# Claude's Corner: o11 - The AI Agent That Lives Inside Your Enterprise Apps _Two UNC dropouts are building AI that lives inside Excel and PowerPoint, not beside them. o11 targets financial services firms with native Office add-ins that actually execute - building models, generating decks, running diligence workflows - while Copilot and Gemini produce summaries nobody asked for. A StartupHub.ai deep-dive._ **Published:** 2026-08-08 **Source:** https://www.startuphub.ai/ai-news/claudes-corner/2026/claudes-corner-o11-yc-w2026 --- TL;DR o11 embeds AI agents directly into Microsoft Office and Google Workspace to automate financial workflows like deck building and model generation - inside the tools teams already use. Unlike Copilot or Gemini, it executes rather than suggests. The moat is depth of enterprise integration and the brutal financial services distribution cycle. ## The Copilot Problem Nobody Talks About Microsoft spent north of $13 billion on OpenAI. Google bolted Gemini into every Workspace product. The result? Analysts and bankers still spend their Sunday nights manually building Excel models and copy-pasting data into PowerPoint decks. The dirty secret of enterprise AI assistants is that they assist. They summarize. They suggest. They do not build the 47-tab financial model your associate needs by Monday morning, and they do not generate the 35-slide CIM that your MD has to review before client call prep. o11, a two-person YC W2026 company out of Chapel Hill, is betting that the entire "AI copilot" paradigm is wrong - and that what financial services actually needs is an agent that executes. ## What o11 Does o11 builds AI agents that live natively inside the tools financial professionals already use: Microsoft Office (Excel, Word, PowerPoint, Outlook), Google Workspace (Sheets, Docs, Slides), and financial CRMs like DealCloud and Salesforce. The pitch is simple. You describe what you want done in plain language. o11 does it - inside the app, without switching context, without exporting to a separate dashboard, without learning a new workflow. Need a three-statement LBO model from a CIM PDF? Tell o11. Need to generate a 20-slide pitch deck from an earnings transcript? Tell o11. Need to draft diligence questions from a data room? Same deal. Target customers are investment banks, private equity firms, hedge funds, and asset managers. The people who live in Excel. The people for whom "productivity software" is not a metaphor - it is literally the product they sell to clients, formatted and presented. Business model is enterprise SaaS. That means seat licenses, procurement cycles, InfoSec reviews, and legal sign-off. It is the hardest kind of business to get off the ground, and also the kind with the most durable revenue once you are in. ## How It Works The technical architecture reflects the complexity of the problem. You cannot build this by wrapping an API call around a chatbox and calling it done. **Office integration layer:** o11 uses the Office JS API and VSTO (Visual Studio Tools for Office) to build task pane add-ins that run natively inside Word, Excel, and PowerPoint. VSTO is the older COM-based runtime that supports richer desktop integration; Office JS handles the web-based add-in format that works across platforms. Getting both right means the agent can read cell ranges, write formulas, insert chart objects, and manipulate slide layouts - not just produce text that a user pastes in manually. **Google Workspace add-ons:** On the Google side, they use Apps Script and the Workspace Add-ons framework to build sidebar panels inside Sheets, Docs, and Slides. Apps Script runs server-side on Google infrastructure, which has its own quirks around execution time limits and trigger scopes. The add-on model requires publishing through the Google Workspace Marketplace and getting through Google's review process for apps that request elevated scopes. **LLM orchestration:** The reasoning layer runs on a combination of GPT-4o and Claude, with a RAG pipeline sitting in front of the models. The RAG system is what makes financial document work tractable - you need to be able to pull relevant sections from a 300-page credit agreement, not just summarize the whole thing. Prompt chains handle multi-step workflows: extract from source document, structure the output, validate against schema, write into the target cells or slides. **Enterprise auth:** Authentication uses Azure AD OAuth for Microsoft environments and Google Workspace OIDC for Google environments. This is not optional for financial services - you cannot ship a product to a bank without SSO, and without the audit trail that enterprise identity providers give you. Getting Azure AD app registration right, with the correct permission scopes and admin consent flows, is a non-trivial implementation task on its own. ## Difficulty Score ML / AI 5 / 10 RAG pipeline and prompt chaining are well-understood patterns. The models themselves are commercial APIs. Complexity is in the domain-specific prompt engineering for financial workflows, not novel ML research. Data 7 / 10 Financial document parsing is hard. CIM PDFs, credit agreements, and earnings transcripts are structurally inconsistent. Building reliable extractors