AI engineers are the whole team now, a leadership risk

The signal. Anekha Sokhal, founder of Moshi, came to VIA I to talk about a recommendation engine for commodity traders. The more useful signal for engineering leaders was what attendees told her: AI engineers are under pressure because management now says, "We don't need a team. You are the team." They are expected to ship a system that normally takes ten people, alone, and then maintain it.

The product. Moshi processes futures time series, technical indicators, supply and demand data, news, X posts, audio, and user-specific information such as LNG ship movements through the Strait of Hormuz. Sokhal claims that in the four seconds it takes to say a sentence, Moshi learns 300 new things. The output is pushed to the trader, not pulled, and she calls the blend "Bloomberg, Claude, and Netflix."

The recommendation layer. The design idea is that a trader looking at oil needs second and third order connections, such as an energy equities company doing something that indirectly affects oil. A flat news stream will not surface that, but a connected graph and a recommendation system can. It is a credible product thesis, but the transcript offers no precision, recall, or production metrics to back it.

The adoption problem. Sokhal says the real barrier after building a dashboard is UI; a dashboard is "your brain in a dashboard," and new users will not incorporate it into their workflow. Platforms like Spotify and Instagram push data and hook users without effort. Moshi's internal thesis is that if you solve for the recommendation system, you are naturally solving for the UI problem too.

The model tradeoff. For messy client PDFs, Moshi built its own vision language model because clients do not want their data sent to third parties. She argues that teams with strong engineering capacity should build, while teams without should buy. That is a sensible starting point, but it is also a vendor's answer to a decision that depends on security, compliance, and long-term maintenance cost.

What's missing. The conversation is light on evidence: no benchmarks, no customer outcomes, no false-positive analysis, and no governance discussion. There is plenty of momentum, including meetings in Houston and 2,500 applicants for a job left open four days. The lasting lesson for engineering leaders is organizational, not architectural: one AI engineer cannot safely own a ten-person system.