Trillions of dollars are flowing into datacenters, defense, energy, and manufacturing. The finance teams processing that capital can't keep up. 800-page invoices, thousand-page contracts, landing every few weeks, often on weekends. The world can't hire specialized accountants fast enough.
Kos is the world's first virtual finance employee for america’s industrial revival. We're processing invoices for some of the fastest-growing infrastructure and defense companies in America. Backed by $12M from 8VC, XYZ, and MVP with industry-leaders as shareholders (Keith Taylor [ex-CFO Equinix], Heather Paduck [CFO STACK], Matt DeNezza [ex-CFO Crusoe], Sachin Katti [Industrial Compute OpenAI]).
Media: The Information covering our launch
You'll own the agent loop, the model orchestration, and the eval infrastructure. Kos is processing billions of dollars of customer work. Your systems are what keep that right. You'll own how that system gets built and how it gets better. Your work lands in production at some of the largest and fastest-growing infrastructure and defense companies in America.
What you'll do:
Own the agent runtime: computer use, tool orchestration, and memory loop that lets Kos learn a workflow from customer demos
Design the ensemble routing layer that picks the right model per step and keeps Kos.ai provider-agnostic
Build the evaluation harness that catches regressions on recorded customer workflows before they reach production
Take frontier agent capabilities from paper to production, with equal attention to what the research says and what runs reliably every day
Partner with the applied controller team so practitioner-caught errors turn into prompt, model, and routing improvements
Set the hiring bar and culture for Kos.ai
Have strong fundamentals: Bachelors / Masters / PhD in CS/EE from top-tier universities, 2+ years of experience
Have shipped an agentic system, computer-use agent, or LLM-powered feature into production. Or you have an equivalent research track record.
Worked on modern LLM orchestration, tool use, eval design for open-ended tasks, and the tradeoffs between fine-tuning, prompting, and retrieval are second nature to you
Can take a problem from "interesting paper" to "runs reliably every day for paying customers" without being told how to scope it
Track agent and LLM research closely. Useful ideas turn into production experiments in your hands quickly.
Care as much about the eval harness as the model weights.
Medical, dental, vision for you and your dependents (little/no premiums, the absolute best we qualify for)
401(k)
Visa sponsorship
Tanuj and Mani met at Stanford as graduate researchers under Sachin Katti.
Tanuj spent his career in and around the operators Kos serves: IT, datacenters, industrials, and hardware. He previously founded Spot AI, a computer vision company which scaled to over 1,000 enterprise customers. Mani's an International Olympiad medalist, earned his Engineering PhD at Stanford, & published leading research at Microsoft Research. He then built product within Microsoft Azure's Strategic Missions and Technologies group.
We built Kos because the finance and operations teams behind America's re-industrialization are talented, overwhelmed, and can't hire their way out. Every CFO we spoke to described the same problem: 800-page pay apps, thousand-page contracts, work piling up on weekends, and a talent pool for inventory, utilities, capex, and construction accounting that's simply too small to meet the moment. The back office of the buildout is the bottleneck, and it's the one nobody's building for. They deserve a partner that can keep pace with their ambition.
Identify high-impact business workflows, redesign them for AI agents, connect systems, and deploy and operate production agents while tracking quality, efficiency, cost, and business outcomes.
Build production services, internal tools, and customer integrations for AI agents, developing reliable primitives and workflows for tasks such as meetings, browsing, search, email, and payments.
Build and ship customer-specific AI automations and agents, turning recurring creative requests into reusable systems while setting quality standards, documenting solutions, and partnering with Product and studio teams.
Deploy ProcessOS into enterprise production, owning customer workflows from initial process files through measurable improvement, while building product features, custom agents, and repeatable deployment practices.
Build and maintain AI agents, automations, and apps embedded with Product, Product Marketing, and Design, turning team needs into adopted tools that improve customer insight, decision-making, and delivery speed.
