Inside Lightfield’s Vision for the AI-Native Business
Alex RampellJoe SchmidtKeith Peiris
- Keith Peiris stopped Tome and pivoted away despite 2 million users a month because “none of us liked the product.” The team could never see a “high-quality, discerning presentation maker” using it indispensably. His answer to why not wait out the model curve is the episode’s sharpest technical claim: the model “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.” Capability scaling alone did not solve the context problem.
- Lightfield (just raised a $47M Series A led by a16z) is built as a “business world model,” not merely a CRM that does tasks. The canonical primitive is a Facebook-timeline-style activity log of every relationship—Keith says three of five founding members came from Facebook, and later describes them as coming from Facebook or Datadog—stored semistructured after fully unstructured search proved too slow, with an effectively schemaless setup: “intelligence is greater than schema.”
- The first pivot product, a go-to-market assistant with daily AE usage, failed on pricing power—“it wasn’t our data”—showing why owning and modeling the underlying data became strategically important. Cold-start was solved with literal negative pricing: free office space for 10 startups that used the four-month-old CRM and Slacked feedback “every two hours.”
- Both pricing extremes failed empirically: seat pricing broke because “the head was using 10,000× more than the tail,” and pure consumption credits froze usage—“the worst three weeks of the company’s life.” The landing: a platform fee plus seats for core CRM work, with consumption for pipeline generation, workflow automation, and intelligence/forecasting. Outcome pricing is not viable for now because “our outcomes are dependent on the strength of your product-market fit”—outbound for OpenAI versus a seed startup with no website.
- The brownfield wedge is understanding, not automation—a direct rejection of the AI-CRM billboard consensus of “do the work, do the work, do the work,” because “Salesforce is going to send emails.” Lightfield gives the product free to everyone at a customer, creating “real company network effects that make it harder to rip and replace us”—so when the Salesforce-trained VP of sales balks, engineering, finance, and the rest of the company push back.
- DIY is “a little overtalked about”: vibe-coding founders get “good luck—call us in five weekends,” and enterprises building a “company brain” return because “maybe the hardest part of it is modeling the customers.” Trust dominates the category—“your CRM is maybe harder to move off of than your bank”—which is why Keith treats Silicon Valley as “efficient marketing capture” for reference logos, not revenue.
- His biggest fear is speed: he cites G2 reports about a16z portfolio company ElevenLabs moving from a startup CRM to Salesforce because dashboards took four months—“we just have to build everything.” The organization is built for it: 40 generalists, no swim lanes, one daily stand-up with stack-ranked problems, prioritized by three-year expansion value of the fastest-growing accounts.
1. Stopping Tome at 2 million users a month: “none of us liked the product”
- Tome launched AI presentations around ChatGPT’s debut and hit explosive growth—2 million users a month, users lining up when inference ran out, and roughly 25 million users or so in total—but Keith and the team stopped and pivoted away anyway. Metrics mattered, but the deeper trigger was instinct: “if you’re a founder you have to love the product,” and the team “could not see the path” to bankers, consultants, or memo-writers using it indispensably. They thought the technology constrained it to individuals and students.
- Alex’s probe was why not shrink the team and wait, given “the leap between GPT-3 and GPT-6 with Astra is extraordinary.” Keith’s answer: 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.” The best case was “a one-shot, pie-in-the-sky presentation,” and they could not get excited about that.
2. The pivot: follow the heat until the data problem reveals itself
- The method was deliberately priorless: “what makes early-stage founders good at this is that you have no priors, you’re willing to ignore the thing you built.” They mined the user base, found B2B concentration in sales and marketing, ran 12 free pilots with large companies, and let customers pull them sideways into research, lead qualification, and expansion analysis. The real discovery was that reconciling disparate systems was the actual work: the call recorder often had “a different view of reality than what the CRM had.”
- The first product, a go-to-market assistant, had daily AE usage but no pricing power because “it wasn’t our data”; 10 other companies were competing for the same underlying opportunity. So they shrank the team and built a CRM from first principles “in the dark for about four months.” They then solved cold-start by posting on X and LinkedIn: free office space if you use our CRM. Ten startups came, were angry about missing features and slowness, but used it every day and sent Slack feedback roughly every two hours.
3. The activity log is the primitive; “intelligence is greater than schema”
- Keith says three of five founding members came from Facebook; later, discussing the architecture, he says three of five came from Facebook or Datadog. The design echoed the Facebook timeline: a canonical chronological log of the relationship—first outreach, meetings, documents, product usage, and payments—which then triggers conventional CRM updates such as fields and stages.
- Fully unstructured storage failed because “the queries just took too long,” creating a needle-in-the-haystack problem. They went semistructured instead: large amounts of unstructured data live in the activity log, which the system traverses to infer causality. That is why Keith says it works better than a data lake. CRM consultants’ most consequential decision is the data model—“you screw that up, it’s over”—so Lightfield is effectively schemaless: fields can be filled or refilled later from the log. “It feels like a consumer product. You just press sync, you wait five minutes, and it’s there.”
- The showcase customer is Power, which works with pharmaceutical companies to find clinical-trial participants and operates a marketplace for people with illnesses or complications seeking frontier treatment. Power modeled both its B2B and B2C sides in Lightfield and built automations that scrape the FDA and ClinicalTrials.gov into “a world model of every trial going on in the world.” Keith says Lightfield helped someone with Alzheimer’s find frontier treatment within days.
4. Brownfield strategy: the wedge is understanding, not “do the work”
- Alex mapped the terrain: “the best companies have hostages, not customers”—SAP has hostages—and cloud beat on-premises partly by redefining the problem. Keith’s honest admission was that “we didn’t have a sharp enough thesis on how to get into brownfield,” so they tried to win new companies first. He and his chief of staff prospected YC companies on LinkedIn and by email, betting that iteration and close listening would reveal the eventual wedge. Some customers went from zero reps to 100 reps while Lightfield observed the problems that emerged.
