# Towards Autonomous Analysts in Finance — Yijia Liang — session 2026-10-14T14:40:00.000Z → 2026-10-14T14:58:00.000Z _35 transcript lines · 53 slides · source: full recording_ ## Transcript **[00:00:14 · 0]** Alright. How's everyone doing? My name's EJai. I'm the cofounder of Quant dot ai. We sell analysts for Five Side Mining, and specifically, we **[00:00:24 · 1]** specialize in long running analysts. Think Cloud agents that Devan or Factory might do for coding agents, but tailoring that for buy side institutions. I was previously an analyst in a previous life. I was a vice president at CodeTwo for about three years, specializing in late stage private software investments. Spent way too much time in Excel, PowerPoint, and late nights doing that type of work, and so decided to split and start building tools to replace my old job. **[00:00:50 · 1]** Quick show of hands. How many people in here have used User Instinct in the last day? Alright. I expected more hands, but, maybe people are too shy to raise them. And how many people know an analyst in finance that fully delegates their work to AI? **[00:01:06 · 1]** Like, fully hands off, running autonomously? Alright. One. But a lot fewer. I think if you think about the question, why did personal agents take off within the last three months? **[00:01:18 · 1]** Internally, we came up with three main answers. The first, reliable background agents. You can delegate a task, book a restaurant, book a flight, and it'll do it reliably. Now, computer use agents have gotten sophisticated. The models themselves have gotten better. **[00:01:35 · 1]** The second is productivity. It'll do things without you asking. It's part of that magical experience when instinct wakes you up saying, it's your wife's anniversary. Go buy some flowers. Or better yet, I already bought the flowers for you. **[00:01:46 · 1]** And finally, integrations. It's connected to the systems of context that matter most for you: email, drive, calendar. But all the factors that make personal agents work so well doesn't really translate to institutional finance today. If we think about background tasks, something as simple as booking a restaurant reservation, The analog equivalent in Biscotti Finance might be how do you go from a data room to a fully prepared set of IC models, slides, and memos, taking a process that would normally take four weeks, multiple rounds of iteration and review, and compressing that to a few days. If we think about productivity, your managing director doesn't really want an agent rogue running around doing a bunch of things and burning tokens without them providing and soliciting input. **[00:02:37 · 1]** And then finally, if you think about the context that matters, it's not just email, it's not just calendar. It could be a data room with 15,000 files. It could be SharePoint box with petabytes of unstructured data the firm has collected in its forty year history. So how do you deal with these problems? So the purpose of today's talk is to discuss a bit more about how we at Compound have approached the problem of solving long running agents in finance. **[00:03:02 · 1]** On one hand, it's creating a real teammate that collaborates with you and the deal team. So that's multiplayer embedding that into the app. On the other hand, it's solving for how can you run long tasks in parallel in the background, reliably execute them. And so for that, we actually rebuilt Office from scratch. And then finally, it's how do you manage the context of everything, everything we just discussed, petabytes of data in SharePoint, in Box. **[00:03:28 · 1]** And we'll talk a little bit about how we've designed a solution for that. So the first question we have to ask, I think, is why does multiplayer actually matter for finance? If you look at the anatomy on a deal team, right, the type of tasks that matter for long running type of workflows, for example, going from a data room into all the sorts of materials that make its way in front of an investment committee, it doesn't work for the single player siloed workflows. Today, for people who have deployed Cloud or ChatGPT in their institutions, it's all single player, people working in a plug in, people working in the web app, producing outputs. But invariably, those people need to collaborate with their team. **[00:04:08 · 1]** And what does that look like? It means downloading the file, uploading it into email, uploading it to Teams, asking for feedback. I think the right way to approach this, and this is the way we've thought about at Compound, is how do you get every person who matters on a deal onto the same thread? How can they all talk to the same agent so that one agent can share the context of everyone who matters for any particular work stream? When we thought about how do we actually bring multiplayer into a familiar context, we looked again towards personal agents. **[00:04:38 · 1]** Why do they work so well? Well, you