Why AI Won't Replace Stock Pickers - Jack Woodruff, Candlestick & Vinoo Ganesh, Kepler
Jack Woodruff, founder and former CIO of Candlestick Capital and former Citadel portfolio manager, talks with Kepler CEO Vinoo Ganesh about what AI actually changes in fundamental investing. The conversation covers why humans remain the bottleneck, how AI compresses hours of prep into minutes and raises the baseline for analysts, why auditability matters more than speed, where edge comes from when every fund has the same tools, how fund teams may be restructured, and why a no-analyst hedge fund is unlikely to beat the best human-led funds.
Humans remain the bottleneck
Jack Woodruff, founder and former CIO of Candlestick Capital, argues that the core of stock picking will not change much, since alpha still comes from finding a divergent view. What is changing is how information is synthesized, what counts as table stakes, and what the traditional roles of portfolio manager and analyst look like. He thinks it is still too early to know how those roles will settle, because the tools are new even to their users.
Investing as time management
Investment management is fundamentally a time allocation problem. One analyst can know a single stock better than anyone, but each additional name dilutes attention and produces diminishing returns. AI’s most immediate effect is compressing preparation work: tasks that once took six to eight hours of background research can be drafted in minutes. That freed time does not mean fewer people or a free lunch. It is redirected toward proprietary research, channel checks, and industry backdrop work, because competitors will use their own time to dig for the divergent view that generates returns. What an analyst was paid for a year or two ago becomes baseline work, and the bar rises.
What junior analysts will still do
Woodruff pushes back on the idea that junior analysts will stop reading 10-Ks and transcripts. Anyone close to a name still needs to absorb its information directly. Summaries can shorten the process, but they cannot replace getting the information into one’s own head. Large-scale data processing and cleaning are a different matter and can reasonably be outsourced. In his view, the more important role for AI is directing attention: agents pointing analysts toward the right names and the right amount of time on each. The winners will be those who use these tools to decide where to spend effort in pursuit of a divergent view.
Smarter or just busier
Freed-up capacity gets filled, not left idle. More data, more proprietary calls, and more industry builds become expected, which lifts the baseline for everyone. Alpha then comes from the investment mosaic: three or four data points, drawn from different sources, that line up to form a view others have not reached. No single data point is the answer.
Auditability above all
Having started as a CPA, Woodruff says he would never sign off on a number without knowing where it came from. At Kepler, the ability to audit every figure is the most exciting part of the work. At Candlestick, human data entry errors were rare but still happened, and a single mistake could cost millions. The same risk applies to AI-generated models. If teams must re-audit every number a model produced, it is unclear how much time was actually saved. Trust in input accuracy and traceability is a precondition for letting AI near trading decisions or automated execution. Woodruff pushes back on the assumption that accuracy will simply improve over time.
Data security
As the buyer of software at many funds, Woodruff defers detailed security questions to his CTO. His instinct is that internal data should not leave the firm, and he would want strict data isolation before any vendor touched it. He frames the current period as a 1.0 phase focused on automation and data build-out. Once the data is in place, the next step is using tools to guide analysts toward where their time is best spent.
Where edge comes from when everyone has the same models
If everyone has the same information, edge comes from how people prompt, how quickly they combine human and machine judgment, and how well they spot what is worth deep work. The field work that once fed a buy decision now serves to identify special projects, names where a deeper look might reveal something others are missing. If AI could do all of that on its own, the question of what people are for becomes pressing. Whatever edge exists is short-lived, since talented, motivated, and highly compensated people quickly arbitrage away inefficiencies.
Fund structure in 2027
A fund launched in 2027 would still need senior stock pickers, but probably fewer analysts, along with new skill sets such as investigative journalists, deep project researchers, and data engineers. The traditional pyramid, with many analysts feeding senior pickers who feed PMs and a CIO, becomes more vertical. Woodruff wants the makers who generate returns and is less sure he needs as many preparers, though he may still need them to gather proprietary information.
Small funds versus large funds
Strategy matters more than size. A five-person team with slower idea velocity will likely run a more concentrated book. Idea velocity and the pipeline of good ideas are central to alpha: when an idea works, it is replaced by another. Covering more names with the same research quality lowers the share of the universe that must go into the portfolio, which improves the odds of holding winners, provided research quality holds across the expanded universe. Large multi-strategy firms face different dynamics, but the same underlying logic applies.
Skepticism about the no-analyst hedge fund
AI agents may be able to replicate a fund as it existed a year ago, but not the best human-led funds of the following year. Humans cannot realistically be removed from the loop in fundamental investing. Quantitative funds have operated without humans in the loop for decades, but fundamental investing without humans would reduce competition to whoever has the most compute and information.