In June 2026, Gumroad founder Sahil Lavingia posted a chart with an unusual milestone: for the first time, his company spent as much on AI as it did on the people who work there. Both numbers landed at roughly $43,000 for the month. One covered Gumroad's small human team, the other paid for the AI models the company used.
Five years earlier, in June 2021, Gumroad's monthly payroll was $419,000, and its AI spending was essentially zero.
What happened in the years between those two numbers turns out to be more complicated than the chart alone suggests.
The Timeline Behind the Milestone
Gumroad's team was not always this small.
Founded in 2011, the company raised roughly $8 million from investors, including a Series A led by Kleiner Perkins, and built out a full venture-backed team. That model didn't last. After failing to raise another round, Lavingia drastically downsized the company. TechCrunch reported in November 2015 that Gumroad had started the year with 22 employees and expected to retain around three.
The downsizing eventually helped Gumroad settle into a much leaner, profitable operating model. The team remained small after that, while Lavingia moved to Utah and stepped back from running Gumroad as a conventional venture-backed startup.
By January 2021, Lavingia was describing Gumroad in public as a company with no full-time employees at all, not even himself, counting 25 people total once every contractor was included, while revenue grew 85% year over year to $11 million annualized.
Gumroad was already operating with almost no traditional employees three to four years before generative AI tools became capable of writing usable code or handling customer support. The company's own SEC filings back this up with harder numbers: a 2020 annual report listed exactly one employee and $9.2 million in revenue for that year, up from $4.9 million in 2019, with net income more than tripling from roughly $288,000 to just over $1 million.
Monthly Payroll Against Monthly AI Spending
June of each year, as plotted in the founder's own chart. The lines meet in 2026, but note where the steepest payroll drop falls.
Source: Lavingia's own chart, posted June 2026. Bar widths are scaled within each column, so payroll and token spending are not plotted against the same axis. None of these figures are audited.
The steepest early drop in payroll, from $406,000 to $258,000 between 2022 and 2023, happened while AI spending was still just $689 a month. That matters because much of Gumroad's payroll reduction had already happened before AI spending became significant.
The most recent figures push the headcount even lower than the June 2026 snapshot suggests. In a reply posted the following month, Lavingia described Gumroad's entire human payroll as three part-time maintainers and two project-based consultants, plus himself at roughly one hour a week.
What Gumroad Actually Uses AI For
Software development
In February 2025, Lavingia announced Gumroad would stop hiring junior or mid-level software engineers. He published "tokens per codebase" figures for Gumroad and several sister products it also runs, including Flexile (contractor payroll) and Helper (customer support), predicting AI would be writing all of Gumroad's own code by the end of 2026.
By late 2025, Lavingia was demonstrating what he called a three-tier workflow: small tasks handed from Slack straight to an AI coding agent, medium tasks moved from GitHub to production, and larger projects built from a blank slate to a shipped feature, all with minimal human coding involved. To push adoption further, Lavingia also described a $33,000 engineering bounty designed to incentivize developers to outperform his own AI-assisted output using tools such as Cursor and Devin.
Gumroad's codebase is also public on GitHub, and the company now pays external contributors bounties for merged pull requests through Flexile. In another post, Lavingia said the company had 41 open bounties worth a combined $119,400 available at once.
Rather than employing a large salaried engineering team, a meaningful share of Gumroad's code now gets written by outside contributors working task by task and paid per contribution rather than per hour.
Customer support
Gumroad said its support SLA had dropped from 24 hours to 15 minutes, while Lavingia reported an average response time of about two minutes. The company also introduced an automated customer support system that gives users a $25 credit when a reported bug gets fixed. Gumroad capped the program at $5,000 a month and designed it to issue credits without human approval.
Asked directly what changed after the shift toward AI-handled support, Lavingia pointed to a mix of measures: customer satisfaction scores, other companies reaching out to copy the approach, product quality, and new creator sign-ups picking back up after a period of decline. These claims come from Lavingia. We found no independent review or customer data that corroborates them.
