AI Agents Replacing Administrative Tasks in Small and Medium-Sized Medical Institutions: The Challenge of Lassie

▶ Original Video: https://www.youtube.com/watch?v=Fpg_8aiBxx4

July 30, 2026

Overview

This episode features "Lassie," a startup that automates administrative management for small and medium-sized medical institutions, including dental clinics, using AI. Co-founders Stein and Frederick, along with representatives from a16z, took the stage to discuss the origins of the business, its technical approach, and its market strategy. Rather than providing software merely as a database for information, Lassie aims to build agentic AI that replaces actual administrative labor itself. Against the backdrop of severe labor shortages in small businesses and inefficient paper-based workflows, the company has begun its expansion from the dental market, with a vision to eventually autonomize the operations of all small and medium-sized enterprises.

Silicon Valley Overvaluation and Iowa Undervaluation

AI is overvalued in Silicon Valley but undervalued in Iowa. Traditional software merely replaced paper documents with databases on green-screen terminals, while the actual work still had to be performed by humans. What Lassie provides is an agent capable of taking over dozens of hours of labor, which medical professionals considering adoption immediately accept as a means of clinic operation. As Dr. Quan put it, Lassie does not replace humans, but rather frees them from a state where they are forced to take on too many roles.

The Origin of the Founding: Meeting Dentist Dr. Quan

The founding of Lassie began when co-founder Stein was shown the behind-the-scenes of his own clinic management by his attending physician, Dr. Quan. Stein had moved from Amsterdam to Silicon Valley in search of a problem to build a business around, and at the time, he was in charge of the growth division at Robinhood. He was deeply shocked to witness the reality that Dr. Quan, who had the highest Yelp rating and whom he visited twice a year, was spending 200 hours a month chasing insurance claims and paperwork, and was working overtime himself due to a labor shortage. Stein's mother worked in a hospital in the 1970s, carrying cash to the bank and processing payments by hand, and he realized that the same problem remained unsolved in modern America. Note that Stein did not have any cavities at the time, and he was able to hear Dr. Quan's story after the examination, not during treatment.

Universality of the Problem: A Gastroenterologist in Scranton and Hundreds of Thousands of Small Businesses

To verify whether Dr. Quan was a special case, both founders began talking to other doctors. When working under a gastroenterologist in Scranton, Pennsylvania, they confirmed the exact same reality of manual labor. It became clear that hundreds of thousands of small business entities were still performing administrative tasks manually, and they became convinced that this was a difficult but worthwhile problem to tackle.

Doctors' Pain and Gaining Full Access

It was an unusual proposal for Stein, from the growth team at Robinhood, and Frederick, who worked on product at Superhuman, to offer doctors that they would "take over billing operations and take charge of finances." However, all the doctors wanted to talk for hours about their problems, and it became clear that this was a pain that kept them up at night and made them want to quit their jobs despite their passion for patient care. Despite concerns about HIPAA, security, and lack of experience, Dr. Quan said, "You can come in at 5 o'clock every night and sit down to work," and Dr. Sha in Scranton also granted access to all information, saying, "I will teach you like my own son."

Evolution of Product Development: Automation Foundation Since 2020 and the AI Tailwinds

Lassie's business was launched in 2020. From the beginning, they were committed to business automation and self-driving, but the models at the time had low performance, and inference models in particular did not exist in the form they do today. Automating any job requires both the business context (such as clinic history data and patient records) and the tools to execute the actual work, which is no different for humans or agents. The company first built the context layer and the tool layer, but the intelligence layer at the time was immature. While advanced reasoning was not required for early basic automation, the models improved significantly thereafter, and the product became smarter by replacing the intelligence layer on top of the existing context and tools, which served as a major tailwind. The core vision is not the provision of tools, but the automation of the work itself.

Practical Limits of Large Models: The Sea of Data and the Void of Business Knowledge

Interestingly, despite being trained on vast amounts of data and being extremely large, the models do not inherently understand how to perform this type of work in practice. Although standard operating procedures and documents exist, such as how to bill specific insurers, the tacit knowledge in the head of an on-site office manager is not in a form that can be easily accessed on the internet. Lassie has the strength of being able to infer such workflows from the ERP history data it has accumulated. When they started using later generations of inference models, they assumed they would naturally know how to execute the work, but in reality, they did not grasp the fine nuances. To bridge this gap, the company is investing a great deal of effort into product UX that makes the agent smarter by obtaining input from staff regarding their preferences and past processing methods. Frederick also expressed hope for the emergence of smaller models that can learn quickly with less data, envisioning a future where intelligence is ubiquitous, like the Pixar lamp.

