Multiplier's Ian McInnis: “If humans survive to 2035, they'll do elite, object-level work”
The Bridgewater investor turned founder on why he doesn't believe in a permanent underclass, why building LLMs for investors is a different business from building LLM investors, and why WithAI is now Multiplier.
Key takeaways
- Multiplier's bet is that the durable human work under advanced AI is elite, object-level analysis — not middle management.
- “LLMs for investors” and “LLM investors” are different businesses: the second centralises advantage inside the largest funds, the first can be sold widely.
- Everything runs inside the client's own cloud, so neither Multiplier nor its model providers can see fund data.
- The pitch is capacity rather than replacement — analysts covering twice as many names at twice the depth, with the human still driving.
- WithAI is now Multiplier, a name taken from a client describing the product as “a massive force multiplier.”
Ian McInnis has had several careers already: Princeton mathematics, classical acting, a novel, and then Bridgewater Associates, where he led the firm's work on how AI would reshape the economy and how large language models could be used to invest. He left Bridgewater in early 2026 to build what is now Multiplier, alongside Ben Finch — a Princeton friend he ran a student investment group with — and Ryan Winkler. The product gives fundamental investment teams AI agents that research stocks and maintain projections, and it runs inside each client's own cloud, so neither Multiplier nor its model providers can see the data. Days before this interview ran, the company announced a $6M seed led by Lux Capital — with Bridgewater co-CIOs Greg Jensen and Karen Karniol-Tambour among the angels — and confirmed the rebrand from WithAI that he discusses below. His answers are published as he sent them.
From Bridgewater to Multiplier
Your bio reads like three different people: Princeton mathematician with a perfect GPA, classically trained Shakespearean actor, novelist, Bridgewater investor. Who is Ian McInnis in your own telling, and how do those threads connect?
I'm a sinner saved by grace. For the past decade, I've intended to subordinate my actions to God. From that flattering perspective, I gave up math, acting, and writing because I wasn't good enough for that to be my highest calling–––the calling where I could do the most good. As an investor, I could save thousands of lives by earning money and directing it to good causes. I'm now making a higher-variance, higher-EV bid at the same.
A less flattering tale is: I do what interests and challenges me, and many things have interested and challenged. But that dilettantism must end–––my cofounders, employees, and investors, count on me. Fortunately, the wandering urge is gone; I've never been so obsessed in my life as I am with this problem.
I've never been so obsessed in my life as I am with this problem.
At Bridgewater you led the questions “how will AI affect the economy” and “how can we use LLMs to invest.” What did you see while answering them that made you leave one of the world's biggest funds to build for everyone else?
First question first.
Internet people misunderstand comparative advantage as applied to human labor under advanced AI. In truth, if you plot human wages on the y-axis and AI capabilities on the x-axis, it looks like the PnL of Long-Term Capital Management. For the purposes of product design, you may as well focus on the first 99%.
That's why I don't believe in a “permanent underclass.” If AI stays subhuman, some tasks will be wholly automated, sure. But in a partial automation equilibrium, automated tasks should fall to ≈commodity prices, while most value is captured in the non-automated tasks.
Most innovation attention right now is paid to the step of automating. When folks look past it to ask “What will not be automated? What task is quintessentially human?” the most common answer seems to be “middle management.” Multiplier is betting otherwise. If humans survive to 2035, they'll do elite, object-level work. I left Bridgewater after realizing how few companies are building for that future.
Second one second.
You can build LLMs for investors or you can build LLM investors. They're different.
The second project rewards clean data, exclusive human data, compute, scaled engagement with the markets. Y Combinator called for a startup to become “the agentic Bridgewater,” or something, but you know who's building the agentic Bridgewater? Bridgewater. When LLMs are used as industrial-scale crystals of investment thought, they're a centralizing technology, much like alt data and factor-neutral portfolio construction.
“LLMs for investors” are a different story. Their benefits can be spread far and wide, complementing independent managers' idiosyncratic alphas. End-to-end “LLM investors” can't be sold to diverse buyers (it ain't alpha if everyone has it), but the natural place for great investor-complementary tech is in a dedicated company. If everyone else is paying for credit card data, it's less urgent for you–––probably mostly priced in. But if everyone else is paying for Bloomberg, you better buy one too so you don't fall behind.
Most saliently, nobody's selling this yet. Bloomberg will wane; so much more is possible in 2026. I left Bridgewater after realizing that there was more than a damn good product to build here, but an enduring business.
You can build LLMs for investors or you can build LLM investors. They're different.
You and Ben Finch became friends at Princeton running an investment group together, then reunited after independently landing on the same startup idea. What was the conversation where you realized you were circling the same problem?
