Which enterprise function should be the first to hand decision-making over to AI?
5 recorded positions from 5 people, first said Apr 28, 2023. They do not agree — the readings below are what each one actually argued.
Legal first because text heavy work matches text in text out models
Sarah Guo · Apr 28, 2023
Law is a strong early AI market because it is a text-in, text-out profession where large models can do much of a first-year legal associate's work end to end.
The profession's inputs and outputs are text, which is exactly what large models handle, and replacing that work is very valuable.
16:19 20VC: In AI Who Wins? Startups or Incumbents? What Happens to Wealth Inequality? Why Will $10BN+ Companies Only Have 10 People | Why Defensibility in Startups is BS & Speed is Everything? Why Large Groups Worsen Decision-Making with Sarah Guo
Lucas Swisher · Feb 23, 2026
Law was one of the clearest early markets for language models because LLMs are text-in-text-out and law is one of the most text-heavy professions, with document generation and analysis as the obvious first use cases.
He had already seen early traction in document generation and analysis, and Winston's articulation matched the thesis exactly.
56:59 20VC: Inside Coatue's $70BN Machine: Why Price Matters Least | Why Mega Markets are the Most Important | How to Assess Durability of Revenue and Margins in AI with Lucas Swisher
Also on the record
Alex Bouaziz · Oct 22, 2025
AI will impact Deel most in pure operations, automating the manual tasks currently done by thousands of operations staff via agents
Deel is an operations-heavy business with a couple thousand operations people worldwide doing tasks that can be automated through agentic software and the infrastructure being built
33:13 Operations is first because manual repetitive tasks are most automatable
Nikesh Arora · Jun 22, 2026
Marketing is the function where enterprises should relinquish control to AI first, because the frontier models already hold the world's best marketing training data
Marketing is by definition public domain, so the training data is already in the models; the only thing left to teach is brand and tone of voice, and models can already flag collateral that is inconsistent with a company's brand
14:02 Marketing first because its training data is already public
Micha Kaufman · Jun 9, 2025
AI in marketing is not yet ripe for automation the way it is in programming, and the past year has been an awakening from the illusion that marketing could be fully automated
Coding copilots automate robotic work developers hated (hunting libraries, copy-pasting from open source), whereas marketing output is consumed by human eyes and ears and requires deep understanding of human behavior
47:58 Marketing lags far behind coding in ai automation readiness
Your assistant can query this graph directly — 5 positions here, 19,646 across the corpus. Add 996.fm over MCP.