The Investor's Edge Lives in Their Head — Anekha Sokhal, Moshi
Read the talk
The Investor’s Edge Lives in Their Head
Anekha Sokhal explains why financial AI should absorb the market-data grind, expose sources and assumptions, and leave consequential trading judgment with the investor.
From a talk by Anekha Sokhal
At a glance
Ideas worth remembering
Investor edge combines earlier awareness, detection of second- and third-order effects, and enough conviction to size a trade.
Widely available models and connectors do not reproduce the tacit context accumulated through years of market experience.
A production financial system must solve security, storage, reliability, current-data, cost, and accountability problems that a weekend dashboard can ignore.
The useful human-machine boundary assigns collection, cleaning, monitoring, and cross-checking to machines while leaving consequential judgment with the investor.
Fresh context is valuable only when claims are traceable and the system exposes the assumptions behind its conclusions.
The hard open problem is not gathering more data; it is determining how much tacit investor context can be represented without erasing human responsibility or apprenticeship.
More market data creates an attention gap
Financial investors face a bad asymmetry: the available data keeps multiplying while the attention available to interpret it does not. Anekha Sokhal, founder and CEO of Moshi and formerly a derivatives trader, opens with the familiar inputs—PDFs, paid subscriptions, podcasts, social feeds, books, and newspapers. Access is no longer the scarce resource. Synthesis is.
The presentation cites 175 zettabytes of data produced in 2025 to dramatize that growth, then draws the more useful relationship: everyone receives a similar flood, but an investor’s attention remains roughly flat. The widening space between those lines contains the work—deciding what matters, connecting events, and discarding noise. The recording does not establish the cited global-data estimate independently, but the operational constraint does not depend on its precise value.
That constraint changes where AI should sit in the decision process. Sokhal contrasts the human brain’s roughly 20-watt operation with the much larger energy demands of machine-learning infrastructure, then asks whether AI should replace the reasoning engine or feed it. The energy comparison is rhetorical rather than a controlled measure of equivalent computation. Its architectural point is sharper: use machines where their scale helps, without assuming that scale transfers responsibility for a financial decision.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
Edge is learned context, not a downloadable component
In this account, an investor’s edge has three parallel parts:
- Know sooner: detect a relevant development before other market participants.
- See blind spots: follow second- and third-order effects rather than stopping at the first event.
- Act with conviction: understand the case well enough to size a trade when the evidence supports it.
A model, prompt, skills file, or collection of MCP connectors may help with those tasks. If everyone can acquire the same component, however, possession of that component alone is not a differentiated edge.
The harder-to-copy part lives in an investor’s accumulated context: years of watching markets, learning beside experienced traders, and remembering how comparable events unfolded. That knowledge can leave a firm when the investor moves or retires. AI therefore has a second possible role beyond answering questions: help make tacit analytical context more usable without pretending that every instinct has already been captured in a database.
Before a trade, the investor continuously monitors markets, relates current events across many data connections, finds historical analogues, asks why those earlier events occurred, observes the market response, and applies internal models. Only then comes the decision. Treating the final click as the whole job hides the research chain that produced conviction—and hides the portion machines can most plausibly accelerate.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
A weekend dashboard is not financial infrastructure
AI makes the old staffing model look deceptively compressible. A conventional desk might combine traders, quants, data scientists, and data engineers; a prototype can appear to replace that stack with one person and a few API calls. The illusion breaks when another person must rely on it. Production introduces security, databases, storage, reliability, and metered inference costs. The dashboard is only the visible layer.
Two tool patterns create different burdens:
- Pull tools: the investor prompts a model and supplies the missing context. If the model has stale temporal knowledge, the user wastes time correcting basic premises—Sokhal’s example is debating whether an event such as an FOMC meeting is today or tomorrow.
- Push tools: the system continuously sends developments to the investor. That can expose useful surprises, but without ranking and filtering it merely recreates the information flood as a data-engineering problem.
The product problem is to combine current information with investor-specific relevance, rather than choosing between stale answers and indiscriminate alerts.
There is also a training cost. Junior staff traditionally sit beside traders, listen to the desk, and absorb repeated patterns—“one ear on the desk and one ear” on the surrounding market. If every professional becomes a one-person pod supported by private agents, fewer people may share the environment where that context is learned. The efficiency calculation can therefore miss the loss of apprenticeship and institutional memory.