Origins and Limits of Software: From Paper to Database, Yet Manual Work Remains

The a16z representative points out that the origin of software was replacing paper filing cabinets with databases. Its roots began with Sabre Systems, a joint project between IBM and American Airlines, which digitized airline reservation management from paper. The fact that Sabre is spelled with two A's comes from the A in American Airlines. Subsequently, similar digitization progressed in every field: PeopleSoft in HR, LexisNexis in law, and Quickbooks or NetSuite in accounting. However, these only made information easier to access, and the actual work was still performed by humans. Comparing 1950 and 2000, the number of people in the HR department of a company of the same size has not changed; the guards protecting paper files have simply been replaced by IT departments and CISOs protecting data.

Market Opportunity for Software That Replaces Labor

The representative argues that current software has reached the stage where it can not only store information but also execute changes to databases—that is, replace the labor itself. In HR, it has become possible for software to perform background checks, onboarding, and benefit explanations; in accounting, it can make collection calls for unpaid invoices. In 2023, AI—that is, next-word prediction technology—could not possibly be entrusted with clinic operations or background checks based solely on statistical inference, but the situation has now changed completely, and the market size has expanded significantly. The representative explains that this is a structure similar to how fintech expanded the market by incorporating payment processing in the form of Toast. Toast could have existed in 1985, but a restaurant with 5 million dollars in annual sales would not have paid 100,000 dollars for MS DOS software. However, by integrating payment processing and earning a 2% fee, it became possible to earn 100,000 dollars in revenue, and the market was established. However, the automation of labor brings opportunities of a scale incomparable to that. What is important is the reality that for many small and medium-sized enterprises, it is not simply a matter of cost, but that they cannot find the necessary talent in the first place.

The Lesson of Dentist Sloop, Who Was Forced to Close His Practice

Dr. Ronald Sloop (75-80 years old, Ashkenazi Jewish), who was the primary dentist for an a16z associate, ran a clinic in Florida. After the key person in charge of his bookkeeping resigned, he could not find a replacement, so he sold the clinic to a younger doctor and retired. He has since moved to Southern California and spends his time playing golf and other activities. Upon learning about the announcement of Lassie, Sloop reportedly told the associate's father, 'If I had this, I wouldn't have retired.' This demonstrates that AI can resolve 'market failures' that occur not just because of cost, but because hiring talent is impossible. When something exists that costs $100 to manufacture while everyone on Earth is only willing to pay $1, that is a market failure; it is 'everything to the right of the supply-demand equilibrium point in an Econ 101 graph.' For example, the reason not every dental clinic in America has a Dutch-speaking receptionist is that while the probability of a patient coming in who only speaks Dutch is only 1 in 100, the clinic would still have to pay 40,000 euros in labor costs for that staff member. If it could be done for free or for $1, every dental receptionist would be able to handle Dutch. The software market is by no means small at $1 trillion, but surrounding it is the financial transaction market, where Visa holds a massive market capitalization, and further outside that exists the labor market, which is of an order of magnitude larger.

The Scale and Business Model of the Dental Clinic Market

There are approximately 160,000 dental clinics in the United States, each bearing about $200,000 in annual administrative costs. As Lassie already handles hundreds of clinics, it witnesses the daily reality of doctors with Harvard degrees being bogged down by administrative tasks until late at night, to the point where they are starting to hate their jobs. Small business owners want to spend their time on patients, not on administrative work or managing staff. However, they cannot even find that staff in the first place. Rather than AI taking jobs, in many cases, finding people is simply impossible. Doctors decide to implement Lassie very quickly after checking its reality by asking friends in study clubs or searching on Google. The reason this business works is that the cost is drawn from the labor budget on the profit and loss statement. The company is already charging five-figure fees for the first agent that provides 30 hours of labor per month, and there is still significant room for automating the 200 hours of monthly labor that Dr. Sloop once needed. Clinic owners recognize Lassie not just as a software tool, but as an entity to which they can outsource the operation of the clinic itself.

The Challenge of Building Autonomous Agents: Eliminating the Human in the Loop

In tasks like insurance payment reconciliation and patient billing, one cannot leave final responsibility to a human. Also, because small businesses do not have dedicated staff to master tools, the software must operate completely autonomously. Many current AI companies still use a 'human-in-the-loop' model where software engineers make the final deployment decisions, but Lassie cannot tolerate that. An agent that takes on insurance payment reconciliation and billing communication with patients must function perfectly, and it took years to build. Lassie took the approach of the founders themselves residing in clinics to perform actual administrative work and automate their own problems. The product works so well precisely because they knew the work inside out. Currently, a mechanism has been built where clinic staff provide instructions on cases the agent cannot handle, and it learns from them. Furthermore, the reason for building an agent that eliminates the human in the loop is to provide software that enables scalable and rapid implementation. In the launch video, Dr. Quan says that thanks to Lassie, he can now coach his child's soccer team and go watch their games.

