Introducing Kos.ai: The World's First AI Accountant

Over the last eighteen months, the concept of an agent has evolved in real time. Early on, it was largely a rebranding of workflow automation: deterministic pipelines with a language model somewhere in the loop. Useful, but not qualitatively different from what came before.

As context windows expanded, task horizons lengthened, and multi-step reasoning became more reliable, something meaningfully changed.

With the arrival of frontier models like Opus 4.6, we are crossing into something genuinely new: AI that operates less like a software tool and more like a human employee in digital form. It takes direction, handles ambiguity, works across the systems your organization already uses, and gets better with feedback. The product modality has changed.

Early Signals

At 8VC, we had an early and close view of this shift through our investment in Cognition, which is building the world's first AI software engineer.

Cognition demonstrated something more fundamental than AI writing code: a long-running agent could be given a goal, do meaningful work across tools and context, and come back with a finished work product. Software engineering was the first function to experience this because it was the most legible to AI systems – code is structured, outputs are testable, feedback loops are tight.

The insight generalizes beyond software. Any function that involves navigating systems, processing documents and making routine judgment calls across an organization is a candidate for this product form. 

Software engineering was first, but it won’t be last. That conviction led us to Tanuj Thapliyal and Mani Kotaru, co-founders of Kos.ai. 

The Founders

Today, we're proud to announce our investment in Kos.ai, the world's first AI accountant, purpose-built for the finance and accounting operations of America's critical infrastructure.

Tanuj previously built Spot AI and brings rare operator depth – ground-level clarity on how complex finance operations run and what fast-growing infrastructure CFOs actually need. Mani has leading-edge AI research experience and a precise understanding of where frontier models are genuinely capable today versus where architecture needs to be purpose-built. 

We first met Tanuj and Mani early in their ideation phase. Working through discovery together, it became clear the right approach was bigger: not a point solution, but the full digital accountant for industrial-stage businesses. They pivoted fast – the data was clear and they followed it. That instinct to update priors in real time and lean into uncomfortable findings rather than rationalize around them, is among the rarest qualities in early-stage founders. 

The Reindustrialization Opportunity

America is building again, yet the accounting function wasn't designed for what comes next.

Trillions of dollars are flowing into the physical backbone of the AI age: datacenters, power plants, factories. The businesses at the center of this moment, which include datacenter developers, power producers, manufacturers and defense primes, are experiencing some of the fastest growth in the economy. But they're also among the most capital-intensive. That combination creates a finance and accounting challenge with no good precedent.

The finance workflows powering America’s reindustrialization (draw requests, payment applications, PO reconciliation, freight and tariff tracking, utilities billing) are among the most complex in any business and largely manual. A single 100MW datacenter building may carry three to five full-time accountants living in spreadsheets, PDFs, and email. Datacenters alone are projected to scale from roughly 8GW of capacity today to 170GW by 2030. The math on linear headcount addition doesn't work.

Underneath all of it is a structural problem: America's accountants are retiring at exactly the moment demand for their work is exploding. Fast-growing industrial businesses are inheriting the finance complexity of capex-heavy incumbents without the decades of staffing infrastructure incumbents built to manage it. 

Kos.ai is starting with America’s AI capex buildout as the need is most acute and complex. This spans from America’s great new defense companies, to the lenders, datacenter developers, energy companies, equipment manufacturers, contractors, and more. The vision is broader though, as the team aims to build a digital accountant for any business that needs one. By starting with highly complex and urgent use cases, Kos.ai can master capabilities that would lend strongly to many other businesses.

A New Product Modality

As model capabilities advance, the most important product challenge is how to make the path from onboarding to value as fast and seamless as possible.

Kos.ai is pushing the frontier here. Kos has no UI at all. You interact with Kos the same way you interact with anyone else at work - email, chat, video calls. Kos even has subscriptions to the same software as you.

The onboarding analogy is intentional and precise. You don't implement Kos, you onboard it. Provision access, share the standard operating procedures, explain the job, provide business context. The same way you'd bring on a new remote employee. When something isn't working, you don't file a support ticket, but get on a Zoom to review the work and walk through feedback. Kos learns the correction and applies it going forward. 

The digital employee modality is reaching its natural conclusion — a product that doesn't ask organizations to change how they work, but shows up and works within it. In this category, a lot of the alpha will be won on adoption: how fast a large organization can onboard, recognize value, and integrate into more core workflows. The employee analogy isn't just a marketing frame, it’s the product strategy. Collapsing implementation timelines from quarters to days gets customers to value fast – and like any good employee, Kos gets better the longer it’s in the role. 

The distinction runs deeper than UX. General-purpose AI assistants hand broad system access to an underlying model that wasn’t designed to manage the security posture industrial finance requires. Kos is provisioned as an isolated tenant with its own managed identity, the same way a vetted remote hire would be onboarded into an organization's IT environment. Auditability, traceability, and strict access controls are native to the architecture, not bolted on. 

That architecture is also a competitive differentiator. The gap isn't access to models. FinOps workflows at scale demand levels of accuracy, reliability, and auditability that a general assistant architecture isn't built for. Kos.ai's purpose-built approach compounds with every deployment.

Building Kos.ai for this moment

Tanuj and Mani are reimagining what it means to build a company in the AI-native era, which extends beyond product to how they operate. 

On one side of the barbell: the vast majority of internal functions (GTM, recruiting, customer discovery, operations) run with dramatically fewer human resources than historically required. AI is doing real work across the business.

On the other side, they’ve leaned into deep, high-touch relationship building with every prospect and customer. Kos.ai sits inside customer processes, learns the business model, and becomes a trusted counterpart to the office of the CFO. In a world where the software layer commoditizes quickly, distribution built on genuine domain expertise and real relationships is the durable moat. 

We couldn't be more excited to partner with Tanuj, Mani, and the Kos.ai team. The digital employee is moving beyond software engineering — Kos.ai is defining what that looks like in finance. If you want to help build the world's first virtual finance employee, they're actively growing the team. Click here to learn more!