Full agent autonomy creates a more immediate boundary. The other side of a financial decision may be a client, a company, or a physical cargo. When something fails, the institution needs a person who can explain why the decision was made. An agent’s apology after leaking a credential is neither accountability nor recovery. The recording warns that small errors can compound, although it does not specify the controls, execution paths, or failure measurements behind that claim.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
Split the grind from the judgment
The proposed correction is to keep the brain in the loop and feed it better. Before deciding, an investor reads, gathers, cross-checks, and watches the market. Conflicting sources create additional questions: what is true, where do accounts diverge, and which source deserves trust? Those activities consume attention before judgment begins.
This creates a practical division of labor. Machines handle the grind: collecting material, cleaning data, monitoring updates, comparing sources, and preparing evidence. The investor handles judgment: deciding what the evidence means, whether it changes the thesis, and whether to trade. The line is not based on a claim that machines never reason; it is based on retaining human responsibility where uncertainty meets consequential action.
That boundary also changes the interface goal. A trader who says “I just want to trade” does not necessarily want a conversational partner for every fact. Useful assistance should reduce avoidable debate, push relevant blind spots, and preserve enough provenance for the investor to challenge the result. Conversation is optional. Inspectability is not.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
Moshi turns continuous monitoring into inspectable context
Moshi applies this split as a continuously running quant analyst for institutional traders. In the recording, six agents monitor markets around the clock and combine external market information with investor material from PDFs, email, and subscriptions. The orchestration is described as deterministic, but the talk does not reveal its routing, retry, consensus, or execution controls, so “deterministic” should not be read as a complete technical specification.
Scale alone is not the product. Sokhal reports that Moshi has processed 2.4 billion data points and learns 300 new things in four seconds. New context becomes useful only if the investor can ask where it came from, whether it is trustworthy, and whether the source’s opinion deserves weight. A confident statement with no reliable basis has no trading value.
The trust layer therefore measures hallucination, checks individual claims for traceability, and exposes the assumptions behind a conclusion. This mirrors the accountability expected from a human analyst: an investor wants both the answer and the path used to reach it. The recording does not specify the hallucination metric, evaluation set, acceptance thresholds, or treatment of conflicting sources, so it establishes the intended mechanism rather than its measured reliability.
The resulting flow keeps execution authority with the investor. Market and private inputs feed monitoring agents; their output passes through claim, source, and assumption checks; the resulting context reaches the person who decides whether and how to trade. Moshi’s job is to expand what that person can inspect, not to become the accountable trader.
Sokhal reports up to 35% greater efficiency, arguing that saved research time creates headroom for conviction and deeper analysis. The 2.4-billion scale figure, the four-second learning rate, and the efficiency claim are company-reported; the talk supplies no study design or denominator. The current supplied Moshi page also describes four specialist agents plus a manager rather than the recording’s six-agent setup, so the recorded topology may represent a different product version.
Push delivery is central to the intended benefit. A prompted model receives context from the user; a monitoring system can instead introduce a cross-asset development the investor did not know to ask about. That surprise can expose a blind spot and trigger a focused deep dive across broad market data and investor-specific material. Relevance matters as much as freshness: an endless stream of unranked novelty would simply rebuild the original attention problem.
Continuous coverage, reliability, and traceability are what turn this from a weekend tool into infrastructure. Other people must be able to trust the system, inspect its claims, and reconstruct the reasoning that informed a decision. The comparison with a human analyst is instructive: the important property is not that the analyst never errs, but that the institution can question the work and assign responsibility for acting on it.
How machine-scale monitoring feeds human judgment
Read the diagram as text
- Market data. inputs — Continuous developments across markets and other external sources.
- Investor data. inputs — PDFs, emails, subscriptions, and other investor-specific material.
- Monitoring agents. processing — The recording describes six agents watching and synthesizing information around the clock.
- Claims, sources, and assumptions. verification — Outputs are measured for hallucination, traced to sources, and accompanied by assumptions.
- Human investor. decision — Interprets the context and retains responsibility for the trade decision.
- Market data → Monitoring agents: continuous input.
- Investor data → Monitoring agents: personal context.
- Monitoring agents → Claims, sources, and assumptions: generated claims and synthesis.
- Claims, sources, and assumptions → Human investor: traceable context for judgment.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
The unresolved question is how much edge can be captured
The ending resists a clean automation claim. If edge lives in accumulated human context, how much of it can a machine capture? The talk leaves that open. Historical comparisons, source preferences, recurring assumptions, and explicit analytical procedures can be represented. Instinct formed through years of market exposure may be harder to extract, and encoding it can change what gets taught to the next generation.