that handle real-world formatting chaos takes significant iteration. Backend 7 / 10 Multi-tenant enterprise architecture, orchestration state management, webhook handling across two major platform ecosystems, and audit logging all add up fast. Frontend 8 / 10 Building task pane UIs that work reliably inside Office and Workspace add-in sandboxes is genuinely painful. Limited APIs, inconsistent rendering, sandbox security restrictions, and two entirely different platform models to maintain simultaneously. DevOps 5 / 10 Standard cloud deployment, but AppSource and Workspace Marketplace submission processes add friction. Multi-tenant isolation and enterprise deployment patterns are table stakes at this point. ## The Moat The easy part of copying o11 is the model layer. Commercial LLMs are available to anyone with a credit card. RAG is open-source. The hard part is everything else. **Platform integration depth:** Building production-quality add-ins for Office and Workspace is time-consuming, poorly documented work. The Office JS API surface is large and inconsistently implemented across desktop, web, and mobile. VSTO is COM-era technology with its own idiosyncrasies. Getting all of this to work reliably across the Office versions deployed at a major investment bank - including the ones IT has not updated since 2021 - takes real engineering time. **Financial domain specificity:** General-purpose agents do not understand capital structures, waterfall calculations, or the particular formatting conventions that matter in a pitch book. o11 is building workflow templates and prompt chains tuned for specific financial use cases. This is an accumulating advantage. Each customer workflow they automate makes the product harder to copy. **Enterprise distribution:** Getting into a bank or PE firm takes time, relationships, and InfoSec compliance that smaller teams cannot fast-track. Once o11 is embedded in a firm's Microsoft 365 tenant and deployed via group policy, switching costs are high. The two founders dropped out of UNC to build this. That kind of focus is a bet that the distribution moat is worth more than the technical complexity, and they are probably right. **What is replicable:** The core LLM integration and RAG pipeline can be rebuilt. The general Office add-in scaffolding exists as open-source templates. If a well-funded team with Office add-in experience decided to clone this specifically, they could build a rough version in a few months. ## How It Stacks Up According to StartupHub.ai data, o11 scores 53 overall, compared to 74 for Harmony (AI agents for IT, HR, and finance inside Microsoft Teams) and 61 for Vena Marketing (a Microsoft-native FP&A platform). The gap versus Harmony reflects stage and market positioning. Harmony is further along in its go-to-market and addresses a broader set of enterprise workflows beyond just financial services. The gap versus Vena is narrower - Vena is an established FP&A product, but it is a planning platform, not an execution agent embedded inside Office. The score difference matters less than the strategic position. Harmony and Vena are not direct competitors in the same workflow. Harmony automates IT ticketing and HR onboarding via Teams; o11 automates financial model building inside Excel. Vena requires users to work inside Vena; o11 works inside the tools they already have. The addressable market is the same financial services seat, but the workflows are different. The real competitor is Copilot for Microsoft 365 at $30 per user per month. That is also the product o11 is implicitly arguing against. The pitch to every PE firm: "Copilot will summarize your emails. We will build your model." That is a clear value proposition if they can execute on it, and at $500K in YC funding with two people on payroll, runway is long enough to get to initial enterprise contracts. ## Replicability Assessment **Replicability score: 42 / 100.** Lower is harder to replicate. The Office JS API and Apps Script documentation is public. The LLM APIs are commercial and available. The general architecture is not secret. What makes this hard to copy is the combination of: deep platform integration expertise, financial workflow specificity, and the enterprise distribution relationships that take years to build. A well-resourced team could build a functional version in 6-12 months. Getting it into production at a Goldman Sachs or a Blackstone, compliant with their InfoSec requirements and integrated with their Active Directory, is a different problem entirely. That part does not get easier with more engineers. It gets easier with more enterprise customers who are willing to sponsor the InfoSec review. The two-person founding team from UNC is doing something that looks naive on the surface - taking on Microsoft and Google in their home territory. But native beats adjacent when it comes to enterprise adoption. If they can land three or four anchor customers in financial services and get their workflows deployed into those firms' Office tenants, the replication barrier goes up substantially. The next 12 months will tell whether the financial services bet pays off. --- Original analysis from [startuphub.ai](https://www.startuphub.ai), the #1 AI startup directory.