- The wedge that emerged rejected the AI-CRM billboard consensus of “do the work, do the work, do the work.” “Salesforce is going to send emails,” Keith said—“I guess they have as of today.” The brownfield wedge instead has to be “better understanding your company so you can steer” it through “the chaotic era of company building.”
- Against the Salesforce-trained VP of sales—Alex’s example was abandoning free SugarCRM after a new VP said, “I’m not using that fucking thing”—Lightfield gives sales-led plans to everyone in the company for free. That creates “real company network effects that make it harder to rip and replace us.” When a seasoned VP says they only know Salesforce, the rest of the company can answer that engineering uses Lightfield to understand customers, finance uses it for revenue recognition, and customer success uses it for account scoring.
- The product strategy is pragmatic rather than religious. Keith still runs his own sales meetings from the spreadsheet view, and Lightfield keeps dashboards and table views. But sequences—once expressed with arrows, conditions, and variables—can become chat: the agent writes a recipe against the company’s world model. Skeptics who “need my knobs” can come around when they find the new workflow more efficient and easier to learn.
5. Pricing: both extremes failed, four buckets survived
- Pure seat pricing matched the Salesforce and HubSpot world but failed because “the head was using 10,000× more than the tail” in consumption. Pure consumption pricing, where everything was a Lightfield credit, produced “the worst three weeks of the company’s life”: signups touched nothing.
- Customer conversations yielded four buckets. Everyday CRM work—capturing meetings, filling records, and updating tasks—should be covered by a fixed platform or seat fee. Customers do not want to budget around the error bounds of their core CRM. Pipeline generation can support consumption pricing because enrichment creates potential revenue alpha. Workflow automation—such as researching an inbound demo request and routing it to the right representative—does real work with an understandable return, so customers expect to pay for it.
- The fourth bucket is perhaps the most undiscovered: intelligence and forecasting, a “crystal ball or snow globe of your company” for scenario planning. Keith says, “I changed my sales process after letting GPT-4 rip on Lightfield for a couple of hours over the weekend.”
- Outcome pricing is not viable for now because “our outcomes are dependent on the strength of your product-market fit.” Outbound prospecting for OpenAI would look highly successful; outbound for a seed-stage startup with no website would be inefficient. So Lightfield charges for the work: a platform fee plus a seat fee for core CRM, and consumption for everything else.
6. Build-it-yourself is overtalked—and trust, not features, closes deals
- Customers that keep Lightfield as their system of record but build their own harness over MCP or CLI “almost always realize” within a few weeks that “your harness was actually doing quite a bit”—including entity recognition, precision and recall, and speed. “That’s our job,” Keith says, while emphasizing that the data belongs to the customer.
- Seed founders who say they can build a CRM “over four weekends” get “good luck—call us in five weekends.” They often return saying the result hallucinates or sends bad emails. Larger companies are less likely to build their own system of record, but often try to build “my own company brain.” Many return after discovering that building a business world model is difficult, especially modeling customers.
- Alex’s observation that AI is “almost overhyped in Silicon Valley” and “massively underhyped outside” meets Keith’s reframe: Silicon Valley focus is “not efficient revenue capture, but efficient marketing capture.” Serve customers that have raised $200 million but still have only three go-to-market people, do excellent work, and carry those reference logos into healthcare, fintech, manufacturing, and other markets.
- Because “your CRM is maybe harder to move off of than your bank,” referenceability and trust matter enormously. Keith says Lightfield’s early security work—signing BAAs and doing penetration testing for years—helped it win complex health-tech deals, where it now benefits from a network of references.
7. Forty generalists, no swim lanes—and the paranoia of speed
- The Tome regret, as repeated from co-founder Henry, was that “we had a lot of people playing house”: leaders defended swim lanes, feedback stayed siloed, and the company slowed to a crawl and became difficult to pivot. Lightfield’s answer is 40 people, one daily stand-up, stack-ranked problems, and continuous planning—“whoever’s free just picks up the problem.” Everyone owns product and customer success; engineers, designers, and CSMs all run projects. Keith’s formulation is, “Man plans and OpenAI lives.”
- Coherence comes from constantly editing the roadmap against the mission. The bar to start a project is low, but the bar to ship is high, and company-wide bug bashes gate releases before they reach customers.
- What worries Keith most is speed. He cites G2 reports about a16z portfolio company ElevenLabs, which used a startup CRM but moved to Salesforce after the startup CRM could not build dashboards quickly enough; after four months of waiting, the team was tired of asking. “We’ve got this incredible wedge,” Keith says, “and we just have to build everything.”
- Prioritization now considers an account’s expansion value over a three-year horizon, skewing toward the fastest-growing customers rather than the average customer. It is “a time of being a little more maximalist than usual.”
- His pivot advice is that “almost none of the noise around you matters.” Find pain, become inspired to solve it, and be maniacally focused on customers. Office nostalgia, uninspiring food, and options-repricing concerns during the pivot were “total noise.”
Full transcript
Three out of our 5 founding members came from Facebook. Building a revenue team is a very rich problem set. It's something that everyone cares about. It could always be done better.
What's the most exciting thing someone's doing with Lightfield today?
This company, Power, has a marketplace where they're aggregating folks who have various illnesses and complications and are looking for frontier treatment. They've modeled all of this in Lightfield, and Lightfield actually helped someone with Alzheimer's find frontier treatment within days.
Is there anything that you've done differently in this AI era?
As a CRM company in a red-ocean space, we have to be an expansion company. If we can help you completely model your business and your customer reality, then the rest will be easy.
What would be the one piece of advice that you'd go back and give yourself if you were just starting again?
1. From Tome to Lightfield: The Pivot Story
Welcome back to the a16z Podcast. I'm Joe Schmidt. I'm joined by my partner, Alex Rampell, and Keith Peiris. Keith is the CEO of Lightfield. Lightfield just raised a $47 million Series A led by us, and they're building a business world model. A business world model turns customer emails, calls, and meetings into a record that AI agents can use to get work done. We'll explore how Keith pivoted, which is very interesting, to Lightfield, how they built the initial product, and what customers can do with it. Keith, thanks for joining us.
