25 September 2026
Heard In AI

Robinhood's Vlad Tenev predicts more lawyers, not fewer, as OpenAI moves into legal work

On the Moonshots podcast, in an episode published September 19, 2026, Robinhood CEO Vlad Tenev predicted that there will be more lawyers and software engineers in ten years than today. He argued that clients value a trusted adviser more than the answers themselves. The discussion followed OpenAI's launches of ChatGPT for Financial Services and Astra for Law. OpenAI reports that Astra for Law was correct on 54.0% of 200 legal research questions, compared with 38.7% for Astra using ordinary web search. Computer scientist Alexander Wissner-Gross agreed that there would be far more legal work but said many future "lawyers" may not be human.

A briefing reports one development at a point in time. We may correct or clarify it later; a new development gets a new briefing. How our formats work

Based on Moonshots with Peter Diamandis, episode published 19 September 2026

Dave Blundin was in a late-night board meeting, negotiating an acquisition, when a complicated legal question came up. His side's lawyers, billing $2,500 an hour, were on the line and started answering. Blundin, founder and general partner of Link Ventures, typed the same question into Google's Gemini chatbot while they spoke. "I swear to god it was word for word the same," he said on the Moonshots podcast.

The episode was published on September 19, 2026. Blundin's story set up a disagreement among the hosts and their guest, Robinhood co-founder and CEO Vlad Tenev. If AI can match an expensive lawyer's answer, will people still need lawyers? Tenev expects that they will, and in greater numbers. His reason is that he doesn't think the answer was ever what clients valued most.

What OpenAI launched

Host Peter Diamandis, founder of XPRIZE and Singularity University, raised the subject by pointing to two OpenAI releases. In his words, "Astra is eating one profession per week."

On September 10, OpenAI announced ChatGPT for Financial Services, with Morgan Stanley and the investment bank Evercore as design partners. The product pairs OpenAI's Astra reasoning model with hosted financial data from providers including Daloopa, PitchBook, LSEG News and Crunchbase. It cites the specific tables and passages behind each figure. The workflows OpenAI describes run from company research and financial analysis to spreadsheets, research notes and pitchbooks (the slide decks bankers present to clients), built from a firm's own templates. The product also comes with controls that large firms need. These include permissions based on each employee's role, settings for how long data is kept, exportable audit logs, and separate workspaces that keep information apart where needed. OpenAI says it does not train on business data by default. The features OpenAI describes cover research and document work. They say nothing about investment performance.

On September 17, OpenAI announced Astra for Law. It combines GPT-6 Astra with instructions for legal analysis and a search index of more than 230 million URLs. The index covers U.S. court cases, statutes, regulations, court rules and administrative decisions. OpenAI tested it on 200 questions from Vals AI's private Legal Research Bench. A benchmark is a standard test used to compare AI systems. This one checks two things: whether a system finds the relevant legal sources, and whether its answer meets criteria written for each question. OpenAI reports 54.0% overall correctness for the legal version, compared with 38.7% for ordinary Astra using web search. Both ran at maximum reasoning effort. OpenAI also says the legal version found 24% more of the reference cases on case-law questions. On an audited set of passages, it retrieved up to 54% more relevant material at the same reasoning effort.

These are OpenAI's own results on legal research questions. They do not measure advising a client or negotiating a deal. Access was initially planned for selected law firms through ChatGPT and Codex, with access for developers through OpenAI's API to follow. The integrations OpenAI lists for firms include agreement analysis, due diligence on transactions and preparation for initial public offerings.

The case against the billable hour

Diamandis framed the stakes around junior lawyers. First-year associates, he said, bill $600 an hour for legal research, and he asked what the "half-life" of that work might be.

Blundin argued that a system this good at law is also good at many everyday money tasks. He named tax-loss harvesting (selling investments at a loss to cut the tax owed on gains), rebalancing a portfolio, and handling wills and trusts. When people talk about AI and trading, he said, they tend to picture fast quantitative trading. For him, the benefit also covers "all the life stuff" that becomes easier when an AI handles it. He called it "one of the great benefits actually for humanity," adding that it was "not great for the legal profession."

Tenev's hot take

Tenev disagreed with that last part. "My hot take is there will be more lawyers in 10 years than today and more software engineers," he said.

He explained his reasoning. The need for lawyers, he said, scales with the number of new businesses, and he expects an explosion in entrepreneurship. He also believes the most valuable thing a great lawyer offers is not the legal advice itself. He compared a good lawyer to a consigliere, a trusted counselor who helps a client think things through, and to a negotiator.

