Daksh Gupta from Greptile - do marketing differently
Greptile founder Daksh Gupta explains why 50 speaking cookie boxes can beat millions of ad impressions, how a technical founder can learn sales like an optimization loop, and why AI DevTools above Git are sold to teams rather than individual developers.
Daksh Gupta is the founder of Greptile, an AI code-understanding API. His team has marketed it with homemade energy drinks, hand-delivered cookies and a willingness to test ideas that sound unreasonable in a boardroom.
The conversation moves from memorable marketing into founder-led sales, enterprise adoption and a framework for where AI changes software work without confusing today’s useful products with a hypothetical fully automated engineer.
The Q&A below is experimental — extracted from the episode transcript with AI. If you spot a mistake, let us know at jack@scalingdevtools.com.
What did the Greptile energy drink teach you about marketing?
We first sandpapered cold-brew cans, added our own labels and took them to a hackathon. People remembered them, so each successful iteration earned more effort: a custom lime drink, then a professionally designed label. The experiment made an important fact feel real—there are no fixed rules for how a developer tool earns attention, and a wild idea should be tested before internal skepticism kills it.
How do you judge the reach of an unusual campaign?
I think about the number of impressions multiplied by how many people retain them. Ads can buy a huge audience cheaply, but the memory may be close to zero. An energy drink or a box that plays Steve Ballmer’s “developers” speech reaches far fewer people, yet nearly everyone remembers it. Fifty carefully targeted boxes can therefore create more useful attention than a large, forgettable campaign.
Why did the quality of the cookies matter?
The small details are sticky and signal how much care sits underneath the visible surface. We could have used cheaper cookies, but then people would associate Greptile with an ordinary experience. Braintrust inspired us by delivering genuinely exceptional pizza rather than merely any pizza. Weeks later, people still discussed not only the gift but how good it was; that feeling became attached to the company that sent it.
How did you find the right offices for the campaign?
We combined Apollo, Clay and an AI spreadsheet called Paradigm. The companies went in one column, then agents researched an address and whether the team worked in person most days. That let us target existing customers and high-fit prospects rather than scatter the boxes. The process was a normal sales-data problem until AI helped complete the difficult last-mile research.
What changed your view of founder-led sales?
I began as a reluctant salesperson because all three founders were engineers and I was the weakest programmer. Sales became interesting once I saw it as finding the people who most need a good product and clearly connecting their problem to it. Every call also supplies an immediate training signal: if the buyer does not understand or care, the loss is high and you adjust the pitch, qualification or product before the next sample.
How do you navigate a larger customer?
The same roles recur: an engineering champion, a pragmatic executive, procurement and information security. Predictable pricing helps procurement, while security questions become repeatable once you have answered the first few thoroughly. At several hundred people, the graph becomes more complex. Rather than forcing the whole organization at once, find the smallest independent unit that can decide and expand piece by piece.
How has the market for AI developer tools changed?
A year earlier, many companies were only trying code generation, so a second-order product around review and internal AI tooling was difficult to sell. Once leaders experienced the productivity improvement, the conversation changed. They no longer needed convincing that AI could help; they needed to know whether this particular product was differentiated and good enough. That prior adoption and executive conviction dramatically shorten the sale.
What is the difference between AI tools above and below Git?
Below Git, an individual developer can choose an editor or coding assistant without everyone matching. Above Git, products such as source control, planning, incident management and documentation coordinate a team, so the organization must share them. That makes the development team—not just a developer—the customer. AI is now expanding from individual code generation into this shared layer where organizational work happens.
What creative work should remain with software engineers?
If a well-defined problem has an objectively correct answer, computers should eventually handle it. Human work remains where there is no single right result: defining the problem, choosing a form, shaping an interface and applying taste. Better abstractions have repeatedly removed apparatus from a craft without removing creativity. AI continues that path by automating refactoring, debugging and deployment so people can spend more time on choices that require judgment.
What advice would you give another founder?
Make something you or someone close to you genuinely wants, launch quickly and charge early enough to prove the demand is real. Choose excellent cofounders—even one very smart, trusted friend changes the journey. Finally, do not raise money merely because it is available. Take capital when it is the actual constraint; strong founders building a strong business will have opportunities to raise later.
Links & references
This episode is brought to you by WorkOS.