Yijia Liang
Co-founder at Compound AI
About Yijia Liang
Yijia Liang is a co-founder of Compound AI, a startup building AI agents for institutional finance. Before Compound, Liang spent about three years as a vice president specializing in late-stage private software investments, and left to build tools that replace the analyst work of building models, slides and memos.
In an AI Engineer talk, Liang explains why Compound rebuilt Office from scratch so agents can run deal-team work in parallel and in the background.
Career Highlights
- Co-founder of Compound AI, which builds agents for private equity and hedge fund deal teams
- Former vice president focused on late-stage private software investments
- Speaker at AI Engineer (2026) on why Compound rebuilt Microsoft Office for AI agents
Notable Positions
On Multiplayer Agents
Finance deployments of ChatGPT and Claude are single player, Liang argues, while deal work involves a whole team. Compound gives every agent an email inbox so the entire deal team can work with one agent on one thread, with authentication for NDA-bound threads and deliberate pacing so replies feel human.
On Owning the Document Engine
Plugins, computer use on virtual machines and libraries like openpyxl couldn't scale. Compound ran coding agents in loops against Office as the oracle, starting from the open-source IronCalc engine, until it had pixel-faithful Excel, PowerPoint and Word engines. Owning the renderer gives agents a render, check and fix loop.
On Swarms Over Vector Databases
For large data rooms, a lead agent slices the directory and swarms of agents read files with grep-style tools, returning findings cited to the source. See Multi-Agent Systems.
Key Quotes
- "For institutional finance, most workflows run through Outlook." (on email-native agents)
- "We ran agents in many loops, spend 1.5 trillion input output tokens and over 8 million lines of code to rewrite Office from scratch." (on rebuilding Office)
Related Reading
- Multi-Agent Systems - Swarms of agents over large file sets
- Agents as Employees - Agents joining teams as teammates