ATRIUMsearch → argument graph
Video · 2026-07-30 · 59m · 6 moments

“Every small business should run itself” | Lassie with a16z

✦ AI generated

timeline · colored by role

01
Claim

Software historically just digitized filing cabinets — it stored information but didn't do the work, so the world didn't get that much more efficient.

Alex argues that traditional software merely took paper-based filing cabinets and made them digital, but people still had to perform the actual work. The result was that efficiency barely improved — the same number of people were still needed.

transcript

Alex Rampell (a16z): I would actually argue that the world didn't get that much more efficient with software because all that software did was like take HR like did Peopleoft and then workday make HR efficient make make HR departments more efficient like I don't think so because the same number of people worked in HR for the exact same size company in like 1950 as probably 2000 and instead of using filing cabinets that are you know that are guarded by Stein and Frederick you know making sure that nobody breaks into the HR files. Now you have an IT department and a CISO to make sure nobody hacks into the IT files or the you know the HR filing cabinet. So nothing really got more efficient.

extends · 1

02
Mechanism

AI can now do the work — not just store information — and that expands the market massively because software can charge for labor, not just for storage or payment processing.

Alex explains that the real breakthrough is software that can execute tasks — edit the filing cabinet, not just hold it. This is orders of magnitude bigger than the storage layer, and far larger than the fintech bundling effect that expanded software markets before.

transcript

Alex Rampell (a16z): What you can now do with software is it can do it can edit the filing cabinet, right? It's no longer just the dumb storage. It's actually like the smart implementation of changes against those things. So like if it's HR, let's do a background check, right? Right? Or let's do an onboarding or let's explain the benefits to this person. If it's accounting, what do you do with the financial statements? Like imagine I'm a dentist and I see I have all these overdue invoices and I can look up look up in QuickBooks. What do I do? Well, I might want to call and say please pay me. Like that's what the filing cabinet should be doing and not just giving you the information. So it just turns out that the work is orders of magnitude bigger than the storage of information that the work is done on. Um so that that's really the been the thesis and that you know you need the technology to catch up so it can actually do it. ... this massively expands the market size. And if you think about fintech, fintech massively expanded the size of many non-financial markets because now you could you could bundle in financial products with non-financial products. ... fintech made the market much much bigger for software because of this bundling effect. Um, and that pales in comparison to now software doing the job of labor because it's like, yeah, fintech made it a little bit bigger, but now instead of just being a dumb pipe for data or a dumb storage of data and instead of just like charging incrementally more by bundling in, you know, financial processing, now we can do work and we can charge for work and we can charge for work at a way that is cheaper than humans, better than humans. But I think both of those sell the opportunity short because in many cases you can't even find a human.

extends · 2gives example · 1provides context · 1

03
Claim

AI is not going to take jobs — in many cases you can't find a person to hire at all, so the real market failure is everything to the right of the supply-demand equilibrium.

Alex counters the 'AI takes jobs' narrative with the reality that many small businesses simply cannot hire. He uses the example of a dentist who retired because his key assistant left, and frames the problem as a market failure — work that people would pay for but no one is available to do at any reasonable cost.

transcript

Alex Rampell (a16z): Part of why he retired was he lost his like, you know, key woman that did the books and everything else. He's like, I can't deal with this anymore. I quit. Yeah. ... And he said that if this had been around, he wouldn't have retired. ... So it's not like oh AI is going to take the jobs. In many cases you can't find somebody. This is the part that people don't realize or you can't find somebody but there's like a imagine that there is something that every human on earth would pay a dollar for but the cost of manufacturing that thing is $100. You just have a market failure and I kind of call this everything to the right of the supply demand equilibrium point on like an econ 101 graph.

explains mechanism · 1extends · 1

04
Claim

The battle between every startup and incumbent comes down to whether the startup gets distribution before the incumbent gets the innovation — and AI changes this because the incumbent can now innovate faster, but many categories had no incumbent software company at all.

Alex argues that the classic startup-incumbent dynamic is shifting: AI makes it easier for incumbents to copy innovations quickly, but many industries where AI does actual labor never had a software incumbent to begin with — the 'incumbent' was a human employee who quit. This creates a huge opening for startups.

transcript

Alex Rampell (a16z): The battle between every startup and incumbent comes down to whether the startup gets the distribution before the incumbent gets the innovation. But with many changes, one change is the incumbent can get the innovation much more quickly. Um uh but the other is that there are a lot of categories where there never was an incumbent software company because the only job to be done was like actual human labor. And that's really exciting because now you don't have to worry about like oh shoot these guys are going to come in and eat my lunch like who right like who does like there are a lot of industries that just don't have an incumbent software solution.

extends · 1provides context · 1

05
Prediction

The end goal is that every small business should run itself — busy work done by agents, with business agents interfacing with consumer agents and insurance company agents.