can easily text one in iMessage, or you can text one in WhatsApp. So for agents at Compound, every single one you create comes with a built in inbox because ultimately, at the end of the day, for institutional finance, most workflows run through Outlook. You'll have chains of emails, 20 or more running. You want an agent to be part of every single one of those conversations. **[00:05:01 · 1]** Designing email for agents is actually nontrivial. It's not just as simple as hooking it up with an email inbox. It requires, a, authenticating, making sure every one part of that thread has read and write access to that agent. Ultimately, at these financial firms, privacy is incredibly important. You're always under NDA no matter what deal you look at, and so you want to make sure everyone on that thread can actually talk to the agent. **[00:05:24 · 1]** But then beyond authentication, you also have to think about how do you design a real human like experience because after all, we're trying to create a real teammate. If multiple people email in the same thread in quick succession, you don't want multiple acknowledgments spamming that group. You actually want to bake in some more latency to make it feel more human. And so there are other design decisions when thinking about how to create real email agents for enterprise finance. But multiplayer aside, we realized even after you solicit the input of everyone who matters, if you wanted to scale up the number of tasks and run them all in parallel, existing off the shelf solutions simply don't cut it. **[00:06:01 · 1]** We evaluated a lot, the desktop plugins of Excel, PowerPoint, and Word. We evaluated computer use using virtual machines to try to scale up the number of agents that could work in Excel, PowerPoint, and Word natively. And we also evaluated a ton of off the shelf engines, whether those are open source packages like OpenPyXL or commercial engines that are available for purchase. None of them really cut it for what we were trying to look at. And with coding agents dramatically improving and bringing the cost of software down, we decided to just build our own office from scratch. **[00:06:33 · 1]** And like a great AI engineer would, we ran agents in many loops, spent 1,500,000,000,000 input output tokens and over 8,000,000 lines of code to rewrite Office from scratch. And I'll walk through a little bit of how we approached it, but it's as simple as pointing agents at an objective function, which is recreate Office in full fidelity and then letting it run. So take Excel as an example. We started with an open source solution called IronCalc. We decided, okay, how can you recreate this perfectly for Excel, use Office as an oracle, and just run loops, benchmarking it on real files from both real world corpus that we grabbed online and also ones that, internally, we created. **[00:07:14 · 1]** And 800,000 lines of Rust later, we had a performant... Much more performant compared to Excel, one for one replica of the product. With this, we're actually able to do that for not only Excel but also PowerPoint and Word. Right? So we ran loops across every single one of these products, and we have faithful recreations. **[00:07:33 · 1]** You can try it for yourself today at getcompound.ai. And it's the same method, loop agents against Office as a judge, using Rust renders to make sure that the rendering of these products was pixel perfect. So what you see in Compound is what you see when you download to Office. We needed one for one parity because, at the end of the day, these firms only trust what they can download and what they can send. Now let's talk about performance. **[00:08:01 · 1]** Why is a headless engine so much better compared to using a VM or using desktop Excel? Well, for one, desktop Excel or PowerPoint or Word is bound to that CPU. With headless engines, you can do tasks in parallel much, much faster. You can build out a complex model in a much shorter amount of time. You can recalculate 20 megabyte excels four x faster. **[00:08:23 · 1]** So every single piece of work an analyst would expect an agent to do because, again, their expectation is quality of a human at much lower speed and much lower cost. You have to have a headless engine to do this. If you tried to scale this up with Excel and PowerPoint, I think you'd be going down the wrong path. But it's not just about performance. Owning the Editor actually allows you to create a much richer product experience that allows you to work and collaborate with agents asynchronously much better than you can with other products. **[00:08:54 · 1]** So this entire slide deck you're seeing today, I built in Compound just by dropping in comments. I had it create a first draft, and then as you can see here on the UI, I dropped in comments for every single piece of formatting For my consulting days, for my code two days, all the learnings I had on what makes a great slide, I imparted with the