OBSERVATION
August 31, 2026
Every capital markets AI conversation eventually circles back to the same question: will it replace traders. It is the wrong question, and asking it is causing firms to miss where the actual change is happening — not at the decision layer, but in everything that feeds into it.
The trading desks seeing genuine productivity gains from AI right now are not using it to make trade decisions. They are using it to compress the research and monitoring work that used to eat most of a trader’s day: scanning news and filings for relevant signals, summarizing analyst notes across dozens of names, and flagging anomalies in position data that would otherwise require a manual review.
That is a meaningfully different claim than “AI is trading,” and it is the one actually backed by what desks are reporting: more time available for the judgment calls that still require a person, not fewer people needed to make them.
Trade decisions carry direct capital risk and sit inside a regulatory and compliance framework that makes full automation a much harder sell than the underlying technology alone would suggest. A model can summarize a 40-page filing reliably; that does not mean a firm is ready to let a model act on that summary without a person in the loop, and for good reason — accountability for a trading decision does not disappear just because a model assisted in reaching it.
The shift shows up less in headcount and more in how time gets allocated. Junior analyst time that used to go into first-pass research and monitoring is increasingly going into validating and refining what an AI layer surfaces, and more senior time is freed up for the calls that genuinely need experience: sizing, timing, and reading situations a model has no context for.
Firms that get this right are not the ones with the most aggressive AI rollout — they are the ones that correctly identified which parts of the desk’s workflow were mechanical enough to hand off, and left everything else alone.
In professional services firms, a surprising share of senior time goes to status updates a well-built client portal would answer without a call. What we have seen when firms fix it.
Most delays we trace in logistics engagements are not transport failures. They are five systems disagreeing about where a shipment was, noticed only when the customer asked.
Firms treat unbilled time as a partner-discipline issue. In practice it is the predictable result of time capture that depends on someone remembering to log it after the work is done.
Across energy and utilities engagements, the root cause behind preventable outages is rarely the asset. It is the gap between the systems that each knew part of the story.
Edge computing in industrial settings gets pitched as a clean architecture upgrade. The reality on most factory floors is messier, and that mess is exactly what determines whether a rollout succeeds.
Travel and hospitality personalization used to mean a tailored homepage. The interesting shift now is happening in the parts of the journey nobody used to bother personalizing.