Moshi’s public proposition is institutional-grade analysis without requiring every investor to operate a personal quant and data-science team. The ambition is economic as well as technical: spread the monitoring and synthesis capacity of a large desk across more individual decision-makers. But the final decision boundary remains the same. Before a trade, the machine can gather, compare, challenge, and explain. The investor still has to stand behind the choice.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
Resources
From the talk
The product page expands on Moshi’s market-monitoring, source-traceability, security, and invite-only institutional workflow. Its current four-specialist-plus-manager description differs from the six-agent architecture stated in the recording.
Related talks
- Build for the Memo, Not the Demo — Notes from 200 Investment Committees
A complementary finance-AI perspective on source weighting, contradictions, linked evidence, and producing defensible investment work rather than impressive demonstrations.
- How Kepler Built Verifiable AI for Financial Services
Explores the same need for auditable financial reasoning by separating probabilistic AI work from deterministic retrieval, computation, verification, and provenance.
- Trading Desks to Clinical Trials: Parallels in Applied Vertical AI
Extends the discussion of durable advantage in vertical AI through domain expertise, proprietary data, observability, and expert judgment.
Read the complete timestamped transcript
- 0:12
Hi, everyone. Thank you so much for coming to my talk. I'm Anekha. I'm the founder and CEO of Moshi, and I'm gonna be talking about a financial investor's context that lives in their head and how we can use that when we're building with machines today. So I wanna do a quick show of hands. How many people in this room are overwhelmed by the amount of data they have when making a decision?
- 0:37
Okay. And how many feel like they've got less attention now today with all the amount of data? And so that's the exact paradox that financial investors have every day. So in 2025 last year, the world made about 175 zettabytes of data. That's expe- exponentially higher than what we had two decades ago. So before we were lucky if we had an internet connection, we had newspapers, books. But
- 1:07
nowadays we have all of that, plus the context that lives in PDFs, subscriptions, lots of different, uh, podcasts, and also a stream of never-ending information coming on X and LinkedIn. So what does that mean for financial investors? Because more data does not equate to more information. It actually just gives everyone the same flood of information, but your attention is the flat line across everything. So for financial professionals who actually are
- 1:37
paid on financial synthesis, what happens between the gap between attention and data? It just gets wider and wider and wider.
- 1:47
So who am I? I've sat on both sides of the table. So I was... Uh, I've spent my career fascinated by data, um, and I've spent my career working in it. So I was a tr- derivative trader for nearly a decade, and then I moved into data science and machine learning. Uh, so I got a Fulbright scholarship to study machine learning, and I got a paper at ICLR, uh, working on machine learning systems with NASA, and now I've married those two. So those two worlds being Moshi. Moshi is your 24/7 quant analyst,
- 2:17
which works watching the markets 24/7 and understanding what is going on and what matters to a financial investor. Having sat on both sides, I can understand what the strengths are for both of them, and that's what I wanna go into deep today.
- 2:34
So I wanna talk about the machine that everyone is trying to beat. So a human brain, to give you context, runs on about 20 watts, which is the power of a dim light bulb. Uh, so it's 86 billion neurons in the brain, and it's making connections that we still can't understand. And that is on the power of a light bulb. And the systems that we are trying to imitate, which we're trying to imitate just a sliver of that power, they
- 3:04
need a whole power plant to, to run. So just in terms of efficiency, the human brain is something like multiple millions, multiple times more powerful than whatever systems that we're trying to... Uh, whatever we're trying to build. And so that actually... What that actually means is it, it's important for where you're trying to point that technology. So at which point in the decision-making, uh, cycle do you point the technology, which is AI, which we're what... We're
- 3:34
building? Because if the most efficient reasoning engine sits between your two ears, then... And it's running on a 20-watt light bulb, where does machine learning actually sit in that? And so do you try and replace it, which is what AI is trying to... What I feel that AI is trying to do every day, or do you actually feed it?
- 3:56
So let's take a few steps back to understand how a trader's brain or a financial investor actually thinks. So what do people actually want? Any investor you speak to is looking for the same thing, edge. What is edge? So edge is in practice just three things. I want to know what's going on before anyone else is. I want to know my blind spots, so tell me the second- and third-order derivatives of what's happening. And also I want to have the conviction on what I'm doing so that I can size up on a trade that I believe in.