agent. This can also work for multiplayer. You can have multiple people part of the same team using the same multiplayer infrastructure we set up, dropping in comments. So for a team of five, you can rip through a 100 page deck in an afternoon rather than sitting in a room revising it for multiple weeks. **[00:09:29 · 1]** Finally, with respect to building your own engine, it also allows us to run iterative checking loops much more as well. So if you have a task that, for example, is writing a complex model, it might take a few hours. But how do you... How can we make that faster? How can we make it more accurate? **[00:09:45 · 1]** If you're using, for example, an OpenPyXL code library, there's no render you can use to check formatting. You're just guessing and hoping that the model writes the right thing, and maybe you do some parsing to see a cell as a particular color or has a particular font. If you try to use Microsoft's own interface, you're going to be limited by what Microsoft provides you, which is admittedly very little. So with an own engine, you actually own the right render, check, and fix loop. So you can rapidly iterate on the output for the user. **[00:10:16 · 1]** So the last thing I sort of want to discuss today beyond multiplayer and building our own office is how to deal with the context that these firms have. For firms on the buy side, they deal with situations where it crosses both breadth and depth of data. You may have situations where you have a data room over 15,000 files, and then files in that data room might be well over 20 megabytes. People in finance complain all the time. Hey, I have an Excel model that's 30 megabytes. **[00:10:43 · 1]** It takes five minutes to recalculate because we have 2,000,000 formulas and five sensitivity tables that need to recalculate every single time. So how do you deal with this? Well, for breadth, I think, again, we take a look at a page out of the coding book and apply agent swarms to the large number of files. No longer do you have to index everything in a vector DB unless it's for a very specific use case. It's actually much preferable to use coding agents to read files, either through grep or through some other methodology, and then have a lead agent slice the contents of the file directory and then have swarms of agents come back to you with insights. **[00:11:19 · 1]** And then you iterate that in a loop until you're satisfied with the output required. Furthermore, through this process, you're actually able to ground every single finding in citations in the source document so that you know exactly what you show on the screen is linked to some number that the analyst can trust. If we think about the depth of data, right, take that 30 megabyte workbook, how can you actually process this more accurately? In today's world, if you were to just use ChatGPT or Claude off the shelf to process that 30 megabyte model, it would write code. It would use open source packages like OpenPyXL. **[00:11:55 · 1]** It would be expensive and slow because it's printing out every single piece of context about that model in code for the model to read. But because we've developed our own engine for Excel, we're actually not only more accurate but a lot faster and cheaper because the engine processes it and then lays it out in a very easy to read and easy to process format for the model. This means we can expose a much richer set of APIs and a much richer set of tools for the agent to work over, and that allows us to process these models in a much faster manner. So if we sort of take a step back and think about, okay, what is all of this context built for? At the end of the day, what we want to do is prevent people from spending time finding files, from reminding the model this is how to do things. **[00:12:42 · 1]** Instead, they should be focused on imparting their judgment on the deal process. They should be focused on tailoring training agents and deploying them on every part of the deal flow. And so it goes from the bottom level, connected sources, off the shelf you know, connectors to SharePoint or Outlook, to the cited findings grounded in the sources that matter. And then finally, embedding in the preferences, company knowledge that matters. How do you do a model? **[00:13:07 · 1]** How should it look? Which differs by every single firm. So that for humans, they spend most of their time imparting judgment. They spend most of their time thinking about is this actually a good deal, not did all the numbers tie? Does the model look right? **[00:13:20 · 1]** Will my VP yell at me because it looks a different way? So... Oh, cursor needs to update. If we take a step back and think about the type of tasks that currently are on autopilot versus what are on copilot, I would say there's already several finance tasks. You can do pretty well with off the shelf tools, spreading comps. **[00:13:45 · 1]** As the Chat GPT head of finance mentioned