Excited to be here.
Maybe we'll start by going back to the Tome journey and how you got to Lightfield. It's a very atypical journey. You got 2 products to explosive scale. Tell us a little bit about that experience and how you ended up at Lightfield.
We started mostly because we were consumer people, and we thought LLMs were going to change the way people communicate. We were working on selfie design at Instagram and Messenger, and we decided to go into the storytelling of ideas. We got this product out to launch around the time of GPT-3.
And what was the product? Tell us a little bit about it, too.
It was a presentation product where you could use GPT to generate presentations, generate pages, and so forth. We launched it around the same time as ChatGPT, and we just got explosive growth. We got 2 million users a month. People were lining up when we didn't have enough inference to support them.
2. Why Not Just Wait for the Technology to Get Better?
Wow. Oh, my gosh. You got that to amazing scale, and then you decided to stop and completely hard-pivot. What was that decision like? How did you make that choice?
I would say there were a lot of metrics behind it, but deep down, at an instinctual level, none of us liked the product, which is kind of a funny thing to say. At the end of the day, if you're a founder, you have to love the product that you're building, and you have to be excited for your customers to use it. We just couldn't, for the life of us, make good presentations.
You know, we could never see the path to a high-quality, discerning presentation maker using this tool in an indispensable way. We couldn't see people like yourselves using it for memos. We couldn't see people using it in investment banking or consulting. We thought the technology constrained us to being a tool for individuals and students.
So interesting. You go from AI presentations to nothing. How did you make the choice of what to build next? Maybe if I can rewind a little bit: Sometimes you get to this point and you're like, "Oh, well, if the technology continues advancing, then it will be good enough."
But that's also a little bit of a danger. Why not just wait it out for the technology to get better? The leap between GPT-3 and GPT-6 with Astra is extraordinary. Obviously, if you're burning money, you don't have the luxury of waiting, but how did you think about the shape of the curve and why not wait out the curve?
We definitely thought about shrinking the team and waiting it out because we were seeing GPT-3.5 to GPT-4 and seeing the leaps being made. But I think the biggest issue was that the model just didn't have enough context to really understand the presenter, the audience, and the relationship between the presenter and the audience. No amount of general reasoning gets you past that.
We thought this was, at best, a great tool for a one-shot, pie-in-the-sky presentation, and we just couldn't figure out how to turn that into something that we were excited about.
How did you think about the problems that you wanted to attack next? I think the hardest part of what you just said, and the most painful part, is that none of you liked the product. That's hard if that's your entire life. How did you decide what product you actually wanted to build? What was that process like?
We had a lot of conversations internally about, "Wait a minute, why did we even build this company?" We built this company because we wanted to help professionals tell expert stories—hard stories. So we thought, "Let's see if we can find the B2B use case."
We looked in our user base. We had 25 million users or something, and we found that the B2B users were in sales and marketing.
Yep.
We reached out to them. We went and got 12 pilots from, call it, 500,000-person companies around here, and we said, "We'll do this for free. We'll do great presentations for you, and let's see where it goes."
We ended up being used by a couple of sales teams. At first, we came in with, "We'll make your new-business decks or your proposals." Then we thought, "Wait a minute. While you're here, can you do other things? Can you do some research? Could you help us qualify leads? Could you help us understand companies for expansion?"
At that point, we were like, "Sure. Let's just follow the heat and see what happens." I think what makes early-stage founders good at this is that you have no priors. You're willing to ignore the thing you built. We thought, "Let's just follow this trail and see where it goes."
The first thing we asked was, "Oh, just give us access to your context," which at the time I thought was Salesforce. We said, "Great. Give us access to your CRM, your call recorder, your data warehouse, and more, and then we'll figure out how to do this work for you."
Then we realized that the hardest part of doing this work was actually making sense of all the data across all of these disparate systems. It was incomplete and conflicting. What the call recorder had was often a different view of reality than what the CRM had. The work required to reorganize it felt like the most important work.
3. Knowing You're Onto Something With the New Product
I think through that we started to realize, "Wait a minute, maybe this is the more interesting problem to solve," which is that if you can reorganize reality for a company in a way that machines can understand and humans can understand, that feels like a much more interesting and enduring company than the one we were working on right now.
When did you know that you were onto something with the new product?
There was a little bit of discontinuity. We didn't know we wanted a CRM first. First, we built a go-to-market assistant. We got people to like it, and then we couldn't get anyone to pay for it.
We had all of these AEs using it every day, but we had no pricing power because it wasn't our data. There were 10 other companies, and all of these companies were competing for it. So we decided to shrink the team, start from scratch, and reimagine the CRM from first principles.
I think we built in the dark for about 4 months. Then we needed to find people to use our CRM, and it turns out no one wants to use your 4-month-old CRM. So we looked at the only asset we had left, which was this giant office space we couldn't get rid of. We thought, "I'm going to post on X and LinkedIn: You can sit in our office space if you use our CRM."
Negative pricing.
Negative pricing. Yeah, exactly. So we found 10 startups to come use the product, and for some reason they were in it every day. They were mad about everything that was missing, and they were mad about how slow it was, but they were in it every day. They were giving us Slack feedback about it every 2 hours.
4. CRM as a Repository: What's Actually Broken Today
I was like, "Oh, this is so different from before, in the sense that we have this barely working, barely finished product that people are in all the time. They care about it so much that they're going to give us feedback about it on an hourly basis."
When you think about CRM, did you start off saying, "I view Salesforce and HubSpot and other things, and here's what I want changed"? How did you triangulate on the job to be done? On the one hand, a CRM is just a repository—it's the repository of all the customer information. On the other hand, it's also this: If you've talked to anybody who's run a sales team, one of the biggest pain points is, "My stupid salespeople don't update the CRM."
