Pancake Built AI Workers for Itself. Then Customers Wanted Them Too

Some startups begin with an idea for a product. Pancake began with something its founders built because they needed it themselves.

Before Pancake existed, its founders were running another AI startup called Basalt. They were spending so much time working with customers that running their own company was becoming increasingly difficult. Their solution was to build AI workers that could take over some of the work they did not have time for.

The agents handled things like outreach, research, content and internal coordination. They worked in the background while the founders focused elsewhere.

Then something unexpected happened: customers started becoming interested in the system Basalt had built for itself.

Eventually, that internal experiment became more interesting than the original business.

Basalt became Pancake.

It started with a completely different startup

Pancake was not the original plan.

Guillaume Marquis and François de Fitte were previously building Basalt, an AI engineering platform designed to help companies test, evaluate and monitor AI products. Basalt itself had emerged from a problem Marquis had encountered while working on his previous startup, where testing and changing prompts was a slow and messy process.

The company gained enough momentum to attract venture funding. In December 2025, Basalt announced a $5 million seed round led by Entourage and Peak, with participation from Alpha Star, Kima Ventures and Hexa. Investors said at the time that Basalt was already being used by companies including Swan and HealthHero. (hexa.com)

But while the team was building software for other AI companies, it had another problem closer to home.

The founders were spending large amounts of time with customers, including travelling between Paris and San Francisco. There was still a company to operate while they were doing all of that.

So they began automating their own work.

Using AI agents, they created a system capable of handling recurring tasks such as research, outreach, content and internal coordination. Instead of opening an AI chatbot every time something needed to be done, the idea was to give different agents ongoing responsibilities.

The distinction mattered.

A normal AI assistant waits for someone to ask it a question. The system Basalt was experimenting with was designed to keep working without requiring a founder to continually start every task.

According to Marquis and de Fitte, it worked well enough that customers started asking about the internal system they were using.

That created an unusual situation: the software being used behind the scenes was attracting attention away from the product the company was actually selling. The founders ultimately decided that this was the more interesting opportunity.

Basalt was turned into Pancake. (Pancake)

The first Pancake was much broader

When Pancake publicly launched on May 28, 2026, the idea was more ambitious than the product it sells today.

The original version was essentially presented as an AI organization that lived inside Slack. Companies could add different AI agents and give them responsibilities across areas including growth, product and operations.

Pancake described the system almost like an organizational chart filled with AI workers.

The company was also using the product on itself. At launch, the founders said they had 23 autonomous agents completing three tasks per day each and sending the results back into Slack. One example involved an AI product team listening to feedback from a customer call and creating a software pull request afterward without being specifically asked to do so. (Product Hunt)

The launch generated some early attention.

Pancake finished as Product Hunt’s number-one Product of the Day on May 28 and number four for the week. It received more than 600 points on the platform and has since accumulated around 1,200 Product Hunt followers. (Product Hunt)

But Pancake did not simply leave the product there.

Over the following months, it became considerably more focused.

Instead of trying to provide AI workers for almost every part of a company, the current version concentrates on one problem that nearly every small business understands:

Finding customers.

Now the AI workers have one main job

Pancake now describes itself as an AI team for small B2B companies.

In simpler terms, it is software designed to continuously look for potential customers and help start conversations with them.

A business first gives Pancake information about what it sells, who its ideal customers are, its positioning, common objections and other relevant information. Pancake combines this into shared knowledge that its different AI agents can use.

Those agents can then look for signals that suggest somebody may be interested in what the company sells.

Instead of simply producing a huge database of random contacts, the idea is to identify people showing some kind of relevant activity. Pancake says its agents can monitor buying signals, qualify potential leads and prepare personalized outreach based on what they find.

The system can also help companies create content intended to appear in Google results and answers produced by AI search tools.

Importantly, the agents are not supposed to operate as completely separate tools. Information generated by one part of the system can be fed back into the shared knowledge used by the others.

If certain types of companies repeatedly respond positively, for example, that information can influence how future prospects are selected and approached.

That is much closer to the original internal experiment than simply adding another AI writing tool to a sales dashboard.

Pancake is trying to turn a job into software.

A five-person company building an AI workforce

There is another part of the story that makes Pancake particularly interesting: the company is applying the small-team philosophy to itself.