- 4:27
All those three together give you edge. So now here's the problem. If you're building in this space and everyone has access to the same models and they have access to the same systems, then what is edge, right? Is it a skills file? Is it a whole bunch of MCP connectors? Is it just a better prompt? Is it a better model? Every con- everyone has access to all of that. So if everyone has access to all of these things, then
- 4:56
what is edge? So the edge is actually what you can't download. It what... It's what sits in an investor's head. And how do they get that? It's decades of working and learning from some of the best traders in the industry and sitting next to them. And here's the concerning part. What happens when you have an investor and everything lives in their head when they move firms or they retire? That IP leaves with them.
- 5:24
So I think before we even start to think about how AI fits into all of this, we have to actually understand what a financial investor is doing. They're keeping up with the markets, which are now moving 24/7. They're trying to understand what is going on and how they relate to each other, so those billions of different data connections. They're trying to compare and understand the historical significance of the event happening right now, what happened in the past, why it happened, and what the market did when it
- 5:54
happened, and trying to piece all of that together and then run it on the models that they have. And then the last part is actually make a decision. So there's a lot that takes place before actually making a financial decision. So I just wanna keep the last bit with you, which is the decision-making. So let's take think about what AI actually changed. So before, if you worked in finance, you'd have a team of quants, data scientists, data engineers, traders
- 6:24
who are all working together to build this decision-making engine. So it's a whole stack of very expensive and very qualified people. However, with AI, these tools made you believe that you actually don't need any of these people. What you need is one person, a few API calls, and you've replaced that whole stack of people. But here's the gap that no one actually talks about. Once you've spent your weekend building a dashboard, the moment you actually want to productionize that and share it with
- 6:54
one more person, you now have to think about security, a database. You have to think about storage, reliability, and the biggest thing in all is credits. How do you actually run the systems? And so what we've seen is pricing has changed. People are no longer pricing by tools, but they're pricing for every API call. So now if everyone is building the same tools, what actually makes it different? And so there's two types of tools in finance. One is the p- tools that you pull data
- 7:24
from, and one, the tools you push data to. And so if you're prompting a model, you are the context. You're explaining to it. You're arguing with a model that's frozen in time, which is telling you, "Hey, go to bed at 2:00 AM. It's late now," when actually it's 3:00 PM on a Wednesday. And all of that, if you think about comparing it to reasoning with a human, it's absolutely no comparison. And then you've got the type of tool which pushes data to you, and that's the tool where it's flooding you with information, and that's a data engineering problem.
- 7:54
So when you have these two tools, and you've got different problems that you're solving for, one is context, and then the other one is the amount of data, you've got to think about what the trader's actually using and what's... what they think about in each of these different tools. And the one thing that it... we're seeing is quietly breaking is that in finance, when you join a bank or you join, join a fund, you sit next to a trader or a team, and you absorb everything that's going on, and you
- 8:23
learn from them, and you just listen to everything. You have one ear on the desk and one ear around everything else that's going on, and it's consistency that pays off. And if everyone is going to be running their own teams as one-person pods, then what happens to that future generation of context that's coming in? And that's the real cost that right now people are not pricing in. And so in finance, you can't just hand the decision-making loop to a team of agents. Why? Because the other side of the trade is a client. If
- 8:53
it's in a commodities trade, it's a cargo, a physical cargo. If it's a company, it's a team of c- people. And so what happens in finance when something goes wrong is you need a human to go and stand up and say, "I made the decision for this reason." If an agent leaks your API key, all it says is, "Oops, I'm sorry. I shouldn't have done that. Please rotate it." And you can't have that in finance because im- like, an agent is just self-correcting, and the one small error is just exponentially getting bigger.
- 9:26
So this is the question that I've been leading to, which is if the brain is the most powerful tool we have, why are we spending everything trying to push it out of the loop? Why don't we flip that problem, and instead of replacing it, why don't we just feed it? And so if you think about what's in a trader's head, they've got everything leading up to the, uh, to the trade itself, but before that, there's a whole bunch of other work that takes place. They're reading. They're gathering information. They're
- 9:55
cross-checking. Uh, they're watching the market. Because we have so much information now, the biggest question people have is what is the truth? What is, what is actually the different sources saying? What is the divergence of information? How can I trust what is out there? And once I've done all of that, then I've actually got the decision. So if we split it into two different parts, we understand that some of it, well, is all the grind, the reading, the gathering, the data cleaning,
- 10:25
and some of it is judgment, which is the decision. And machines are incredible at the grind. They're incredible at cleaning data. They're incredible at gathering data. And then the human is incredible at the decision-making. So why not just split those into two separate parts? And this is what brings us to Moshi, which is what we've built. It's a 24/7 quant analyst for institutional traders, and what that means is we are running a team of six different agents, which are built in
- 10:55
a deter- de- deterministic manner, which are watching the market 24/7, more data than is humanly possible, plus the investor's data, which lives in PDFs, emails, different subscriptions, understanding all of that in a way that is making the t- the trader a better decision-maker. So Moshi has read 2.4 billion data points, and in the four seconds it took you to read that sentence, Moshi has learnt 300 new things.
