o11 Wants the AI Agent to Do the Work, Not Just Talk About It

Almost every software company shipped an AI sidebar in the last two years. A little chat window bolted onto the corner of the app, ready to answer questions. o11, a startup in Y Combinator's Winter 2026 batch, looked at that pattern and asked a blunter question: what if the agent just finished the work? Not a summary of the spreadsheet, but the built model. Not advice about the deck, but the deck.

That is the whole idea behind o11, and the founders describe it in one line - "the AI agent inside of every enterprise app." The company builds AI agents that live inside the tools finance teams already open every day: Microsoft PowerPoint, Excel, and Word, plus Google Slides, Sheets, and Docs. Instead of exporting your data to a new platform, you type a request where you already are, and the agent produces the deliverable.

The distinction sounds small until you sit next to someone doing the job. A managing director does not want a chatbot's opinion on a comparable-company analysis; a first-year analyst does not want a suggestion about how to structure a slide. They want the tab filled in, the footnotes tied to the filing, the deck formatted the way the firm has always formatted its decks. o11's argument is that the last mile of enterprise AI is not intelligence at all - it is delivery. The model can be brilliant, but if it hands the work back to a human to assemble, most of the time has already been spent.

01 / The ProblemLegacy software isn't going anywhere

o11's founders, Aryah Oztanir and Ajay Misra, met the problem up close. Surrounded by finance students heading into internships, they watched capable people spend nights doing manual grunt work inside decades-old software - reformatting slides, wiring up spreadsheets, retyping numbers from filings. The tools were entrenched, and the AI features grafted onto them were, in the founders' telling, mostly decoration.

The statistic o11 keeps returning to is that roughly 62% of companies still run on legacy software. That is not a market waiting to be replaced; it is a market too large and too load-bearing to rip out. So the incumbents - Microsoft with Copilot, Google with Gemini - added AI assistants to the side of those tools. Useful for a quick answer, less useful when the actual job is to produce a 40-page committee memo grounded in current filings.

There is a reason the incumbents stopped at the sidebar. Doing the work is riskier than talking about it. A wrong sentence in a chat window is a shrug; a wrong number in a financial model is a mispriced deal. Finance is one of the least forgiving places to ship an AI product, which is exactly why most general-purpose assistants hedge - they surface information and leave the judgment, and the liability, to the person at the keyboard. o11 is choosing the harder half of the problem on purpose, betting that the firms who feel the grunt-work pain most acutely will also be the ones willing to pay for a tool that actually removes it.

02 / What It DoesAn agent that ships the work

In practice, o11's agents take a natural-language prompt and return finished finance work. The launch materials and a case study with data provider Valyu describe outputs like three-statement models, investment committee memos, confidential information memoranda (CIMs), buyer lists, and presentation decks - the kind of documents that normally consume an analyst's evening. The difference from a chatbot is that these are not descriptions of the work. They are the work, produced inside Excel, Word, or PowerPoint.

The customers are specific. o11 says its agents are already used by investment banks, private equity firms, hedge funds, and asset managers, with workflows tuned to each. The pitch is not "AI for everyone" - it is AI for the person who has to produce a diligence pack by morning. That focus is a strategy as much as a market. Finance deliverables are highly templated, repeated thousands of times a year, and expensive when done by hand at analyst salaries. It is a category where "the agent did it in the format we always use" is worth real money, and where the buyer can measure the value in hours saved.

03 / The Hard PartReal numbers or nothing

Finance work is unforgiving about data. A model has to be grounded in 10-Ks, a memo has to cite current filings, a market summary has to reflect what happened this week. Building that data infrastructure in-house is its own multi-month engineering project - licensing feeds, ingestion pipelines, indexing filings, normalizing formats. o11 decided not to do that.

Instead it wired in Valyu's Search API as its retrieval layer, pulling SEC filings, company research, financial data, and news directly into the agent stack. According to the case study, o11 got its data layer running in days rather than months, with every output traceable back to a primary source.

Data layer: build vs. buy (time to running)

Build in-house

months

o11 via Valyu

days

04 / The FoundersA habit of shipping early

Oztanir and Misra both left the University of North Carolina's computer science program for YC. What stands out in their backgrounds is less the pedigree than the pattern of shipping real products young. Oztanir built a Discord bot with more than 10,000 users at 14, and an Apollo.io-style tool that saw 20,000 uses at 19. Misra led AI at a startup he says reached $40M in ARR and 10 million users while he was 19, and published cancer-detection research at the Mayo Clinic.

That history shows up in how o11 operates. The choice to buy the data layer instead of building it is the same instinct - do the smallest thing that ships the product, then move. For a company selling to investment banks with a team of roughly three people, that instinct is close to a survival trait. A small team cannot win by out-engineering a bank's internal technology group on data infrastructure, so it doesn't try. It spends its scarce hours on the one thing it can be best in the world at: the agent that turns a prompt into a finished document inside the app.

The business model follows the customer. o11 sells software to financial firms - B2B, subscription in shape - and rides on top of the Microsoft and Google tools those firms have already standardized on. There is no rip-and-replace pitch, no migration project, no year-long procurement fight to swap out Excel. The agent shows up inside the software the firm already licenses. That lowers the barrier to a first "yes," which for an early company selling into conservative institutions may matter as much as the product itself.

05 / The MarketWhere o11 sits

o11 lands in a crowded but oddly divided market. On one side sit the platform giants, Copilot and Gemini, whose AI reaches everyone but rarely finishes a specialized deliverable. On the other side sit finance-specific AI tools and general agent platforms, powerful but usually asking users to work somewhere new. o11's bet is that the winning position is the least glamorous one: inside the apps people already use, doing the parts of the job they least want to do.

Its stated direction is to keep going past productivity suites - to automate work inside any of the thousands of legacy enterprise applications businesses depend on. Whether that expansion holds is the open question for a company this young. But the initial wedge is clear, and it is narrow on purpose: the finance analyst's late night, handed to an agent that lives in the same file.

The expertise that makes that possible is less about any single model and more about the plumbing around it. To produce a usable financial deliverable, an agent has to understand a firm's templates, respect the format of the destination app, reach for the right filing, and keep every figure tied to a source a compliance team can check. o11's decision to lean on Valyu for retrieval and to focus its own effort on the agent-inside-the-app layer is a statement about where it thinks the defensible work lives. Data can be bought. The last mile - the thing that turns a prompt into a document a managing director will actually send - is the part o11 is trying to own.

None of this guarantees the outcome. o11 is early, small, and selling into an industry that moves slowly and audits everything. The incumbents it is stepping around have distribution o11 can only dream of, and a general agent platform could always decide to specialize. But the company has picked a real problem, a customer that feels it every day, and a position - inside the tools, not beside them - that the giants have so far declined to take. For a three-person team out of YC W26, that is a coherent place to start.

The name is a small tell about the ambition. "o11" reads like shorthand - the kind of compressed, engineer's label a team picks when it plans to be typed a lot. The company would rather be the thing inside your tools than the tool itself. If the sidebar era was about AI you notice, o11 is chasing the version you stop noticing because the work is simply done.