Trace was acquired in late 2023. Immediately, I started using AI non-stop for two things: managing my own portfolio and prototyping startup ideas. I was working on a couple projects, but figured I'd build an AI product for CFOs.
2026 · Letter No. 1
BenchmarkAI began as a personal project. My career has been in tech – systems engineer at Facebook, founder of Trace, a software business sold to Paylocity – but my first love was markets and investing.
·The test·
To learn and to build credibility with the market, I created a benchmark of the frontier models. I already had a prompt library for investing tasks that I'd copy/paste across Claude, Gemini, ChatGPT, and Grok, so we built a system around it. Instead of measuring the agents on individual tasks, we put them in competition and told them to beat the S&P 500.
·Dec 1, 2025·
In the summer of 2025, reasoning agents became good enough to build real systems on. We started with the foundational data infrastructure – pipelines, workflows, and agents. On December 1, 2025, we launched four agentic investment teams, one per frontier model, each running three strategies: Risk-Adjusted Return, Long-Term Return, and Max Alpha Return.
A month in, I realized how lucky I was to be working on a problem I was obsessed with (the market) and doing what I loved (building product). We started asking a question to figure out which frontier model would deliver the best investing performance. We'll be working on that for a long time and it's now the company.
We are building an AI-native investment firm on proprietary systems and frontier models. AI has moved fast over the last 12 months. We've moved with it.
Here are the calls we made – and would make again:
- Put the leading U.S. frontier models in competition.
- Use code for determinism, agents for intelligence.
- Pick architectures and frameworks for agents, not humans.
- Develop opinionated context, workflows, and evaluations.
- Give agents maximum tools, capabilities, thinking, and time.
·Focus·
What we won't do matters as much as what we will:
- U.S. equities, long only
- No debt or leverage
- No options
- No prediction markets
The questions I care most about are still open.
When the agents recommend changes to themselves and to the systems they run inside, and we adopt them, can we tell which ones moved the portfolios? When we give an agent more time to decide, with more layers of agents working beneath it, do the returns change? Every piece of research is written down, and we keep a written rationale for every trade. I previously founded a startup called Trace because we wanted to help CFOs track every investment from plan to actuals. With BenchmarkAI, we're on a similar pursuit to understand how AI traces to financial returns.
·The mission·
We're using AI to build the portfolios, but we also want to use AI to build the firm. We believe the teams that use it best will build something that lasts. The people and the mission behind ours matter just as much.
Managing other people's money is a responsibility I don't take lightly. I can't promise a result. I can promise that we'll make well-reasoned decisions, work extremely hard, and communicate clearly and consistently. More on my story below.
Mike Gonzalez
The founder
At the Kelley School of Business, I studied finance and IT. I thought I was getting my start in “financial management consulting.” Really, I was building Oracle systems for Fortune 500 FP&A leaders – GE, Simon Property, Quanta Services, and my first client, TIAA-CREF, which brought me back every quarter to help run its planning cycle.
·Facebook·
From there, I joined Facebook to partner with FP&A leaders to build and manage their planning and long-range forecasting systems. I worried the role was too much tech. I wanted to run finance, not build systems for the people who did – so I left Facebook and joined the FP&A team at Zenefits, then one of the fastest-growing software companies of its era.
·Zenefits·
I worked closely with David Sacks, and when he took over as CEO he put me on the executive team to lead finance. I was 27, and I led a restructuring of a business that grew from 600 to 1,600 employees in a year and needed to figure out margins. Zenefits taught me two important things. First, I loved the chaos of startups, but wanted agency. Second, enterprise software lagged behind what I'd known at Facebook and our teams suffered.
·Trace·
That gap became Trace – a tool to put finance and the business teams on the same page about how a company hires and spends. We were the system behind procurement and headcount planning for leading public and enterprise businesses. I made first-time-founder mistakes: too much product and cross-functional complexity in our sales cycles. We weren't ready for growth funding, but the work became the foundation for Paylocity's move into the Office of the CFO.
·Paylocity·
I led product there for a year, and after our $325M acquisition of Airbase, we had a full spend and headcount suite.
·Indiana·
It was rough where I grew up in Northwest Indiana, but I had smart, hard-working parents who were equal parts demanding and supportive. We've always been a tight-knit family, so when I sold Trace, the first thing I did was fund an investment account for every child in our family tree. Back in 2018, when we founded Trace, we dreamed of a Salesforce-style giving program to fund investment accounts for underprivileged children. When Invest America was introduced in 2020 – an investment account for every American child – it was exactly the idea we'd hoped for. Now that it's live, we're honored that CFO AI, Inc., dba BenchmarkAI, intends to donate 3% of net revenue to fund Invest America accounts.