- 11:25
And in those 300 new things that it's learnt, that's new context for the model. The issue with that is new context is worthless if you don't actually know is it trustworthy, where does it come from, do I even value this analyst's opinion? Because in finance, even if it's a really confident answer, if it's wrong, then it's worth nothing. And so what do we do? We measure it. We measure the hallucination. We measure every claim. We sh- see how it's
- 11:54
traceable, and we also, uh, give the user the assumptions that we've had and how we've led to that decision. Because if an investor asks you about your opinion, they don't just wanna know what you think. They want to know how you came to that reasoning
- 12:11
And so what we do is we leave the decision-making with the human because we believe that that is what humans do best. There's no machine more powerful than the brain. But everything else that you need to run is what we give you. So we believe that every single firm is running these desks of quants, of data engineers, of data scientists, and that's what we give you because it's impossible to run that per human. It's impossible to run your own quant and data science team, and it also doesn't make sense. Economics-wise, it
- 12:41
just is too much to run per person. And so that's what we do. We run an institutional-grade anal- uh, analyst how it works in bigger institutions to actually give you that conviction to make better trades. So what does that actually buy you? It buys you more efficiency, up to 35% more, so that you can actually spend time to have the head space to have more conviction on the trades that you wanna do. So if you speak to a lot of traders, they'll say to you, "Hey, I don't wanna speak to anyone all day. I
- 13:11
just want to trade." So why are you giving these people just LLMs to talk to and to debate, "Hey, is FOMC today or is it tomorrow?" And then the LLM is saying it's tomorrow when it's today. That's just a complete waste of time. And so that's the opposite of what investors want. They want something that can give them the power to reason more and use their brain. So Moshi isn't you. Moshi isn't a model. It's, um-- it is you on steroids, which is the same instinct, but
- 13:41
fed by all the data that lives in different parts so that you can actually spend your time decision-making instead of debating when FOMC is.
- 13:52
In, in terms of the scale, we watch over two point four billion data points, and it's growing every millisecond. And so that's decades of derivatives trading knowledge around the clock, which is what it-- no, no one person can build. And then it's the pushing of the data. So when you're pulling data from an LLM, you are giving it the context on an old model. When you're getting pushed new data, it's forcing your brain to think about new things and make connections in ways that you didn't
- 14:21
initially think about. So that means that you're then thinking about these blind spots which you've been told about, which you are then trying to understand, "Okay, well, I wouldn't have caught this, so what am I gonna do with this new information?" And then it's the deep dive, the ability to deep dive into one thing with a 360-degree view of information that you wouldn't have had before. And so we do that with the big data and all of the specific data.
- 14:50
And so how does it differ from a tool that you're building on the weekend when a tool that you need to build 24/7 on every market on big data, which is running in a way that you can understand the context, which is reliable and trustworthy, that becomes infrastructure. It's no longer a weekend tool because you need it to be something that other people can trust and you can trace back because that's what you have with a human, which is someone being able to stand up and say, "Hey, I made this decision because of XYZ,"
- 15:20
not calling an agent who's never-- who's just gonna apologize to you.
- 15:25
So the one question that I think we haven't answered yet, which is if real edge lives in a human brain, how much of a machine can it act- how much can a machine actually capture? And I don't think we have an answer to that yet, um, and I think whoever answers it is not building a product. They're building a new way of thinking in finance, and that's what we're trying to explore.
- 15:49
So we've been building very quietly with, uh, a few of the biggest players in the trading industry in finance. And so this is the first time we're opening up to talk to the public, uh, because we believe that every single person should have access to institutional-grade analysis without having to run their own data science or quant team. Um, so if that's interesting to you, come talk to me. I love talking about everything to do with data and finance and financial engineering. Um, and yeah, before you trade, we want people
- 16:19
to talk to Moshi. Thank you.