earlier, they have now their own products for this. But I think there are a lot of longer running tasks that are not capable with existing solutions. How do you draft a 50 page IC presentation off the cuff? How do you get agents to build out a model exactly the way you want it for a private equity professional or a hedge fund professional. I think those are tasks that, with off the shelf tools, still exist as co... **[00:14:11 · 1]** As sort of within the Copilot realm. People want to be embedded deeply in that process without letting go. But with some of the tools we've built today, I think we're able to expand the aperture for which Autopilot can take over in finance firms. So rather than just using tools embedded in Excel as plug ins, hand holding along the way, you can start delegating. You can start becoming more ambitious about what AI is capable of. **[00:14:37 · 1]** If I think about where we are today in the world of finance, it very much reminds me of when Opus 4.5 launched last December where people stopped using code as a simple copilot to generate the next file or the next line of code, but they started delegating far more ambitious tasks. They started realizing, maybe I can just truly vibe it. If I have sufficient checks, if I have sufficient tests, maybe I can just trust all the code it writes. I think with some of the tooling we've built today along with what we're planning in the future, these are the type of problems you have to solve before you get to that point for finance. How can analysts... **[00:15:15 · 1]** How can the entire deal team become more ambitious with AI? How can they just drop in a data room and decide, okay, I'm gonna trust every single one of the outputs? And then let's do that tenfold so we can see more deals, so we can expand the aperture of what matters. So to conclude, enterprise agents are certainly coming to finance. I strongly believe we're at an inflection point in the adoption curve and that people are slowly becoming more ambitious and they're moving away from copilots to autopilots. **[00:15:42 · 1]** And so this was sort of how we've approached it here at Compound. If you work at a firm and you're interested in deploying our agents to drive more of your deal work, please reach out. Or if you're an engineer and you're curious about talking about our approach and sharing ideas, please either reach out online or I'll be around if you want to find me. Thanks so much. ## Slides ### 00:00:00–00:00:20 # WE'LL BE RIGHT BACK ## AI Engineer NEW YORK View Full Schedule at [ai.engineer/nyc/2026/schedule](ai.engineer/nyc/2026/schedule) [Stylized silhouette of the New York City skyline at night, featuring prominent buildings like One World Trade Center, Empire State Building, and the Statue of Liberty.] ### 00:00:20–00:00:40 Compound # Towards Autonomous Analysts in Finance Yijia Liang, Co-Founder AI Engineer conference, New York, October 14, 2026 ### 00:00:40–00:01:00 # Towards Autonomous Analysts in Finance Compound Yijia Liang, Co-Founder AI Engineer conference, New York, October 14, 2026 ### 00:01:00–00:01:20 # How many people have used Muse or Instinct? ## Towards Autonomous Analysts in Finance Yijia Liang / Co-Founder Compound [Blue M-shaped logo for Muse, followed by the ### 00:01:20–00:02:00 # Why did personal agents take off in the last 3 months? ## Three pillars of what made personal agents so successful 1. **Reliable background agents** Reliably finish ### 00:02:00–00:02:20 # Most people in finance are not "handing off tasks" ## All three factors that make personal agents work, fall short inside an investment firm - **Background tasks** - Personal agent ### 00:02:20–00:02:40 # Most people in finance are not "handing off tasks" All three factors that make personal agents work, fall short inside an investment firm - **Background tasks** - Personal agent: ### 00:02:40–00:03:00 # Most people in finance are not "handing off tasks" All three factors that make personal agents work, fall short inside an investment firm - **Background tasks** - Personal ### 00:03:00–00:03:20 # The bar for firms is a real teammate ## How Compound is approaching long running agents in finance ### Deal-team Analyst - Collaborates the way a deal team does - Executes many tasks in parallel and in the background - Manages the context of everything ### What we built - Part 1 **Multiplayer** - Part 2 **Compound Office** - Part 3 **Massive, parallel context** ### Towards Autonomous Analysts in Finance Yijia Liang / Co-Founder Compound [Diagram showing three points describing a "Deal-team Analyst" connected by arrows to three corresponding parts of "What we built"] ### 00:03:20–00:03:40 # The bar for firms is a real teammate ## How Compound is approaching long running agents in finance ### Deal-team Analyst - Collaborates the way a deal team does - Executes many tasks ### 00:03:40–00:04:00 # Agents that listen to everyone on the deal team - Single player AI works for siloed