And it's often a stale repository. But I guess what were the governing principles, if you will, around what was broken with the world today? What were you going to do differently? Or was it, “Let's just have people with negative pricing in my office, and I'll figure out what they're complaining about and build around that”?
I think one of the things that benefited us was honestly how naive we were about the space. If you're in sort of a growth-stage company, the CRM is the tool that reps use not to forget things. It's the tool that powers low-level automation, and it's also the tool that powers forecasts.
I think our naive view was that the most important thing here is the latter: if we can help you completely model your business and your customer reality, then the rest will be easy. The rest should just be prompts and tool calls. Because of that, I think a lot of the world ran toward CRM for work. We ran toward high-fidelity business modeling.
5. The Architecture Choices That Make Lightfield Work
Because of that, we were like, what gets in the way of this? It's the rep's manual entry, the API quality, and we were sort of on our island trying to world-model instead of sending emails.
6. Company Culture, Velocity & Shipping Speed at Lightfield
Yeah. Can you talk a little bit about the very intentional decisions you made on the architecture and how you built the primitives that allow this to happen? It's interesting now: it's beautiful and the experience is incredible, but it only works because you made the right choices early on. Talk a little bit about how you did that from first principles and what it now enables.
Yeah. Actually, 3 out of our 5 founding members came from Facebook or Datadog, and we had this very naive view that the most important thing in the CRM is modeling the relationship. We looked at the Facebook timeline and thought, “We just need to model the chronological relationship between your business and this business.”
We built out the activity log first: When did you first reach out? What did you say to them? What did they say to you? What meetings did you have? What documents were sent back and forth? Eventually, what are they doing in your product, and how are they paying you? We thought working off this activity log was the right primitive.
Interesting.
So we have a system where it builds the activity log for your relationship, and then it uses that to trigger the traditional CRM updates that you'd expect—updating fields, updating stages, and so on. You always have this canonical log of the relationship that everything is built on top of.
And then how did you think about—there's the chronological view, but then there's all of this other metadata that's never existed in any other CRM? This was one of the things I remember initially really jiving with you on. There's so much other context, and how do you then think about that as part of this record? I don't know when that came in, or was there a certain workflow that you were trying to enable that triggered that?
We actually tried going fully unstructured, and we found that the queries just took too long. You have the needle-in-the-haystack problem. So we ended up finding this semistructured approach where we store gobs of unstructured data in the activity log, and then the system can use the activity log to infer causality and work through from there. That's why it works so much better than a data lake.
Of course, you can put anything in it. You can put your Snowflake records in it.
Well, maybe give an example of what that means. How is it actually used at the end state?
A good example would be that we have a lot of customer success folks on Lightfield, and they might be tasked with, “Is this account ready for expansion?” If you ask Lightfield this open question—what should we sell to them, and when should we sell to them?—it can now go through everything this account has done.
First, it's the people interactions: What did they say to you? What do their tickets look like? Then it can also go to the product usage, which is stored as activity log entries. You're like, “Well, they haven't logged in in a month. Maybe you should try to sell them more stuff.”
Yeah.
If you want to compare—which, you know, the most common question is, “Which of these accounts should I work on for expansion?”—now it can traverse the CRM schema and then dive deep into the log of each customer to give you a good answer.
How did you think about making the initial experience and the initial configuration of all of this intuitive for a small, midsize, or even large company?
We had this view that—it's funny, we spoke to a lot of CRM consultants, and we started to write down, “What is it that you do?” We found that the biggest, most consequential decision they help you with is your data model.
Yeah. You screw that up, it's over.
It's over, right? If you get the wrong stages or the wrong fields, you can't get the reps to go back in time and fill it out. It's over. We were like, can we be effectively schemaless? Can we just say, “We'll connect you to your emails, give you a call recorder, connect you to your data warehouse, and then assemble your relationships for you”? You can fill out the fields later. If you change your mind about the fields, you can just traverse the activity log and refill them.
So we ended up with this schemaless setup where you log in, connect your email, and more. We have built-in enrichment sources, and then it just assembles everything for you in real time.
Yeah.
Basically, intelligence is greater than schema. That's the crux. It feels like a consumer product: You just press sync, wait 5 minutes, and it's there.
So cool.
Well, if you go back to the really old days, it was to save space, because these were all just relational databases, right? To save space in a given table, each column—if you know SQL—was VARCHAR, or variable character. You would actually predefine how many characters that particular column could have in the table. You'd say, “Name would be VARCHAR(25).”
It's like, “Oh, shoot, this person has too long of a name.” It's just funny how far this has gotten, because back then, when you were really managing every bit and every byte, you would literally predefine the maximum number of characters for a column in a table. Now it's like, no schema. It's just funny.
Yes. The times have changed.
Yes, yes. Intelligence is greater than schema. Exactly. I'm drafting off that intelligence thought. So much of what's interesting right now is that you're giving someone the kind of chassis to apply intelligence to really hard problems.
In that example you gave of a CSM saying, “How do I upsell this customer?” what's the most exciting, random, or interesting thing someone's doing with Lightfield today that they couldn't have done in the AI world or another world before this one?
One of the things that I'm really proud of is that we started with fully arbitrary schema—custom objects, custom relationships—because it just doesn't matter in Lightfield the way it does in other CRMs. I love our customers that have strange business models where they just need to model different things.
We have this company, Power, and they work with pharmaceutical companies to help them find clinical trial participants. On the other side, they have this marketplace where they're aggregating folks who have various illnesses and complications and are looking for frontier treatment.
They've modeled all of this in Lightfield—the B2C side and the B2B side—and they've built automations to scrape the FDA and ClinicalTrials.gov to give them a world model of every trial going on in the world. Then they do matching on the B2C side as well as with the right pharmaceutical company. So there are collisions happening.