Pancake currently lists just five people on its team.

Guillaume Marquis is co-founder and CEO, François de Fitte is co-founder and COO, Tristan Comte handles go-to-market work, while Zakaria Benhadi and Théophile Cousin are founding engineers. The company is based in San Francisco. (Pancake)

That small headcount is deliberate.

Marquis and de Fitte have argued that AI could allow companies to become much larger without building the kind of large workforce that would traditionally have been required.

Their argument is not simply that AI makes employees slightly faster. It is that some recurring responsibilities can increasingly be handed to software altogether, leaving humans to concentrate on areas such as strategy, relationships, creativity and judgment.

Pancake itself provides a small example of that idea.

The company inherited the funding and history of Basalt, but rather than immediately building a large team around its new direction, it is operating with five people while developing software specifically intended to let other small teams do more with fewer employees.

There is an obvious marketing benefit to that story, so the company’s claims about what AI can replace should not be taken as independent proof that companies no longer need employees. Pancake’s own website makes some aggressive arguments about founders delaying early hires and running more of their companies with AI.

But the underlying experiment is still worth watching.

A five-person startup is building AI workers while simultaneously trying to use AI to remain a five-person startup.

$99 a month changes the comparison

Pancake’s current price also says a lot about where the company wants to position itself.

The product costs $99 per month, with its current customer-finding agents included in the plan. (Pancake)

That puts Pancake in a very different category from hiring another employee.

It also makes the proposition easy to understand.

A founder does not have to decide whether an AI agent can completely replace a salesperson or marketer. The more immediate question is whether software costing $99 per month can find enough relevant prospects, start enough useful conversations or create enough visibility to justify keeping it running.

If it can, the economics become interesting very quickly.

But price alone does not prove that the software works.

Finding a possible customer is relatively easy. Finding the right customer at the right moment is harder. Writing an outreach message is easy. Writing one that somebody actually wants to answer is much harder.

The same applies to automated content. Producing more pages for search engines does not necessarily mean those pages will rank, attract the right visitors or create customers.

Those are the areas where Pancake ultimately has to prove itself.

Its early Product Hunt reviews are positive, but the sample remains small. Product Hunt currently displays only a handful of reviews, including comments from early users who say they have used the product for autonomous marketing and sales work. That is encouraging early evidence, but it is not the same as a large body of independent customer results. (Product Hunt)

The company’s Product Hunt customer page currently highlights Guideflow, while Pancake’s own material naturally makes broader claims about how its agents can be used. (Product Hunt)

For a product launched only a few months ago and then significantly repositioned, that uncertainty is understandable.

It also makes the next stage more interesting.

The pivot may be more important than the product

There are now countless startups adding AI to sales, marketing and customer acquisition.

That alone does not make Pancake unusual.

Its origin does.

The founders did not begin by deciding that businesses needed another AI sales platform. They were building an entirely different company, created internal AI workers because they had too much work to do, and discovered that people around them were interested in those workers.

Then they changed the company around that discovery.

Even Pancake’s evolution since its May launch follows the same pattern.

The first version tried to cover much of a company’s organization with AI agents. By late August, Pancake had narrowed its focus toward customer acquisition, and its older articles were updated to explain the new positioning. Today, its agents concentrate on finding potential buyers, outreach and visibility in traditional and AI-powered search. (Pancake)

That is a much easier proposition to test.

A startup claiming it can provide an entire AI company creates dozens of questions about what the software actually does. A startup claiming it can help find customers has a much clearer measure of success.

Either it produces useful opportunities or it does not.

The larger idea behind Pancake has not disappeared, however.

The founders still believe that future companies will be able to operate with unusually small human teams because AI agents will perform more of the repetitive work that once required additional employees.

Pancake is simply starting with one of the areas where the result matters most.

For an early-stage company, saving time is useful. Writing faster is useful. Automating administrative work is useful.

Finding another customer is directly connected to whether the company survives.

That makes customer acquisition an interesting place to test whether autonomous AI workers are becoming something more than another layer of productivity software.

Pancake began by building AI workers because its own team needed help running a startup. Customers noticed, the founders pivoted, and the internal system eventually became the product.

Now Pancake has to demonstrate that the AI workers useful enough to change its own company can find customers for everyone else’s.