Gabriel Paunescu

A teenager cornered the wine-yeast-oil market. Then he came for enterprise software.

By nineteen, Gabriel Paunescu's contracts represented roughly half the world's trade in wine yeast oil. He had started two years earlier, at seventeen, brokering essential oils out of Romanian factories. Most people learn supply chains from a textbook. He learned them by living inside one, calling factory managers, and discovering that the slowest part of any business is rarely the work - it is the software wrapped around the work.

That insight never left him. Today he runs Naologic, a no-code platform that lets a company build and reshape its own enterprise software - ERP, CRM, billing, dashboards, workflow apps - without hiring a development team. The pitch is blunt: connect your data sources, sit AI on top of the systems you already paid for, and stop waiting on a roadmap that belongs to someone else.

It is easy to file him as another no-code evangelist, and easy to be wrong about it. The no-code field is crowded with tools that make simple things simpler and complicated things impossible. Paunescu aims at the opposite end - the gnarly, regulated, mission-critical software that companies usually assume only a development team can touch. He has spent years on the parts of that problem nobody demos: workflow automation, real-time financial journal entries, inventory segmentation, the quiet machinery that breaks expensively when it breaks.

What Naologic actually does

Most enterprise software ships as one frozen application that every customer rents and nobody owns. Naologic inverts that. Everything is configurable through a visual interface, and the platform generates a unique source code base for each customer that matches their existing data models. A manager can deploy an app for every business unit without a developer in the room.

The newer chapter is AI. Naologic's apps plug into legacy ERP systems, ecommerce platforms, and other environments to capture data and apply AI across the business - what Paunescu calls "business autopilots." The work has been notable enough to land him in a Google Cloud case study and a MongoDB case study built around his AI retrieval-augmented-generation framework.

The detail that matters is the one most vendors hide: there is no single Naologic app. The platform spins up a distinct codebase per customer, shaped around the data models a business already runs on. That is closer to a tailor than a department store, and it is the whole argument. If your operations are unique, the reasoning goes, your software should be too - and changing it should not require a quarter of engineering time and a budget approval.

A blunt thesis about supply chains

Paunescu keeps returning to supply chains, and it is not nostalgia for the oil-broker years. It is where the no-code argument bites hardest. A warehouse manager knows exactly which transfer rule is broken or which inventory segment is mislabeled. Under the old model, that manager files a ticket and waits for a developer who has never seen the warehouse. Naologic's bet is that the person closest to the problem should be able to fix the software themselves, in an afternoon, through a visual interface - no implementation project, no consultants billing by the hour.

He has made this case on podcast after podcast, from "This Week in Innovation" to supply-chain shows where the audience is operators rather than engineers. The message lands because he speaks both languages. He can talk negative-stock reconciliation and demand forecasting with a logistics lead, then turn around and talk prompt failure modes with an AI researcher, without changing register.

The engineer underneath the salesman

Before Naologic, Paunescu built things that did not look like SaaS at all. He created DragonJS, an event-driven TypeScript framework for spinning up backend applications fast. Under his mentor Reynaldo Gil, he developed a device-mesh technology that used IoT sensors, local analytics, and autonomous swarms to cut industrial water and gas consumption. That work put him alongside Siemens, Pacific Gas, and Huawei - companies that do not buy ideas, only results.

So when he talks about AI, he talks like someone who has shipped, not someone who has watched. His view of large language models is unsentimental: they are tools with specific failure modes, not magic. His teams have taught models to debug their own hallucinations and built systems that catch when a model is trying too hard to please the human asking the question.

That last idea is more subversive than it sounds. A model that flatters its user is a model that lies politely. Paunescu's interest is in the opposite reflex - software that knows the boundary of its own competence and says so. For enterprise systems, where a confident wrong number can move real money, that humility is not a nicety. It is the difference between an autopilot you can trust and a demo you cannot.

The architecture he favors is multi-agent rather than monolithic: several specialized agents, each with a narrow job, checking and routing around one another instead of one large model asked to do everything at once. He presented exactly this at the Arize Conference, and it is the spine of the RAG framework MongoDB built a case study around. It is also the natural extension of the DragonJS instinct - small, event-driven pieces that compose, not one giant block of logic.

From the factory floor to the main stage

The Romanian kid who exported oils now keynotes for forty-five minutes on the main stage at Google Next in Las Vegas. He judged the Mistral Hackathon in Paris. He presented multi-agent RAG architectures at the Arize Conference. He lectured at Draper University on the unglamorous-but-essential craft of JSON prompt engineering. And he gives a year of his time to UC Berkeley, mentoring computer science students through the messy work of building their own startups.

There is a through-line. Every stop is about handing builders more leverage - whether that builder is a Berkeley sophomore, a supply-chain manager who has never written a line of code, or an AI agent that needs to know when it is wrong. Paunescu's whole career is a long argument that the people who use software should also get to shape it.

The Berkeley commitment is the tell. A year is a lot to give when you are also running a venture-backed company. He spends it walking computer science students through the unglamorous middle of starting up - the part after the idea and before the success, where most founders quietly drown. It is the same generosity that shows up in the Draper University lecture on JSON prompt engineering, a topic no one keynotes for the applause. He teaches the plumbing because he believes the plumbing is where leverage actually lives.

He studied at Stanford in the mid-2000s, has founded seven companies, exited two, and holds patents as a named inventor. But the resume is almost beside the point. What he is really selling is a posture: refuse the frozen product, demand the editable one, and assume the tools should bend to you rather than the other way around.

It is a contrarian bet in a market that has spent decades convincing buyers to accept what they are given. Whether enterprises fully follow him there is still being written. But the argument is clear, the track record is real, and the founder making it has been betting against the default since he was seventeen.

Ask him for a tidy origin story and you get a market statistic about wine yeast oil instead. That is the tell. He does not lead with where he came from or how impressive any of it sounds. He leads with the strange, specific detail and lets you do the math. The horse-versus-car line is the same move in miniature: not a mission statement, just a small true thing that, once you see it, you cannot unsee. Most of his work runs on that bet - that the right specific detail beats the grand general claim, in software and in people both.