Lightfield actually helped someone with Alzheimer's find frontier treatment within days.
7. Greenfield vs Brownfield: Cracking the CRM Market
That's incredible. One question that we talk about a lot here is this greenfield-versus-brownfield thing. If you think about startups normally, selling into the brownfield is hard just because it's brownfield. What does brownfield mean? It means that it's been trampled by an incumbent—hence brown.
Try selling ERP: You have a product that's much, much better than SAP, but there's a saying that Joe has heard me use a million times: the best companies have hostages, not customers. SAP has hostages. That's a brownfield.
Sometimes you can break in. I mean, look at cloud versus on-premises. What did cloud do? It was brownfield, but it kind of just redefined the problem and said, “You know what? What you're using—if you go back to the CRM days, you're using Siebel Systems running on your IBM AS/400 mainframe in your office.”
You're tired. The guy who maintained it quit; he was 92 years old. Maybe now you should use a cloud-based vendor. That's partially how that brownfield was done, from on-prem to cloud. But the other strategy is going greenfield: saying, “I'm not going to bother with the hostages. I'm just going to build the best product in the world,” and then brand-new companies untethered from any existing software solution will just use me.
And I guess, when you were thinking through—going back to the early days of negative pricing, with the free office space being the negative price—how did you think about who the right customers were? How did you get to what we call an ICP, an ideal customer profile? Maybe talk about that a little bit.
Just to be totally real, when we were starting the company, we didn't have a sharp enough thesis on how to get into brownfield. So I think we had this perspective that building a revenue team is a very rich problem set. It's something that everyone cares about; it could always be done better. So we believed there was something in here, but we weren't sure exactly what it was. So we figured, let's just try to win a new company first.
Yeah.
How do we win a new company? It was actually my chief of staff and me doing LinkedIn prospecting, emailing YC companies, being like, “Hey, can we beat one of these startup CRMs?” And that was how we got started. We figured, by listening really deeply, we would find the sort of wedge required to go brownfield eventually.
I think we sort of found it over the past few months, which is to say you get lots of iteration. One of the nice things about serving an early-stage startup right now is that it's never been faster to go from pre-seed to seed, seed to A, and A to B. We now have customers that had 0 reps when they joined us and now have 100 reps.
We've been able to look inside and figure out what problems we're solving for you that you actually care about. It seemed like it was actually a little higher-level than we had anticipated. If you looked at all of the AI CRM billboards around Silicon Valley in the past couple of years, they've always been about “do the work, do the work, do the work,” right? Lead scoring, sending out outbound emails, and more.
I think our team always had this perspective that that wasn't really the wedge to do brownfield. 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, so you can steer your company through the chaotic era of company building. But it took us, honestly, 6 months of having customers and staring at them for that to emerge.
Yeah.
Right. One of the things that also emerges is, if you're creating a better product and you have all of these people who are consumers of the product, of course—but let's just take the greenfield versus brownfield distinction. The VP of sales you hire at your greenfield company has been acclimated and trained to use this thing.
I remember when I started one of my first companies, I was so adamant about not paying $85 a month for Salesforce that I used something called SugarCRM, which was free. And finally I gave up. I finally started paying for Salesforce. Why? Not because of product gaps or anything, but I hired this VP of sales, and he was like, “I'm not using that fucking thing.” I was like, “It's easier, it's better,” but that was also challenging.
It's not like a new company needs to be in the business of training people per se, but there's also that gap. One of the things that's seldom understood about greenfield versus brownfield is that the greenfield people hire brownfield VPs who actually make product—or, sorry, make purchasing decisions.
I feel adamantly that this is a much, much, much better product, but how do you overcome some of the objections, if you will, from people who are like, “You know what? I'm just perfectly fine using something that doesn't work as well”?
It's a good question. One of our design principles early on was that we figured we'd be great at convincing the founder, the engineering leader, and the product leader to use frontier tech to understand and serve customers. But it was probably going to be a lot of work to convince the VP of sales who joins that this is the way.
So we had this idea: for all of our sales-led plans, we're going to give this thing away for free to anyone in the company. One, it helps Lightfield understand what engineering's doing, what customer support's doing, and what finance is doing. The other thing is it'll create sort of real company network effects that make it harder to rip and replace us.
We've been at a few companies now where the seasoned VP of sales comes in and says, “This is cool, but I only know how to use Salesforce because I've been trained that way.” The rest of the company is like, “Well, hold on. This is how the engineers understand customers. This is how finance does revenue recognition. This is how customer success does account scoring. Can you try harder to figure this out?”
In which case, we have the relationship, we jump in, we show them how we're going to make their lives easier, and we've got a fighting chance.
8. Skeuomorphic vs Natural Language: The Product Trade-Off
There are so many different ways to now do those tasks that were just a table in the past. Do you want to try to give those tables to people and make it very skeuomorphic—here's the CRM that you're used to, and it's also blue and blah blah blah—or is it, “Hey, actually, you can now use natural language to do a lot of this. You don't actually have to look at all these dashboards. Lightfield will just tell you when something happens”? How do you think about those product trade-offs, which I think a lot of systems of record are actually trying to figure out? How do we bring people to the real future, which is now here?
Yeah. I think we've actually been on the side of pragmatism here. Right now, I'm the de facto sales manager at the company, and I still run all of our meetings from the spreadsheet view. We had this view that if we really want to be a real enterprise CRM, we can't be religious about the way people work.
So we've leaned into having great dashboards and great table views. You can use those things if you want, and then, if you're at the frontier, you can do it all—
And you want a CLI or whatever.
Yeah, exactly. You can also do that.
Yeah, totally. I find that—it's funny. I just bought a new car. I came from a BMW, which had 9 million buttons. If you go in an airplane, right, you look at the cockpit and think, “How does the pilot know what these are?” There were 9,000 switches—
Straight out of Apollo.
And the really cool thing—it's actually even scary—is, I don't know if you know this, but the newest Teslas don't even have a stalk to change gears.