Growth Through Word-of-Mouth from Dentist to Dentist

There is still a large gap between the desire to implement Lassie and actually introducing AI to handle payment processing. However, as a result of achieving results that bridge that gap, most of the growth has come from word-of-mouth, that is, recommendations from one dentist to another. The background to achieving such high performance is the existence of extremely high standards for shipping, which the founders cultivated in the world of consumer products like Robinhood and Superhuman.

Phased Automation Strategy: Business Optimization After Reaching 95%

Lassie does not aim for 100% automation with a single product, but rather takes a strategy of returning tasks to the clinic side once an automation rate of over 95% is reached, and then moving on to automating the next task. If a task that required 200 ledger updates per week is reduced to a few, it saves 10 to 20 hours, making management significantly easier. Reflecting on running his third company, Stein points out that just as finance, payroll, and HR tools have saved vast amounts of time compared to 10-15 years ago, even if they are not fully autonomous, there is no need to wait for complete automation. Currently, thousands of clinic staff provide input on the few percent of cases that agents cannot handle, contributing to further accuracy improvements. Also, Lassie adopts an approach of integrating with existing clinic management systems, so there is no burden on dentists to switch between numerous software programs.

The Fable of TiVo: Innovation Competition Between Startups and Incumbents

An a16z associate recalls that about four years after founding TrialPay, they realized they should have built a service like Stripe. At the time, Stripe had five employees, but they understood the structure of starting with a commodity business like boring payment processing to acquire customers ahead of time, and by owning that pipeline, they could build higher-revenue features like offer-based payments on top of it.

This lesson also applies to the examples of TiVo and ReplayTV. Both companies invented digital video recorders, but even with superior technology, they met harsh ends as businesses. When acquired by existing giants like Comcast, a discount on control is applied to the acquirer because competitors will ensure that the technology does not work. Even if a licensing agreement is signed, the incumbent with the customer base takes most of the economic benefits, or the incumbent releases a poor copycat product a few years later. As a result, the competition between startups and incumbents is decided by 'whether the startup acquires distribution first, or the incumbent acquires innovation first.'

In the AI era, incumbents can improve their bad engineers with AI, so their ability to imitate increases. For example, the 'great idea' of conducting background checks on employees on top of an existing massive HR information system like Workday is at extreme risk of being imitated by Workday. In the past, there was a company called X1 that provided excellent search to replace the terrible search in Microsoft Outlook, but it was highly likely to be imitated by Microsoft and ended up being acquired by Yahoo!. However, in Lassie's case, a major advantage is that there are no massive incumbents like Workday in dental software. The company competes with administrative staff like 'Betty,' who quit a few weeks ago, or external contractors.

Furthermore, building an ontology that unifies the definitions of insurance claims and patient payments across multiple systems requires years of work, which creates a strong defensive moat. The public market is evaluating that 'software is dead' and 'AI is wonderful,' which is the exact opposite of VCs, but AI is essentially software, and software is AI. It is tantamount to 'dereliction of duty' for existing software companies like Workday and NetSuite not to implement AI features when customers are demanding them.

Distribution Strategy for Small and Medium-Sized Businesses and Self-Service Adoption

How to reach busy, non-technical owners and get them to adopt it is a key challenge for Lassie. Dentists like Dr. Sloop live busy lives, returning to the office at 7 PM to handle emergency patients and having to order supplies themselves after dinner with their children because the staff member who resigned is gone. Therefore, making contact itself is extremely difficult. The company has built a self-service adoption process close to consumer products like Stripe and Rippling, and just as Robinhood pioneered the automation of account opening, it abstracts bank account linking and insurance portal authentication. Currently, it is approaching a stage where, if the doctor gives consent, the agent is set up almost automatically.

Data-Driven Market Entry: Mapping Dentists Nationwide and Proprietary Playbooks

Lassie's market entry is fundamentally different from what is considered the state-of-the-art method today, namely a model where sales representatives hold meetings with large clients to secure contracts and record $10 million in annual recurring revenue at once. The company must literally find thousands, tens of thousands, and hundreds of thousands of small business owners. To do this, they have developed a proprietary playbook that maps the locations, owners, systems used, and even intent signals like posting job openings on Indeed for dentists across the country, and makes contact with messages that cut through the noise and resonate. Conventional sales methods like enriching leads with Clay or Apollo and finding them on LinkedIn do not work because Dr. Sloop does not exist in such databases and is not on LinkedIn. An entirely new market entry model is required to deliver AI to mainstream American businesses.