Stein lays out a three-step master plan: first dominate dental practices ($1B revenue opportunity), then expand to other doctor office types, and finally build AI agents that can run any small business. The ultimate vision is a network of agents — business-side, consumer-side, and insurance-side — all interoperating autonomously.

transcript

Stein (Lassie co-founder): Yeah. I think three steps like the end goal here is that uh every small business should run itself, right? And the busy work is done by uh agents. uh we want to build agent for the business that then will interface with the personal agent of a consu consumer highly likely that then will interface with an agent at the insurance company or other parties that the business needs to interface and interact like with. Uh but step one is to Alexis's point like there are 160,000 dental practices in the US alone $200,000 in labor that Dr. Sloop and others can't find. So like um serving that market first. You're looking at a $1 billion in like a recurring revenue as a market um so we're that's like step one and then uh likely like we will pick another doctor office type um that like is not well served and has a big temp um and and needs consumer-l like product right because what we discussed a big part of not this you get the AI to work 95% accurateish it needs to really work and the onboarding needs to be as simple as onboarding on Coinbase. um or stripe. Um so like it will likely be another like doctor office type like um and then [sighs] uh the last uh part there is I think uh we've then trained AI agents uh to like run the small business and all small businesses at an abstract level like have a system of record they need to read and write into. Um they all have customers in doctor offices they happen to be patients but it's interacting and transacting around payments. You need to book appointments. So I think the end goal is if we served all the the doctor offices that we help all the small businesses across the world because we're just the best in you know building AI agents that uh salt of the earth people or people in Iowa and Paduka Kentucky and hopefully in Amsterdam like down the line where I'm from um and Germany Hamburg where Frederick is from can can start like using as well.

explains mechanism · 2extends · 1

06
Mechanism

The models are trained on so much data and yet they actually don't know how to do any of this work — the workflows are not encoded in the training data, so you need to build the context layer and tools yourself.

Frederick explains that despite their vast training, models don't inherently know how to run dental billing workflows. The office managers' knowledge — how to submit claims to specific payers, what narratives to include — is not accessible on the internet. Lassie's advantage comes from historical ERP data and actually doing the work itself to infer those workflows.

transcript

Frederick (Lassie co-founder): I think one thing that's interesting is that the models are trained on so much data and they're they're so large and yet they actually don't really know how to do any of this work. Like they don't have the uh workflows encoded in any way. Um so for example, we're working on a product now where we we have to like collect all of these like basically SOPs on like and documents about like how are you supposed to bill insurance claims to certain payers and all this kind of stuff. uh which uh to some extent humans would do the same but there's also a big amount of um just like human knowledge that is encoded in say these office managers and they just like know how to do this work uh that's weirdly not that accessible on the internet um I think we have a big advantage there because we have um all of this like historical data out of their ERPs that we can look at and kind of infer you know some of these workflows from um but that's something we notice a lot I think actually when we started using some of the, you know, later reasoning models, uh, we kind of assumed like, oh, they probably just know how to do this work because like why would they not, right? Like they're trained they're trained on on all of this data. Um, but it it it turns out that they they don't know all the intricacies of most of these workflows.

extends · 1rebuts · 1

Highlight slides
Software digitized the filing cabinet, not the work✦ from: Software historically just digitized filing cabinets — it stored information but didn't do the work, so the world didn't get that much more efficient.Headcount unchanged — HR shifted from filing to cybersecurity✦ from: Software historically just digitized filing cabinets — it stored information but didn't do the work, so the world didn't get that much more efficient.Software shifts from storage to execution✦ from: AI can now do the work — not just store information — and that expands the market massively because software can charge for labor, not just for storage or payment processing.The labor market dwarfs prior expansion waves✦ from: AI can now do the work — not just store information — and that expands the market massively because software can charge for labor, not just for storage or payment processing.End goal: every small business runs itself✦ from: The end goal is that every small business should run itself — busy work done by agents, with business agents interfacing with consumer agents and insurance company agents.Step 1: Dominate dental practices✦ from: The end goal is that every small business should run itself — busy work done by agents, with business agents interfacing with consumer agents and insurance company agents.Steps 2-3: Expand then generalize✦ from: The end goal is that every small business should run itself — busy work done by agents, with business agents interfacing with consumer agents and insurance company agents.
Related episodes