From $750M in Sales to Starting Over: Razor's Founder on His New AI Bet

Tushar Ahluwalia has seen operational complexity at its most brutal. With Razor Group, he built a company that peaked at more than $750 million in sales, with 500 suppliers and over 100,000 shipments in a single week. Now he's starting over, building GeneralMind to automate exactly that kind of operational complexity with AI. His pitch is as simple as it is ambitious: "We buy tokens, we build a harness, and we sell an outcome."

In this conversation with Fabian, Tushar explains why he started GeneralMind with six co-founders, how he balances forward deployment with generalization, how GeneralMind prices its AI autopilots, and what it actually takes to reach $100 million in ARR.

Building a Founding Team for Speed

Tushar's core thesis on speed: it's not about working harder, it's about working in parallel — "a function of working in parallel on different topics at the same time." That's exactly why he started GeneralMind with six co-founders, even though that might look unconventional at first glance.

One concrete example of this principle in action: a deployment CTO and a platform CTO deliberately pull in opposite directions. One wants to ship as fast as possible for a single customer, the other wants to generalize the system. This tension isn't a bug — it's a feature. It forces the company to take both perspectives seriously.

Large founding teams, Tushar argues, only work if equity is properly differentiated and the first 100 hires can see a genuinely life-changing outcome ahead of them. Without that differentiation and without a real upside for early employees, the model falls apart.

Owning the Unstructured Layer

The idea behind GeneralMind grew out of a very concrete problem at Razor Group: 50 people between Berlin and China were manually typing supplier emails and WeChat chats into NetSuite. Tushar calls this "human glue" — people holding together systems that were never designed to fit.

His strategy at GeneralMind: don't replace the ERP, own the unstructured layer sitting on top of it. He describes this workflow work as "quite anthropic proof" — fairly resistant to pure model progress — because it demands real closeness to the customer. Having a better language model isn't enough; you need to actually understand how the customer works day to day.

Forward Deployment Without Turning Into Consulting

One of the biggest risks for a company like GeneralMind is doing too much customer-specific custom work. That quietly turns a product company into a consulting business stuck at a few million in ARR. Tushar is well aware of this trap and makes "time to value" the central KPI — the next deployment should take hours, not weeks.

The key to pulling that off is memory that compounds over time. One example from the field: an oat-milk customer's boat cafés only accept deliveries on Tuesdays and Thursdays. Once that kind of knowledge is captured, the system gets faster and more precise with every new, similar case. It's exactly this compounding of experience that turns individual projects into a scalable product.

Pricing and the Path to $100M ARR

When it comes to pricing, GeneralMind starts with a very concrete question: what does the payroll for this repetitive workflow cost today? From that FTE cost baseline, the price for the AI autopilot is derived.

Customers, according to Tushar, can expect at least a 10x return on investment. One particularly striking example: for a fleet customer, analyzing invoices revealed roughly 5% overbilling — a figure that would have simply gone unnoticed without systematic, automated review.

Asked what stands between GeneralMind and $100 million in ARR, Tushar is clear: "There is no demand constraint." The bottleneck isn't finding customers — it's generalizing fast enough, building solutions that work across more and more customers and industries without starting from scratch every time.

Takeaway

Tushar Ahluwalia's journey from Razor Group to GeneralMind is also a lesson in how to turn real operational pain points into a scalable product. Instead of replacing ERPs, GeneralMind claims the unstructured layer in between — the space where humans used to act as glue between systems. The real challenge isn't finding demand; it's balancing forward deployment and generalization well enough that individual customer projects turn into an actual product with network effects driven by memory.