mostly a seed-stage phenomenon, not a real threat from larger companies — and even at the low end it self-corrects quickly. "We would often hear, look, we can either pay [Lightfield] or we can do this over like four weekends... Call us in five weekends... they were like, ah, like, none of this works right. It's hallucinating" 00:47:59. On larger companies attempting to build proprietary "company brains": "maybe the hardest part of it is modeling the customers. So we tried and we don't like our results. So we'll come to you now" 00:48:47.

Unstructured Data Storage Sounds Modern But Fails in Practice

Given LLM capabilities, a fully unstructured data lake seems like the obvious AI-native architecture. Lightfield tried it and rejected it: "we actually tried going fully unstructured and we found that the queries just took too long... you sort of have the like needle in the haystack problem. So we ended up finding this like semi-structured approach" 00:13:11. This runs against a common assumption that more LLM-native = less structure is always better.

You Cannot Charge for Outcomes in B2B Sales Tooling — Yet

Outcome-based pricing is treated in venture circles as the inevitable future of AI-native software. Peiris pushes back specifically for go-to-market tools because outcomes are hostage to variables outside the vendor's control. "It would be incredibly efficient for us to do outbound prospecting for OpenAI... And it would be incredibly inefficient for me to do outbound prospecting for a seed stage startup with no website... So I think where we've landed at the moment is we have to charge for the work. We can't quite charge for the outcome in this space for now" 00:35:17.

3. Companies Identified

Lightfield — AI-native CRM / "business world model" company built by Keith Peiris and team (including Henry, co-founder), which raised a $47 million Series A led by a16z. Described as turning "customer emails, calls, and meetings into a record that AI agents can use to get work done" 00:01:47. Mentioned throughout as the case study for the episode; distinguishing features include schema-less onboarding, an activity-log data architecture, company-wide free distribution, and hybrid platform/consumption pricing.

Tome — Peiris and Henry's earlier company, an AI presentation-generation tool that reached 25 million users and 2 million monthly users at peak, launched around the same time as ChatGPT. Mentioned as the pivot origin story: "We got this product out to launch around the time of GPT-3... we just got explosive growth" 00:02:30, but ultimately abandoned because the founders lost conviction in the product's ceiling.

Power — A healthcare marketplace/clinical-trials company built on Lightfield, cited as the most exciting current use case. "They have this marketplace where they're aggregating folks that have various illnesses and complications that are looking for frontier treatment... they've built automations to go and scrape the FDA and clinicaltrials.gov to give them a world model of every trial going on in the world... Lightfield actually helped someone with Alzheimer's find frontier treatment within days" 00:00:00.

Eleven Labs — Referenced as a cautionary tale about startup-CRM churn risk: they moved off a "flavor of the month" startup CRM to Salesforce after the vendor couldn't build dashboards fast enough ("it was like four months") 00:40:10. Used by Peiris as the paranoia-inducing example driving Lightfield's obsession with shipping speed.

Salesforce / HubSpot — Referenced repeatedly as the incumbent "Brownfield" CRM standard that new/growing companies eventually default back to if a challenger doesn't move fast enough, and as the benchmark for per-seat pricing models.

Sugar CRM — Mentioned by Alex Rampell as the free CRM he used at an earlier startup to avoid paying $85/month for Salesforce, illustrating the Greenfield/Brownfield hiring-driven purchasing dynamic.

4. People Identified

Keith Peiris — Co-founder and CEO of Lightfield; previously co-founded and scaled Tome to 25 million users before shutting it down to pivot. Identified as the guest of the episode for his rare willingness to walk away from hypergrowth and rebuild from first principles. "Keith Perez built Tome to 25 million users. Then he decided to start over" 00:00:51.

Henry — Peiris's co-founder at both Tome and Lightfield. Credited with the internal cultural diagnosis that shaped Lightfield's organizational design: "Henry, my co-founder, says this all the time. It's like we had a lot of people playing house" 00:36:45. Also noted as leading deep-tech-focused sales/customer relationships ("it should go to Henry because... it's a deep tech company") 00:34:18.

