Vishvesh Bhat, co-founder of CoreThink AI, leading the company’s enterprise strategy.
The playbook for AI startups rarely changes: build fast, launch wide, scale aggressively. For deep tech companies targeting enterprises, the path is less clear. Some chase broad adoption and run out of runway. Others narrow their focus so much they become consultancies rather than product companies.
Vishvesh Bhat, co-founder of CoreThink AI, chose a middle path. When his company developed General Symbolics, a hybrid reasoning engine, he opted against a broad launch. Instead, CoreThink pursued design partnerships with a small number of enterprises willing to co-develop solutions in production.
A year later, two partnerships have converted into commitments. The company has a seven-figure revenue pipeline and operates with nine people. Whether this validates a repeatable strategy or simply reflects good execution on specific cases remains an open question.
The Enterprise AI Pilot Problem
The statistics on enterprise AI adoption are sobering. According to MIT Sloan research, over 60% of pilots fail to reach production, primarily due to integration complexity and trust gaps. (MIT Sloan Management Review) The technology often works in controlled environments but breaks when exposed to messy real-world data and organizational friction.
CoreThink’s product is designed for complex reasoning in regulated environments where explainability matters. Finance, healthcare, supply chain, software development. These are verticals where AI adoption has been slower precisely because the stakes are high and error tolerance is low.
The challenge is proving value before running out of capital. Broad launches generate visibility but rarely revenue. Narrow pilots generate learning but consume resources. Vishvesh’s bet was that a handful of deep partnerships would be more valuable than either extreme.
The Mechanics of Co-Development
CoreThink’s design partnerships meant something specific: enterprise clients granted access to proprietary workflows and data, dedicated internal engineering resources, and provided detailed feedback. In exchange, CoreThink offered hands-on support, rapid iteration, and customization.
Vishvesh personally led these engagements, which is both a strength and a constraint. Having the CEO directly involved in pilots builds credibility and ensures quality feedback loops. It’s also a model that doesn’t scale beyond a handful of clients.
The approach yielded specific results: 30% error reduction in drug discovery simulations. 70% reduction in LLM spend. Integration time dropping from weeks to days.
These are meaningful improvements, though they come with caveats. The metrics are based on internal case studies rather than independent verification. The cost savings assume certain usage patterns. And the integration improvements may reflect as much about SDK design as the underlying reasoning engine.
The Conversion Question
Two design partnerships converted into commitments. That’s a 100% conversion rate if those were the only two pilots, or a more modest figure if others didn’t work out. The company hasn’t disclosed how many partnerships they pursued that didn’t convert, which makes it difficult to assess whether this approach is reliably effective or worked in these specific cases.
What’s clear is that conversions happened because Vishvesh maintained personal involvement throughout each pilot. Enterprise clients weren’t buying software from a sales team; they were buying access to the person who understood the technology most deeply.
The pilots addressed real operational problems: fraud detection, logistics optimization, predictive accuracy. Because each demonstrated measurable business impact, the business case for production deployment was straightforward. According to McKinsey research, integration complexity remains the single largest barrier to enterprise AI adoption, with 45% of organizations citing it as the reason pilots stall. (McKinsey) CoreThink’s approach addresses this directly, which likely explains much of its success.
The Scalability Problem
The tension CoreThink now faces is common among companies that succeed through deep customer relationships: how do you scale an inherently unscalable approach?
Vishvesh can’t personally lead every future pilot. The company will need to build a sales engineering team, develop automated onboarding, and create self-service documentation. Each step risks diluting the personal touch that made initial partnerships work.
There’s also the market size question. How many enterprises have problems complex enough to justify CoreThink’s reasoning engine but also have the technical sophistication and organizational commitment to deploy it? The addressable market for highly specialized reasoning systems may be smaller than for general-purpose AI tools.
What This Means for Other Founders
For other founders, the lessons from CoreThink’s path are situational rather than universal.
Design partnerships work when you’re solving genuinely novel problems for customers who understand they have those problems. They don’t work when you’re trying to create demand for something people don’t yet know they need.
Founder-led sales work in early stages when technical credibility is the primary barrier. They become a constraint once product-market fit is established and the bottleneck shifts to execution capacity.
Quantifiable metrics matter because they make value propositions defensible. But metrics need to be independently verifiable, or they risk being dismissed as marketing claims.
The Uncomfortable Truth About Growth
CoreThink’s path wasn’t flashy, and that’s often presented as virtue. But slow, methodical execution is only valuable if it’s building toward something that can eventually accelerate. The company has proven it can win individual enterprise clients through deep partnership. Whether it can do so at scale, with less founder involvement, and with positive unit economics remains to be demonstrated.
The AI industry celebrates companies that achieve rapid scale, but it also needs companies solving hard problems in complex domains. CoreThink appears to be in the latter category. The question is whether solving hard problems slowly is sufficient in a market where capital, talent, and customer attention are finite resources.
Two commitments and a seven-figure pipeline represent tangible progress. Whether they represent the foundation of a scalable business or a successful consulting practice with product characteristics is something only time will reveal.