I was riding to a church retreat in a taxi, and he called me. When Ben Finch calls, you answer. For the next hour until cell service dropped, we hashed out the future of investing.
What actually changes for an analyst
For readers who don't live in finance, describe an analyst's ordinary Tuesday before your platform, and the same Tuesday after. Where does the time actually go now?
Before: root through news, dealer research, and expert calls to see if there's anything new, true, and useful about your forty stocks. Read those articles and try to remember them. Report on yesterday's earnings calls and ping your portfolio manager. Keep building that model in Excel. Get lunch. Et cetera.
Today: glance at your dashboard with articles pre-filtered and updates pre-drafted. You still read the articles and review the updates––you're still the driver here, and your understanding matters–––but you can cover twice as many stocks, at twice the depth. Instead of building the Excel model yourself, AI can execute the gruntwork for you, checking in when it's time to make subjective decisions.
Obviously, that's just illustrative–––analysts have dozens of discrete tasks, and Multiplier helps with all of them. (Well, not lunch.) Some of their spare time is spent on new tasks they never would have done before: defining a framework to first-pass screen all the stocks in the world, running quantitative-qualitative analyses of their own stockpicking history, designing custom dashboards–––but the main effect is for coverage to expand and deepen. Each analyst becomes three analysts.
Each analyst becomes three analysts.
Everything runs on the client's own cloud, where neither you nor Anthropic can see their data. That is a harder way to build software. Why was it non-negotiable from day one?
It's just one element of our extreme defensive posture. The cybersecurity news has been crazy, but you ain't seen nothing yet. Ben, Ryan, and I all bought N95s in January 2020. That's the type of security mindset we bring; we're paranoid so our customers don't have to be. Attacks on hedge funds are already ramping up. “Minimize your surface area” is just common sense now.
We're paranoid so our customers don't have to be.
You sit face to face with investors and interview them so the AI can learn each firm's unwritten knowledge. What has surprised you most about how great investors actually think?
“What's our source of edge” doesn't keep most great investors up at night. They know their edge: out-think the next guy. This has always been tenuous and ephemeral. They know it will get even harder as AI makes cognition abundant.
Earning trust in a secretive industry
Three months after launch you were live with four hedge funds at over $100k ARR, and your angels include Greg Jensen and Karen Karniol-Tambour, the co-CIOs of your former employer Bridgewater. How did you earn that much trust that fast in an industry built on secrecy?
Greg and Karen had seen my work; Mercator (our first client) had seen Ben's. Beyond that, there's an ever-growing swath of investors that hear our pitch and really Get It, really understand what the future needs to look like. Folks with that spirit are excited to get in early.
WithAI becomes Multiplier
Now the news: WithAI is becoming Multiplier, the name of your platform, which itself came from a client calling it “a massive force multiplier.” Walk our readers through the rebrand. Why now, and what should the new name tell people about where you're headed?
'WithAI,' although pleasantly simple, wasn't as evocative to customers as it was to us. To me it was an alternative to “ByAI” or “AIInstead”; it evoked human-AI symbiosis (or at least collaboration). 'Multiplier' seems to do a better job. We turn one analyst into three. We make good investors great. We make great investors superhuman. (Conversely, if you have no alpha, don't contract us! No multiplication can turn zero positive.) That's our telos.
We turn one analyst into three. We make good investors great. We make great investors superhuman.
What changes for your clients under the Multiplier name, and what stays exactly the same? Was there any internal debate about retiring a company name that literally says what you do?
When we realized that “WithAI” wasn't even causing fun who's-on-first-type situations, that was the last straw. I will miss the W/ logo I designed, but it's inferior to Ben's M-that-evokes-an-X-and-also-evokes-duplication. (I'll also miss how surprisingly easy the SEO for “WithAI” was.)
The long view
A lighter one for the actor in you: if the first year of this company were a play, which one would it be, and which act are you in right now?
You know how Jeff Bezos popularized “Always Day 1”? We're going for “Always Act IV, Scene 3, of Henry V.” I have delivered the St. Crispin's Day speech more than once in our office.
We're going for “Always Act IV, Scene 3, of Henry V.”
You've argued that we shouldn't build superintelligence, but that near-term AI doing real work has enormous value. Ten years from now, what would Multiplier need to have become for you to feel that argument was proven right?
Multiplier has nothing to do with it. No human business survives superintelligence, even a benevolent one. (Obviously, we're talking about Real-Deal Superintelligence here, not Marketing Term Superintelligence.) A permanently benevolent one, obviously, would prove me wrong. As would our AI not-so-useful, but that's already falsified.