The Persistence of Paper Payments and the Digitalization Revolution

Even today, about 70% of small and medium-sized enterprises receive payments via paper checks, and some insurance companies, like Cigna, intentionally do not offer online enrollment and instead send paper checks. With the federal government mandating a shift from paper checks to digital payments, a significant tailwind is blowing across the entire industry. Lassie has turned the process of converting paper payments to digital into a product itself, building an engine that executes the switch automatically using only business information and tax IDs. The complex psychological factor of staff preferring manual paper-based work was one of the reasons this transition had not progressed.

Interaction with the Insurance Industry: Claim Optimization and Incentive Alignment

AI agents accurately process the 50 to 100 types of procedure codes handled by dental clinics and submit them to insurance companies along with the appropriate X-ray images and treatment plans. Insurance companies have clear rules and simply make payments according to them. Furthermore, insurance companies have an incentive to keep high-quality dentists within their networks. This is because when employers renew their dental insurance plans each year, employees will move to another insurance company if there are no good dentists in the network.

Master Plan: From Dentistry to Autonomous Operations for All Small Businesses

Mr. Stein explained Lassie's vision in three stages. The first stage is a focus on the approximately 160,000 dental clinics in the United States. The market for automating approximately $200,000 in annual administrative costs per clinic reaches a $1 billion scale in recurring revenue alone. In the second stage, they plan to expand horizontally to other types of medical institutions, such as physical therapists. As a third stage, they aim to provide AI agents that autonomize the operations of all small and medium-sized enterprises based on the knowledge accumulated. Ultimately, they envision a world where corporate agents abstract away reading and writing to record systems, dialogue with patients and consumers, payments, and scheduling, allowing operations to proceed fully automatically while interacting with the AI agents of consumers and insurance companies. The ultimate goal of this vision is to support small and medium-sized enterprises around the world, not just in Iowa or Paducah, Kentucky, but also in Mr. Stein's hometown of Amsterdam and Mr. Frederick's hometown of Hamburg.

Organizational Building in the AI Era and the Transformation of Small Business Management

Mr. Frederick and Mr. Stein point out that the need for top-tier talent remains unchanged in the AI era. Ability and ambition in the top 5% of engineering or sales, as well as being AI-native, are factors they emphasize particularly during hiring. What they are aiming for is a 2026-version large-scale organization, with the goal of shipping twice the output of other companies and moving four times as fast. How to build a team that incorporates AI into every function, starting with the finance department, is a major challenge for the future. The fundamentals of running a company do not change, just as Olympic-level running requires training twice a day and effort to the limit; it is not something everyone can do.

From their experience at Superhuman, they learned to focus strictly on the right Ideal Customer Profile (ICP), as it becomes extremely difficult to churn if you onboard the wrong clinic and cannot automate their processes. In onboarding, they are obsessed with reaching core product value quickly, measuring checkpoints set like a movie script within a specific timeframe.

On the other hand, if anyone can hire a "Betty," it theoretically makes starting a small business easier, but it could also create a paradox where competition intensifies and management becomes even more difficult. While replicating software has become surprisingly easy with tools like Replit, Lovable, and Claude, replicating "Dr. Slope's clinic" itself is far more difficult. However, both are not pessimistic about this change. They believe that the demand for dental clinics and plumbers is twice the current supply, and that ensuring everyone can receive high-quality dental care twice a year, or realizing the kind of high-quality primary care they experienced in the Netherlands here in America, is a major opportunity for this technology. This is a market where demand far exceeds supply, the exact opposite of the situation described by the late Yankees legend Yogi Berra: "Nobody goes there anymore; it's too crowded."

Lassie is currently actively hiring, and details can be found at lassie.ai.

▶ Previous in the same series: The Next Frontier of AI: Spatial Intelligence—Fei-Fei Li on World Labs' World Models and Expansion into Robotics https://note.com/yondo/n/n27d1c5b1f1b8

#AIDevelopment #Podcast #a16z #Lassie #AIAgent #Healthcare #Dentistry #SMB_DX #BusinessAutomation #Robinhood #Stripe #Coinbase #GenerativeAI #LaborSubstitution #SabreSystems #DigitalPayment #SMB #InferenceModel #Ontology #TiVo #GoToMarketPlaybook #WordOfMouthGrowth #MarketFailure #LaborShortage #PrimaryCare