Yeah.
There isn't one.
Yeah.
It's the most physical of physical buttons, and it's completely gone. But actually, it's so simple and so intuitive that a 5-year-old can use it. Hopefully a 5-year-old isn't driving a car, but you get where I'm going.
It's another way of thinking about schema versus schemaless, right? If you have all these knobs and switches, it's great. It's so advanced—the VP of sales, or if you talk to finance people, they know all their buttons in Bloomberg—but one of the ways you're able to build something great is just to make it so intuitive—
That even a 5-year-old can do it.
Yeah. There's a whole category on Reddit called ELI5—“Explain Like I'm 5,” right? So it kind of feels like the cool thing—I don't know if you're the boss, not me, but it feels like that's what you have, right? You have this old era with lots of knobs and switches and everything else, but part of the benefit of natural-language intelligence is that you don't need any of those things. It's almost rendered anachronistic.
Yeah, this is true. And we run into this in different ways, right? We haven't won the war on deterministic dashboards. I think everyone wants to look at the same dashboard every morning when they're drinking their coffee. Me too.
But they're all—
Hopefully it's revenue going up.
It's always up.
But I think, for some of these old workflows, we've completely reimagined them. One of the most classic sales workflows is the idea of a sequence, right? You send out 5 emails at different times, on different triggers, to someone who might have expressed interest in your product, and you used to have to express these things—
With arrows.
—and conditions and variables.
Whereas in Lightfield, we're like, no, you just chat with the agent. The agent writes out a recipe for you. The recipe takes into account what's in your world model, and it sort of runs with it.
At first, we had some sales leaders who were like, “I don't trust this thing. I need my knobs. I need my switches.” But then they sort of came around to, “Wait a minute, this is actually more efficient, better, and I have to learn less.” I think we're running into this in sort of every aspect of running a go-to-market team.
Yeah. There's, of course, this whole movement right now where everyone thinks they can build basically everything themselves. Software is dead: What can I do myself versus not?
How do you talk to customers that want to try to build elements of this, or are trying to build elements of this? Say a little bit more about that movement and how you're responding at Lightfield.
First of all, I think that just being a system of record means we have to be open-minded. It's your data, not ours, right? We tell our customers that we take a lot of pride in the modeling and the accuracy of your business. We built our own email sync, our own Slack sync, and our own data warehouse sync, and we are proud of how high-performance our database is. You can take that wherever you want.
We've actually had a couple of customers that would use Lightfield as the core system of record, and then they'd try to build their own harness on top. Maybe they've got a company harness, right? We support that. You could push it all through MCP or CLI. But after a few weeks, they almost always realize, “Wait a minute, your harness was actually doing quite a bit.”
Yeah.
It had better entity recognition, better precision and recall, and it was faster. We're like, “Yeah, that's our job, you know.” Our take is, it's your data; do what you want. But we're going to work hard to earn the right to your productivity every day.
9. Pricing in the AI Era & Outcome-Based Models
Yeah. Where do you think—kind of on that topic—pricing is such a crazy question right now? There's seat-based pricing, where the seats don't make sense anymore. Imagine running—thank God you run Lightfield and not Zendesk, right? How many seats do you need? Maybe 0. That's pretty scary.
Then there's outcome-based pricing. But what is an outcome? Am I selling the thing? What if I just have a happy customer forever? I'm still using CRM for that, so outcomes are kind of tricky.
Against all of that, you have this idea—which I don't agree with, obviously—that software is dead, the whole SaaS apocalypse. You're also going to say supermarkets are dead and you're going to grow your own food; car manufacturers are dead and you're going to weld your own aluminum. You could take this to the extreme.
I guess where do you think about this? How did you come up with pricing? Maybe that's question number 1. And then what do you think about outcome-based pricing in this space? Support almost makes sense, where it's a cost: Can I bring down the cost? Of course, you can have software answer questions like a knowledge base. There's a fixed number of answers, and I kind of match it up with the question.
Whereas it almost feels like the last job standing for humans will be sales, right? Anyway, pricing, outcomes, competition—I'm just curious how you thought about it. How did you get to your pricing, from negative pricing—which I would not recommend you keep in perpetuity [laughter]—and how do you think about everything that's going on in that triangle?
Yeah, it's a good question. We started with both extremes, and we figured that was the fastest path to find the efficient frontier. We started with pure seat pricing because it just matched the Salesforce and HubSpot world.
I think it was received really well by our customers. But the head was using 10,000× more than the tail.
Yeah.
In terms of consumption. We realized that we weren't going to be in business for a long time with per-seat pricing.
So then we tried pure consumption pricing, where everything is a credit in Lightfield. Then we found that nobody touched anything.
Yeah. [laughter]
You could imagine it was the worst 3 weeks of the company's life. We had all of these signups, but they weren't doing anything.
Yeah. This is not good.
Then we started talking to customers. Joe, actually, we divided the work that Lightfield does into sort of 4 buckets, and we're able to distinguish between them.
The first was your everyday CRM work: capture a meeting, fill out records, update tasks. I think most of our customers just expect that to be covered in a platform fee, or expect that to be covered in seat-based pricing. They're like, “I don't want to think about the error bounds of my core CRM when I look at my budget over the next year.” They're like, “Make that fixed.”
Then we found there were 2 other buckets. One was pipeline generation, where you're pretty willing to pay consumption pricing. With pipeline generation, you realize there's enrichment and there's actually alpha for you. You'll get more meetings that can convert to revenue.
Then there's this other bucket of workflow automations, where I think there's just an expectation that you pay for those.
What's a good example of that, by the way?
Someone signs up on our website for a demo, and then Lightfield does a little bit of research and realizes, “Oh, this should go to Henry because it's a deep-tech company,” or, “It should go to Matt because it's a health-tech company.”
I think that's doing real work, right? You can imagine the ROI on that, so you'll pay for it.
