Software program groups are transferring sooner than ever, but the programs they construct and preserve had been by no means designed for AI brokers to function inside them. Codebases span tens of millions of traces, context is scattered throughout dozens of instruments, and important data lives within the heads of some senior engineers.
Potpie was constructed to vary that. The corporate introduced a $2.2 million pre-seed spherical to assist engineering groups unify context throughout their total stack and make AI brokers genuinely helpful in complicated software program environments.
The spherical was led by Emergent Ventures with participation from All In Capital, DeVC and Level One Capital. The capital will probably be used to help early enterprise deployments, broaden the engineering workforce, and proceed constructing Potpie’s core context and agent infrastructure.
As generative AI adoption accelerates, most instruments concentrate on surface-level code technology whereas ignoring the deeper drawback of context. Giant language fashions are highly effective, however with out entry to system-level understanding, tooling historical past, and architectural intent, they wrestle in actual manufacturing environments.
Conventional approaches depend on senior engineers to manually maintain this context collectively, a mannequin that breaks down at scale and fails totally when AI brokers are launched.
Potpie addresses this by unifying context throughout your complete engineering stack and enabling spec pushed improvement. It pulls in info from supply code, tickets, logs, documentation, and opinions, hyperlinks it collectively, and makes it usable by brokers.
With Potpie, the spec turns into the supply of reality. Brokers plan the function finish to finish first by turning necessities into a transparent implementation plan, mapping dependencies and edge circumstances, and aligning exams and rollout steps earlier than writing a single line of code.
The precept is easy. An agent is just as efficient as the knowledge it could entry and the instruments it could use. Potpie focuses on each.
Aditi Kothari, CEO and co-founder of Potpie, says: “As AI makes code technology simpler, the actual problem shifts to reasoning throughout large, interconnected programs. Potpie is our reply to that shift, an ontology-first layer that helps enterprises actually perceive and handle their software program.”
The platform permits groups to automate high-impact and non trivial use circumstances throughout the software program improvement lifecycle, like debugging cross-service failures, sustaining and writing end-to-end exams, blast radius detection and system design.
It’s designed for enterprise firms with giant and sophisticated codebases, beginning at round a million traces of code and scaling to lots of of tens of millions.
Moderately than appearing as one other coding assistant, Potpie builds a graphical illustration of software program programs, infers habits and patterns throughout modules, and creates structured artifacts that enable brokers to function constantly and safely.
Potpie additionally actively creates context as programs evolve. When pull requests are created, it could replace documentation and tickets robotically.
When tickets are opened, it could generate system designs. The platform robotically generates structured habits definitions for every AI agent, outlining how they need to function inside a selected codebase.
On the identical time, it builds a searchable, tagged index throughout APIs, companies, databases, and parts, narrowing the search area and considerably bettering reliability.
The corporate was based by Aditi Kothari and Dhiren Mathur, who started engaged on the issue in October 2023, in the beginning of the primary wave of generative AI adoption.
Whereas a lot of the trade centered on data employees, they noticed that builders confronted a essentially completely different problem. Code is non-linear, deeply interconnected, and unfold throughout giant programs.
They spent practically two years constructing the foundational layer that understands codebases and creates the underlying data graph, earlier than launching Potpie publicly final 12 months in January 2025.
Early deployments replicate the dimensions of the issue Potpie is addressing. One buyer with a codebase exceeding 40 million traces diminished root trigger evaluation for manufacturing points from practically every week to round half-hour, with engineers appearing as reviewers as a substitute of investigators.
One other buyer sustaining decades-old programs used Potpie to replace and generate exams within the background, compressing work that beforehand took a number of sprints right into a a lot shorter cycle.
Anupam Rastogi, Managing Companion at Emergent Ventures, mentioned: “In giant enterprises, the actual problem shouldn’t be producing code, it’s understanding the system deeply sufficient to vary it safely.
“Potpie’s ontology-first structure, mixed with rigorous context curation and spec-driven improvement, creates a structured mannequin of your complete engineering ecosystem.
“This permits AI brokers to cause throughout companies, dependencies, tickets, and manufacturing indicators with the readability of a senior engineer. That’s what makes Potpie uniquely able to fixing complicated RCA, affect evaluation, and high-risk function work even in codebases exceeding 50 million traces.”
Potpie at the moment works with Fortune 500 and publicly listed firms in regulated industries, together with healthcare and insurtech. Its open-source initiatives have surpassed 5,000 stars on GitHub, creating a powerful magnet for enterprise adoption.
Kothari added: “AI readiness shouldn’t be about selecting the correct mannequin. It’s about constructing programs that may help intelligence over time. Our objective is to make Potpie the foundational layer engineering groups depend on to construct, function, and evolve complicated software program with AI inbuilt from the beginning.”
