Morgan Stanley's AI head thinks tech skills are terminal. Is he right?
The notion that generative AI being used to 'replace' developers is a sentiment that's been perpetuated by a variety of figures in finance, from UBS' chief economist to Goldman Sachs MDs in user experience. The latest to chime in is Jeff McMillan, head of AI at Morgan Stanley.
McMillan thinks that "the demand for deep technical expertise will decline with the rise of people skilled to leverage [AI] tools." He suggests that the key to successfully implementing it is "a robust training program", which will teach staff how to negate the need for that expertise.
How far will AI reduce headcount? Barclays' head of FIG banking has said that AI could reduce engineering headcounts by almost 90% in its current state. Not everyone agrees, however, including the CEO of a startup focused on training LLMs.
Nicholas Khami, CEO of RAG infrastructure firm Trieve believes software developers are being "gas-lit" by the consensus that coding is a stranded asset.
One of his key arguments is that LLMs won't understand the "downstream effect of fixing [a] problem" on your codebase. Because of this, he fundamentally disputes the fact that AI will "build your whole software product, and you need drastically fewer people."
Responses to Khami on HackerNews echo this sentiment. One says the focus of AI is "very narrow" at present and functions better as a "super robust auto complete." Another said that "If I had to give a junior the level of direction I give an AI bot every day, that person would be on a PIP pretty quickly."
Goldman Sachs seems to be in agreement with Khami. It published a report on AI this year, in which equity research head Jim Covello said AI works best only as an efficiency booster, but "estimates of even these efficiency improvements have declined." These efficiency gains will likely result in fewer opportunities for engineers, but perhaps not to the extent widely prophesized.
However, AI still has exponents who think AI will reduce demand for engineers. One HackerNews respondent says Khami "seriously underestimates the pace of progress in AI," and that specific tasks it can't do today may be done easily tomorrow. Another said, "if you can't get an LLM to do something, you probably aren't thinking creatively enough."
In the hedge fund space, Man Group has been one of the biggest proponents of GenAI, going so far as to build its own chatbot, ManGPT. The firm's head of data and machine learning Tim Mace said at the Quant Strats 2024 conference today that around half of its 600 active coders use GenAI today. However, Mace said it's not not about replacing coders but "making them more effective."
Success will also depend on the LLM you're using to code. Ben Reesman, a former engineering manager at Meta (then Facebook), says Anthropic's Sonnet models are "a class apart" for code generation. Tim Shelton, who previously went viral for automating 633 Leetcode problems in 24 hours, had a similar opinion. Mace of Man Group says the fund has started to move from GPT-4o to Anthropic's models for ManGPT for this reason.
Love it or hate it, AI is worming its way into the workflow of almost every engineer. 76% of respondents in the 2024 Stack Overflow survey said they use the tech, and only 6.4% of respondents were actively unfavourable on their use. It still has a long way to go, however; only 2.7% of respondents said they highly trust AI tools.
Have a confidential story, tip, or comment you’d like to share? Contact: Telegram: @AlexMcMurray, WhatsApp: (+1 269 237 3950). Click here to fill in our anonymous form, or email editortips@efinancialcareers.com. Signal also available.
Bear with us if you leave a comment at the bottom of this article: all our comments are moderated by human beings. Sometimes these humans might be asleep, or away from their desks, so it may take a while for your comment to appear. Eventually it will – unless it’s offensive or libelous (in which case it won’t.)