The last piece is honestly intelligence and forecasting, which is maybe the most undiscovered part of Lightfield. You've got this little crystal ball or snow globe of your company, and now you're going to deploy frontier intelligence to it, start scenario planning, and figure out what's next. In that world, you'll definitely pay for the alpha there.
I changed my sales process after letting GPT-4 rip on Lightfield for a couple of hours over the weekend. So, all of that is to say, we sort of landed on a platform fee plus a seat fee for core CRM, and we do consumption for everything else. It's landed pretty well.
To your other question about outcomes, I think the hardest thing about being a sales company is that our outcomes are dependent on the strength of your product-market fit, right?
Of course.
It would be incredibly efficient for us to do outbound prospecting for OpenAI [laughter], and they'd have incredible outcomes.
And they are interested.
Yeah. And it would be incredibly inefficient for me to do outbound prospecting for a seed-stage startup with no website, right?
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 the space, for now.
Yeah. Maybe from when you started Tome to today, how has the actual work changed? Something that I found amazing and really fascinating about the company when we first met over a year ago was just how fast you guys were already shipping, and it feels like the velocity has just gone through the roof.
I'd be curious how you've designed company culture and tooling, and all the decision-making around velocity. Talk a little bit about that step-function change from before and after.
Yeah. I think one of the things I regret about Tome that we definitely fixed with Lightfield is—Henry, my co-founder, says this all the time—we had a lot of people playing house. You had the product leader, the marketing leader, and the customer-success leader, and they all had their swim lanes. They'd get really mad if someone gave them feedback about something in their swim lane.
I think because of that, we were slowed to a crawl. Maybe put another way, it was impossible to pivot, right? You had all of these appendages that weren't talking to the brain.
We also tried to plan too far in advance, which I think is challenging in this era of company building. So we're like, none of that. No one has a swim lane.
The expression is, “Man plans and God laughs.” Now it's, “Man plans and OpenAI lives.”
Yeah. [laughter]
Yeah, exactly. So now—
It's funny: everyone owns product and everyone owns customer success.
Which is kind of an interesting setup. We have 40 people at the company. Everyone shows up to the same stand-up every morning. We stack-rank the most important problems. Some of them are delivery, and some of them are engineering.
Some of them are CS. And then whoever's free just takes them.
Interesting.
We do continuous planning, so every day the list can change. Every week, we reassess the list, and then whoever's free just picks up the problem. But because of the age that we're in, anyone can ramp up on a customer through Lightfield. Anyone can ramp up on our design system library because the LLM can touch Figma, and anyone can auto-create tasks in Linear because of Lightfield's connectivity to Linear.
We have this environment where basically everyone is a generalist, and engineers run projects, designers run projects, and CSMs run projects. If you're the specialist, you work on more of the similar projects than not.
Yeah. How do you keep everyone as a generalist, with everyone constantly coming up with something and working on the most important thing, aligned with this very coherent view of the future? I just don't know how you manage that process.
I would say that it's always this push and pull. Each week, we meet about road mapping in GTM and edit it. We're like, "Yes, we had this customer that asked for this. How does it fit into the mission? It doesn't."
I think it's just this constant editing to make sure that what's coming out of the company is coherent. 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 when I was there. The company needs to like this before it goes to customers.
Yeah.
10. What Worries Keith Most Right Now
I feel like those things combined keep us moving pretty fast.
So cool. What's the thing that worries you the most?
I would say speed. We're in this really interesting environment. Maybe I'll share something from the past.
Before we started Lightfield, I read all of these G2 reports about people moving off a flavor-of-the-month CRM to Salesforce. I remember reading one about one of your portfolio companies, ElevenLabs, where they run one of these startup CRMs. They couldn't get that company to build dashboards fast enough. It was 4 months, and the folks at ElevenLabs were tired of asking for dashboards, so they moved to Salesforce.
That's actually what makes me the most paranoid. I think we've got this incredible wedge of being just the best CRM for new companies, and we have to build everything so those folks never feel the desire to go to the old world.
How do you actually prioritize on that? This is the craziness about the new world, and it goes back to your point on everybody being a generalist. Once upon a time, you'd have a 10-person product, design, and engineering pod—or pick your number, but not 100, not 2.
You'd have the product manager, you'd have the designer, and then you'd have 9 engineers. You would have things that would be booked out until 2028. Now anybody can just prompt their way into a product, right?
If everybody understands what we need to do, on the one hand, you could build everything much more quickly. But you still have to make tradeoffs: "Am I prioritizing this customer that I know? I have this amazing company, and they're going to churn unless I build this." That's a really good reason to build that thing, but it might not be a good thing because you have 15 other customers that you're not going to sign up unless you build the other thing.
This has been a problem since the B.C. era, but what have you learned and done differently in this AI era, where anybody can build anything and therefore you could theoretically have unlimited product managers and engineers in your company?
It's funny. At Lightfield, we built this skill where we look at the expansion potential of an account. I think, as a CRM company in a red-ocean space, we have to be an expansion company.
In a competitive space, you often don't get the initial land that you would have expected because it's just important to win every company. But you're like, by year 3 or by year 5, this is not going to matter because you'll be a huge company and we'll be an important part of you.
We take this perspective: What's the expansion value of this account over a 3-year time horizon, and how do we prioritize? I think because of that, we've leaned more toward building for our fastest-growing customers than our average customer.
At the same time, we have to build everything, and it's a time of being a little more maximalist than usual.
The other thing that's kind of interesting is my observation that AI is almost overhyped in Silicon Valley. I don't think it is, but it's massively underhyped outside this little region of the world.
Part of it is that people haven't tried things in years. It's, "Oh, I tried ChatGPT in November 2022 and it hallucinated something. Oh, that doesn't work." Then they just put it to bed. Now they think it's going to kill them somehow. [laughter]
But how do you think about the long term? It's related to Greenfield versus Brownfield, but you can get Silicon Valley, and it's kind of crossing the chasm into the rest of the country and the rest of the world. It's the same type of customers: They have products, employees, and salespeople; they sell things; they want reports. But they're not as plugged in. You can't just advertise on 101 and reach them.