Rebuilding the codebase around AI
Reporting on Lavingia's workflow describes him deliberately simplifying the codebase, deleting thousands of lines of older CSS and adopting more standardized component libraries, specifically because cleaner, more predictable code performs better when an AI model is the one editing it.
He has also described moving away from lengthy product requirement documents in favor of working prototypes as the de facto specification, on the reasoning that when a feature can be built in two hours, a ten-page planning document stops making sense.
The Financial Picture Isn't a Straight Line
Public estimates of Gumroad's revenue vary by source. Sacra estimates that revenue rose from about $11 million in 2022 to $21 million in 2023, while GetLatka reports $23.8 million for 2024. These figures are third-party estimates rather than audited financial results.
Sacra attributed much of Gumroad's 2023 revenue jump to a change in how the company charged creators. Early in 2023, Gumroad moved from a variable transaction fee of 3.5% to 8.5% to a flat 10% fee on every sale, a change that pushed monthly revenue from roughly $1 million to $1.8 million almost immediately and flipped the company from burning cash to a net profit of about $9 million for the year.
Revenue Rose While Creator Volume Fell
Gross merchandise volume is the money flowing to creators. Revenue is Gumroad's cut of it. The two moved in opposite directions.
Source: Sacra estimates. GMV and revenue are plotted on separate scales and are not directly comparable with one another. These are third-party estimates, not audited results.
In other words, the pricing change appears to have been a major driver of the revenue increase, separate from any savings Gumroad achieved by operating with fewer people.
The same analysis found that gross merchandise volume, the total amount of money actually flowing to creators through Gumroad's platform, fell from a pandemic-era peak of $185 million in 2021 to $171 million in 2023. Meanwhile, Sacra reported that monthly revenue was down roughly 7% year over year as of July 2024.
Between 2021 and 2023, Gumroad's estimated revenue and profit increased while Sacra estimates that GMV declined, suggesting that a higher take rate contributed significantly to the improved financial picture.
Equal Spending, Unequal Proof: What Did AI Actually Change?
The $43,000 against $43,000 chart is compelling because it reduces Gumroad's transformation to one striking comparison: AI spending had caught up with human payroll dollar for dollar.
Gumroad's actual financial history tells a less tidy story than that image suggests.
The clearest revenue jump in the company's recent history, the near-doubling of monthly revenue in 2023, coincided with a major change to how much Gumroad charged creators. Sacra attributes much of that increase to the new pricing structure rather than changes in staffing or AI use.
That pricing decision happened before AI token spending was even a measurable line on the chart.
Payroll was shrinking during a period when Gumroad's estimated GMV also fell, from $185 million in 2021 to $171 million in 2023. Paying humans less did not, on its own, translate into Gumroad earning more.
Gumroad's tiny workforce predates the current AI push. The company was already describing itself as having no full-time employees back in January 2021, three to four years before generative AI tools could write usable code or handle customer support on their own.
Gumroad didn't need AI to become a tiny company. Its shift toward a leaner operating model began with the 2015 downsizing, long before token spending became a line item. What AI appears to have done is compress an already-small operation even further.
The strongest evidence that AI has replaced work inside Gumroad comes from Lavingia's descriptions of what changed day to day: faster customer support response times, an automated bug-bounty system, code increasingly written by AI agents instead of salaried engineers. Much of the evidence for those day-to-day changes still comes from Lavingia's own posts and interviews. Independent evidence showing how much AI, rather than Gumroad's longer-running lean operating model, contributed to the reduction in payroll remains limited.
None of this means AI played no role in Gumroad's transformation. The company is clearly using AI to write code, automate support, and operate with fewer people. What the evidence doesn't show is that AI created Gumroad's lean workforce or caused its biggest recent revenue gains.
The $43,000 milestone is still significant. It shows just how far Gumroad has shifted its spending from traditional labor toward AI. Whether that shift produces a stronger business is a question the current numbers cannot yet answer. For another small team running a business far larger than its headcount suggests, see our investigation into how Submagic reached $8M ARR with 13 people.