workflows but doesn't scale for long-running tasks (i.e. a deal) ## ### 00:04:00–00:04:20 # Agents that listen to everyone on the deal team - Single player AI works for siloed workflows but doesn't scale for long-running tasks (i.e. a deal) ## Single-player: one person's chat - Maya: Keep leverage where it is - Peter's 'hold the exit at 5.0x' sits in another chat - Agent: Maya: leverage is unchanged ## Multiplayer: on the thread, sees everyone's - **Frankie** 9:12 AM: The Senteve downside case is built. FY27E growth runs at 38 ### 00:04:20–00:04:40 # Agents that listen to everyone on the deal team - Single player AI works for siloed workflows but doesn't scale for long-running tasks (i.e. a deal) ### 00:04:40–00:05:00 # Building multiplayer in familiar environments - All agents created in Compound come with a built-in inbox ### Personal: iMessage or WhatsApp > Fresh slate it is. What's ### 00:05:00–00:05:20 # Building multiplayer in familiar environments - All agents created in Compound come with a built-in inbox ### Personal: iMessage or WhatsApp "Fresh slate it is. What's on ### 00:05:20–00:05:40 ### Opening > Multiplayer > Our own Office > Massive context > Map and close # Identity, membership and threading keep it safe ## How it works under the hood: mimicking behavior of a human ### 00:05:40–00:06:00 # Identity, membership and threading keep it safe ## How it works under the hood: mimicking behavior of a human - **cc'd on the email** - **Lands in the ### 00:06:00–00:06:20 # Nothing off the shelf could run our cloud agents ## Why we ended up building our own engine - **Goal:** Cloud agents: in the background, many at once - **Options considered ### 00:06:20–00:06:40 # Nothing off the shelf could run our cloud agents ### Why we ended up building our own engine - **Goal:** Cloud agents: in the background, many at once - **Rejected options ### 00:06:40–00:07:00 # Building our own Office took 1.5 trillion tokens Rough order of magnitude for all of Compound Office, built with agents - **1.5 trillion** input and output tokens - ### 00:07:00–00:07:20 # We rewrote Excel in four months, entirely with AI We built the tool for agents by using agents, using Excel as a judge - Starting point - IronCalc: open source, missing features, slow - 4 months - of agentic loops - 800,000 lines of Rust - our spreadsheet engine today Source: Peter's LinkedIn article, 3 Aug 2026 ### Towards Autonomous Analysts in Finance Yijia Liang / Co-Founder Compound [Diagram showing a development process: a vertical timeline from "Starting point: IronCalc" to "800,000 lines of Rust" over "4 months of agentic loops". To the right, a triangular loop diagram with "Agentic loops" (with an Excel icon) in the center, connected to "Match Excel", "Real files", and "Speed".] ### 00:07:20–00:07:40 ## We rewrote Excel in four months, entirely with AI - We built the tool for agents by using agents, using Excel as a judge - **Starting point** - IronCalc ### 00:07:40–00:08:00 # Excel was the first engine, not the only one → The same method for PowerPoint and Word, with Office itself as the judge ### Excel The first engine Rebuilt against Excel's own calculated values ### PowerPoint 0.975 median match to PowerPoint's own render, across 4,402 slides in 169 decks ### Word The same method A typed document model, drawn by the same renderer - One method: loop agents against Office as judge - One Rust renderer: same pixels in browser and server ## Towards Autonomous Analysts in Finance Yijia Liang / Co-Founder ◊ Compound [Screenshot of an Excel spreadsheet with data and a bar chart] [Screenshot of a PowerPoint slide showing financial metrics and text] [Screenshot of a Word document with a project summary] ### 00:08:00–00:08:20 # A headless engine runs many tasks at once ## Impact 1: performance on many tasks, in parallel, in the background ### Operations a second, 20 files open at once - Compound engine: **225,000** - Commercial engine: **19,000** ### Performance Comparisons - **Model build** - Timed out at 300 s - **0.87 s** - **Each write, 24 MB model** - Every write timed out - **About 1 s** - **61 workbooks, 8.5M formulas** - LibreOffice: 183.0 s - **43.3 s** (4.2x faster) Source: internal benchmarks dated June 2026 (commercial-engine comparisons) [Horizontal bar chart comparing "Compound engine" (225,000) and "Commercial engine" (19,000) for operations per second with 20 files open at once] [Three comparison boxes showing performance improvements: Model build (0.87s vs timed out), Each write 24MB model (About 1s vs timed out), and 61 workbooks 8.5M formulas (43.3s vs 183.0s in LibreOffice, a 4.2x speedup)] ### 00:08:20–00:08:40 # A headless engine runs many tasks at once - Impact 1: performance on many tasks, in parallel, in the background ## Operations a second, 20 files open at once ### 00:08:40–00:09:00 # A headless engine runs many tasks at once ## Impact 1: performance on many tasks, in parallel, in the background ### Operations a