How do you think about Silicon Valley overhype versus the rest of the world underhype?
I think for most of these system-of-record companies, you end up getting most of your revenue scale from the 50 miles out of here and more. I've always thought about this moment in company building as a trick to get reference logos.
We have customers that have raised $200 million and have 3 go-to-market people, and they're going to be huge one day. We just need to do amazing work for them so we can take their logos when we go to the rest of the world and be like, "All right, we've got healthcare," or "We've got fintech," or whatever it is.
I see the focus on Silicon Valley not as efficient revenue capture, but as efficient marketing capture to go out there and be like, "All right, we're excited to get into manufacturing. We're actually working with a bunch of great manufacturing companies here, and now we can do that for you."
What's funny is how customers think, right? It's like, "All right, you're using the exact same software, but somehow, if Joe is in a different industry than me and he's a super-happy customer, I want the exact same thing." I'll use the Lightfield example, so just bear with me. "I'm doing the exact same stuff, but I don't know if I can trust you because he sells toothbrushes and I sell Coca-Cola."
But if he sells toothbrushes, "Oh, I must have the thing that the other toothbrush seller sells." It's really peculiar. I always found that it's somehow related to referenceability. Customers just want to know that it works, like the old "Nobody gets fired for buying IBM."
They want to know that this works, especially for something that is a system of record, and they want to know that somebody who is a competitor, a friend, or somebody in the same vertical is using the product before they're willing to make the jump.
Do you find that that's true? You mentioned manufacturing versus X. At the end of the day, a customer is a customer. The context and the contact information, all that stuff is the same, but why do they care? Is there just so much trust associated with the product?
I think so. In many ways, your CRM is maybe harder to move off of than your bank.
Yeah.
And—
It's a Lightfield.
Yeah.
Yeah. Exactly. And then we make it easy.
Yeah, we've made it a lot easier.
But I think in terms of the mental overhead, you just do not want to choose the wrong one.
Yes.
Because of that, I think referenceability goes a long way. On the way in, I was telling Joe that we do great with healthcare and health tech.
Part of it is that we do really well in these complex deals where there are 50 different stakeholders. Our context engineering is done really well, and we were hardcore about security at the start. We signed BAAs, and we did penetration testing for years.
Now it's so easy for us to win a health tech deal, and I think it's because we have this network that we can reference and move on. I think that's just part of the game. If I imagine even myself, 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 I never have to think about this after I buy it.
Yeah. Going back to an earlier question, do you run into people in this crazy era saying, “No, I’m not going to use anything. I’m just going to build it myself”? Because that’s the craziest, right? It’s like, am I going to be around? Is the person working at my company who’s decided to vibe-code a CRM—it's free, right? But is the support free? The salary that I’m paying them isn’t free. Do you run into people who are trying to DIY, or is that kind of overtalked about?
I think it’s a little overtalked about. We heard it a lot when our ICP was the seed-stage founder.
Yeah.
We would often hear, “Look, we can either pay you X, or we can do this over 4 weekends,” and we were like, “Great.”
Good luck.
Good luck. You should try it. Call us in 5 weekends.
Yeah, exactly.
Then they were like, “None of this works, right? It’s hallucinating. It’s sending out bad emails.”
I think on the larger-company side, we don’t hear, “I’m going to build my own system of record.” We hear, “I’m going to build my own company brain,” a lot. We’ve had so many of these folks come back and say, “Actually, building a company brain or building a business world model is really hard.” I think maybe the hardest part of it is modeling the customers. “We tried, and we don’t like our results, so we’ll come to you now.”
Yeah. What are you most excited about for the future?
I think the thing I’m most excited about—it’s funny—is I’m stoked about Lightfield becoming your sort of crystal ball for scenario planning. The thing that gets me up in the morning is having folks use Lightfield to think about, “How many reps should I hire? What product should I build next? Where do I go?”
One of our customers who sells to enterprise discovered through Lightfield that he needs to build a mid-market product and built this whole new product line because of this thing he found. I’m just excited about us being that sort of mirror.
Yes, for a company to make its hardest, most consequential decision. I think this goes back to the point Alex was making: in this world of maximalism, where you can build for anyone, actually the answer is having a tool like Lightfield. If you just have all of this data and all this context, you can actually be data-driven, and this is the exact opportunity to throw frontier intelligence at a really complicated decision.
Historically, this would have required really smart people, a lot of operations folks, SQL, and a bunch of other things. Instead, now you just talk to Lightfield for an afternoon or a weekend and come up with a really intelligent plan to move forward. I think that’s one of the reasons you can run your company the way you run it, which is really cool.
Maybe my last question for you is this: There are a lot of people—I think there will at least be a handful of people—who are going through some pivot moment, watching this, and thinking about the future. They’re in Keith’s position, talking about their new, amazing company. What would be the 1 piece of advice you’d go back and give yourself if you were just starting the pivot journey again, with the benefit of hindsight?
I think the most important thing to remember is that almost none of the noise around you matters when you’re in a pivot. You just need to find pain. You need to be inspired to build a product or service that solves that pain, and you need to be maniacally focused on your customers. Then the rest is total noise.
I remember when we were going through this, people were talking about how our office reminded them of the good old days. They were talking about how the food wasn’t inspiring. They were talking about, “Well, how are my options going to get repriced?” Honestly, none of that matters. I think you can just put the blinders on and focus on the core.
Well, we could not be more impressed with what you’ve done and the business you’ve built. The product is absolutely incredible. If you haven’t tried it, you need to try it. It’s incredible. Go check it out at lightfield.app. Keith, it’s an absolute pleasure to work with you, and we’re so excited for the future.
Thanks for the honor. Thanks for having us.
Yeah. Thanks, Keith.