Matt — A Lightfield team member handling health-tech-oriented accounts, mentioned in the same routing example: "it should go to Matt because it's a health tech company" 00:34:18.

Alex Rampell — a16z General Partner and co-host; led Lightfield's Series A. Contributes the Greenfield/Brownfield framework, the "hostages not customers" heuristic, and probing questions about pricing, DIY-building, and referenceability throughout.

Joe Schmidt — a16z partner and co-host, guiding the narrative arc of the conversation and closing with the reflective question about advice for founders mid-pivot.

5. Operating Insights

Give the Product Away Free Internally to Build Structural Retention

Rather than treating free seats as pure loss-leader marketing, Lightfield uses total company-wide free access as a deliberate anti-churn mechanism, because it embeds the tool into cross-functional workflows a single department head can't rip out. "It'll create sort of real company network effects that make it harder to rip and replace us" 00:24:23 — and in practice this converts skeptical new sales VPs because "the rest of the company" pressures them to stay 00:24:52.

Segment Pricing by the Job, Not by a Single Universal Metric

Instead of picking seat-based or consumption-based pricing as a philosophy, Lightfield reverse-engineered pricing from customer psychology about budget predictability versus ROI-driven spend: fixed platform/seat fee for "everyday CRM work" that customers want off their risk radar, consumption pricing for pipeline generation and workflow automation where ROI is visible and provable [00:33:19–00:34:48].

Use Bug Bashes and a Low Bar to Start, High Bar to Ship

To keep a flat, generalist organization from shipping chaos, Lightfield keeps project initiation cheap but enforces a rigorous shared quality gate before release — a practice Peiris explicitly imported from his time at Instagram. "The bar to start a project is very low at Lightfield, but the bar to ship the project is pretty high... we still do company bug bashes. It was something I learned at Instagram" 00:39:11.

Prioritize Product Investment by Three-Year Expansion Potential, Not Initial Deal Size

In a competitive ("red ocean") category, Lightfield explicitly models the future expansion value of an account rather than optimizing for immediate ACV, deliberately building for fastest-growing customers over the "average" customer. "We take this perspective of what's sort of the expansion value of this account over a three-year time and how do we prioritize... we've leaned more towards building for our fastest-growing customers than our average customer" 00:42:41.

Use Silicon Valley Customers as Marketing Infrastructure, Not the Revenue Engine

Rather than optimizing sales effort toward the highest-visibility local accounts, deliberately treat them as reference-generation investments for penetrating unglamorous, higher-volume verticals elsewhere. "I just see the focus on Silicon Valley not as efficient revenue capture, but actually just like efficient marketing capture" 00:44:32.

6. Overlooked Insights

The Real Product Wasn't the CRM — It Was Reorganizing Broken Cross-System Data

Buried in the pivot narrative is an important, easy-to-miss structural insight: Lightfield's entire company thesis emerged not from a top-down "CRM is broken" observation but from an operational discovery mid-project — that the CRM, call recorder, and data warehouse actively disagreed with each other, and reconciling that conflict was the highest-value work. "The call recorder often had a different view of reality than the CRM. Actually, the work required to reorganize, it felt like the most important work" 00:06:54. This is a much deeper claim than "better CRM UX" — it implies that most system-of-record categories (ERP, support, HRIS) may have the same latent reconciliation problem waiting for an AI-native rebuild, which is a broader whitespace thesis than either speaker calls out explicitly.

The Failed Go-to-Market Assistant Reveals a Structural Rule About Data Ownership and Pricing Power

Easy to skim past: before landing on the CRM idea, Lightfield built a genuinely well-liked go-to-market assistant that users loved daily but wouldn't pay for — specifically because it sat on top of someone else's data/system of record. "We had all of these AEs using it every day, but we had no pricing power because it wasn't our data. And there were 10 other companies and all of these companies competing for it" 00:07:53. This is a significant, generalizable investment/operating principle understated in the conversation: product love and usage frequency are worthless economically if you don't own the underlying data layer — an "application layer without a data moat" warning that applies broadly to today's wave of AI wrapper startups building on top of others' systems of record.