second, 20 files open at ### 00:09:00–00:09:20 # Owning the editors makes agent work reviewable - Impact 2: Custom UI allows us to create better experiences aligned with analyst workflows 1. Comments get picked up 2. Team ### 00:09:20–00:09:40 # Owning the editors makes agent work reviewable - Impact 2: Custom UI allows us to create better experiences aligned with analyst workflows 1. Comments get picked up 2. Team and agent in one file 3. Every save is a reviewable version 4. Figures link back to their source This deck, open in Compound: a comment on this slide and the agent's reply ## Towards Autonomous Analysts in Finance Yijia Liang / Co-Founder Compound ### 00:09:40–00:10:00 # Our engine checks at every step, not at the end - Impact 3: the agent checks its own work as it goes **A code library** Nothing to look at **Microsoft's own interfaces** Every look needs a host **Our engine** A check at every step Slides are checked after every script the agent runs; spreadsheets at every save ## Towards Autonomous Analysts in Finance Yijia Liang / Co-Founder Compound [Flow diagram showing "Write" connected to "Deliver"] [Circular flow diagram with "Host" in the center, connected to "Write" and "Look" in a loop] [Circular flow diagram with "every step" in the center, surrounded by "Write", "Render", "Check", and "Fix" in a cycle] ### 00:10:00–00:10:20 # Our engine checks at every step, not at the end - Impact 3: the agent checks its own work as it goes ## A code library Nothing to look at ## ### 00:10:20–00:10:40 # Part 3: Massive, parallel context Up next: an agent that manages the context of everything ## Deal-team Analyst - Collaborates the way a deal team does - Executes ### 00:10:40–00:11:00 # A firm's context has both breadth and depth ### Massive context - Neither a deal's depth nor its breadth fits in a context window ## Breadth: one data ### 00:11:00–00:11:20 # Applying agent swarms to large number of files - Breadth: swarms of subagents over a data room 1. The lead agent slices the data room 2. Subagents read like analysts: list, search, open 3. Findings come back cited; the lead keeps conclusions, not files 4. Citations only after a re-read; mismatched quotes are refused Launch film frame: a lead agent and its subagents; illustrative Built for scale: work runs for hours and survives restarts; archives of up to 50,000 files and 40 GiB ## Towards Autonomous Analysts in Finance Yijia Liang / Co-Founder Compound [Diagram showing a central node connected to many smaller nodes, each labeled with a document or task name such as "Q3 LP exhibits", "Diligence tracker", "Synergy case", "Precedent deals", "Management deck", "IC memo draft"] ### 00:11:20–00:11:40 # Applying agent swarms to large number of files ## Breadth: swarms of subagents over a data room 1. The lead agent slices the data room 2. Subagents read like analysts: list, search, open 3. Findings come back cited; the lead keeps conclusions, not files 4. Citations only after a re-read; mismatched quotes are refused Launch film frame: a lead agent and its subagents; illustrative Built for scale: work runs for hours and survives restarts; archives of up to 50,000 files and 40 GiB ### Towards Autonomous Analysts in Finance Yijia Liang / Co-Founder Compound [Network diagram showing a central agent connected to multiple subagents, each labeled with a document or data source name.] ### 00:11:40–00:12:00 # Understanding a workbook is an indexing problem - Depth: a more accurate map of every model, built faster and cheaper A real workbook -> Index once, in the engine -> The map (Inputs, ### 00:12:00–00:12:20 # Understanding a workbook is an indexing problem ## Depth: a more accurate map of every model, built faster and cheaper [Flowchart diagram showing a spreadsheet image leading to a process: "Index ### 00:12:20–00:12:40 # Understanding a workbook is an indexing problem ## Depth: a more accurate map of every model, built faster and cheaper Built once per file, by code, before the model reads a ### 00:12:40–00:13:00 # Every deal team draws on the firm's knowledge → The company brain: organizational knowledge and human capital, built over years, put to work ### What accumulates, from files up ### 00:13:00–00:13:20 # Every deal team draws on the firm's knowledge → The company brain: organizational knowledge and human capital, built over years, put to work ## What accumulates, from files up to judgment - **The firm's judgment** - Built over years, put to work - **Conventions and templates** - Written by its people and read by the agent on every task - **Preferences and corrections** - From each deal team - **Cited findings** - Every cited finding from every deal - **Connected sources** - Such as SharePoint and Outlook ## What people stop doing - Finding ### 00:13:20–00:13:40 # Every deal team draws on the firm's knowledge → The company brain: organizational knowledge and human capital, built over years, put to work ## What accumulates, from files up to judgment ### 00:13:40–00:14:00 # Where can people fully delegate vs. handholding? → Delegate vs. copilot: which analyst workflows are ready to hand over ## Autopilot - Spreading financials - Comps refresh - Data room extraction - Deck roll-forward - Formatting and tie-out ## Copilot - An IC memo first draft - A downside case - A novel model structure - The investment thesis - The pitch narrative ### In plain terms - **Spreading financials**: keying statements into a template - **Comps refresh**: updating peer valuations - **Deck roll-forward**: last period's deck, new numbers - **Downside case**: the pessimistic scenario ## Towards Autonomous Analysts in Finance Yijia Liang / Co-Founder Compound [2x2 matrix chart plotting "Can the output be checked mechanically?" (Y-axis, higher is easier) against "How much deal judgment it needs" (X-axis, low to high), categorizing tasks into Autopilot and Copilot quadrants.] ### 00:14:00–00:14:20 # Where can people fully delegate vs. handholding? ## Delegate vs. copilot: which analyst workflows are ready to hand over ### Can the output be checked mechanically? Higher is easier ### Autopilot - Spreading financials - Comps refresh - Data room extraction - Deck roll-forward - Formatting and tie-out ### Cop ### 00:14:20–00:14:40 # Three things push the delegate line outward ## What moves a workflow across the line: the three parts of this talk 1. **Part 1** Solicit input as a deal team member 2. **Part 2** Checks you can run 3. **Part 3** Context you can reach ### Towards Autonomous Analysts in Finance Yijia Liang / Co-Founder Compound [Quadrant chart with Y-axis "Can the output be checked mechanically? Higher is easier" and X-axis "How much deal judgment it needs: low to high". The chart shows a "Delegate" region, a "Copilot" region, and points labeled "IC memo draft" and "Downside case" with an arrow indicating movement from "IC memo draft" to a point in the "Copilot" region.] ### 00:14:40–00:15:00 # Three things push the delegate line outward ## What moves a workflow across the line: the three parts of this talk 1. **Part 1** Solicit input as a deal ### 00:15:00–00:15:20 # Three things push the delegate line outward What moves a workflow across the line: the three parts of this talk ## Copilot 1. **Part 1**: Solicit input as ### 00:15:20–00:15:40 # Three things push the delegate line outward ## What moves a workflow across the line: the three parts of this talk 1. **Part 1** Solicit input as a deal team member 2. **Part 2** Checks you can run 3. **Part 3** Context you can reach [2x2 matrix showing workflow delegation, with the Y-axis "Can the output be checked mechanically? Higher is easier" and the X-axis "How much deal judgment it needs: low to high". The top-left quadrant is labeled "Delegate", and the top-right is "Where the line moves". Points like "IC memo draft" and "Downside case" are plotted, with an arrow indicating movement. The chart is associated with "Copilot".] ### 00:15:40–00:16:00 # Enterprise agents are coming for finance ## How we're building long horizon analysts ### Where the firm fell short - Background work must be IC-ready Excel and PowerPoint - A deal is run by a team in a shared thread - Context means 15,000-file data rooms and years of models ### 00:16:00–00:16:20 # Let's talk At an investment firm? I'd love to chat about putting agents to work on your deals. **Yijia Liang** Co-Founder, Compound ### 00:16:20–00:16:40 # The Highest-Signal Technical AI Conferences ### AI Engineer AI Engineer hosts AI developer conferences around the world for engineers, breakout startup founders, and AI architects advancing AI engineering at the world's leading ### 00:16:40–00:17:00 # AI Engineer Europe 2027 ## ExCeL London Europe's flagship technical AI conference returns to ExCeL London for three days, February 17-19, 2027 — bringing the world's leading AI labs, platforms, and tools to Europe's builders. - **Dates:** 02.17 - 02.19 2027 - **Duration & Location:** 3 DAYS - LONDON, UK - **Website:** [ai.engineer/europe](ai.engineer/europe) [Black ### 00:17:00–00:17:20 # AI Engineer NEW YORK ## Stay in the Loop Scan the QR code or view the schedule at [ai.engineer/schedule](https://ai.engineer/schedule) [QR code linking to https://ai.engineer/schedule] ### 00:17:20–00:17:40 # PLATINUM SPONSORS - Oracle AI Database - Sonar Makers of SonarQube - WorkOS - exa View Full Schedule at [ai.engineer/nyc/2026/schedule](ai.engineer/nyc/2026/schedule) [Logos for Oracle AI Database, Sonar, WorkOS, and exa] ### 00:17:40–00:18:00 # AI Engineer Europe 2027 ## ExCeL London ### 02.17 - 02.19 2027 Europe's flagship technical AI conference returns to ExCeL London for three days, February 17-19, 2027 — bringing the world's leading AI labs, platforms, and tools to Europe's builders. ai.engineer/europe [Black and white image of Big Ben and Westminster Bridge] [Black and white image of a grand building in London] [Black and white image of the London Eye]