Tenev saw the same pattern in personal finance. Robo-advisers are automated investment services. Once they arrived, he said, they could harvest tax losses and rebalance portfolios "better than any human." Even so, he said, the market for human financial advisers grew much faster than the market for robo-advisers. In his view, people pay for someone "that you can fully delegate and trust" with every part of their financial life, not for the advice alone. The podcast did not cite figures for either market.

Blundin largely agreed. For someone living a life like Tenev's or Elon Musk's, who has an idea in the morning and wants to act on it that afternoon, he said the consigliere comparison fits, and the cost of that person is "such a rounding error." Everything will be AI-assisted, he said, but people probably won't want to cut a human they trust out of their lives. Over the next three or four years, he expects AI to make human advisers much more useful. Clients' requests are often obscure and tied to a particular will, a divorce or a liquidity event, such as a sudden large payout. Acting on them used to take advisers a long time because the details were not at hand. With direct access to a client's accounts and AI to act on them, Blundin said, "the feeling of value add from the financial advisor is up at least a factor of 10." He acknowledged that Wissner-Gross might disagree over a 10- or 20-year view.

The counterargument: lawyers that aren't human

Alexander Wissner-Gross, a computer scientist and founder of Reified, first said he agreed with Tenev. Then came his twist: "How many of those 10x lawyers in the future will be human?" He accepted that there will be "10x lawyers" but added, "I just think many of them won't be natural humans." Tenev called it "a good question."

Wissner-Gross also argued that replacement can come well after the technology is ready. The Stone Age, he said, did not end for a lack of stones. He pointed to the "paperless office" promised in the 1980s. Office paper use did eventually fall, but not as soon as personal computers arrived. He expects "the lawyerless office or the lawyerless company" to arrive the same way: "It just takes a little bit longer after the capability first comes online."

That leaves two views of what cheaper expertise does. Tenev and Blundin describe more companies and more demand for trusted people who use AI, with Blundin speaking specifically about the next few years. Wissner-Gross accepts that demand will grow but doubts that the trusted advisers will stay human, even if the change comes later than the technology.

Diamandis ended the segment by returning to Tenev's point about entrepreneurship. He urged listeners who are looking for a job to "stop looking and start building."

Share this article

Go to the original

Sources & further reading

  1. 01
  2. 02

Connected ideas and articles

From the conversation

Podcast episodes

Article history

Updates to this article

Tags

Box's two rules for software in the agent era: beat the generic agent, then let it in

On Sequoia's Training Data podcast, Box CEO Aaron Levie said any company sitting on customers' data now has two obligations: build an agent measurably better than an off-the-shelf one at its own workflows, and expose the same capabilities to outside assistants like Claude and ChatGPT. He described the tuned search-and-retrieval harness behind Box's agent, the evaluations that track model progress, and his bet that within five years roughly 90% of enterprise tokens will be spent on work nobody asked for directly.

· Updated 8 min read

Would she pay the real price? Zitron's test for AI adoption

On The Diary of a CEO, critic Ed Zitron praises a chatbot for reading a troubleshooting log and for helping fix his son's Minecraft mod, then argues that neither is worth a trillion dollars. The host counters with his fiancée's one-woman business and his chief of staff's inbox. The argument turns on tokens, subscription rate limits and who is paying the real bill.

· Updated 7 min read

GPT-6 Astra tops a simulated vending-business test, but Dave Blundin says founders still have to manage the work

On an episode of Moonshots with Peter Diamandis recorded September 16, 2026, the hosts discussed three tests of AI-built businesses. The first was a GrokBot company-building livestream that Peter Diamandis described. The second was Andon Labs' Vending-Bench 2, in which GPT-6 Astra averaged about $15,515 at the end of a simulated year that began with $500. The third was the Build with Gemini XPRIZE, which requires real users and revenue. Salim Ismail stressed that the vending result came from a simulation, not a real company. Dave Blundin argued that AI fails at big tasks until a human provides a framework and breaks the work into small pieces. In his view, that makes entrepreneurs more powerful but leaves them with the job of managing the work.

7 min read

Andrew McAfee says he was wrong about AI and jobs, and still expects worker shortages

On The Diary of a CEO, MIT research scientist Andrew McAfee said he was "dead flat wrong" about the job and wage pressure he expected AI to put on white-collar workers. He expects employers to still be short of workers ten years from now, not facing mass unemployment. The payroll study he cites shows a narrower change, but not simply slower growth. Its August 2026 revision finds that employment of 22- to 25-year-olds in highly AI-exposed jobs is 19% below the path of comparable workers, mainly because of weaker hiring. In absolute terms, their employment in the most exposed jobs fell about 11% between November 2022 and June 2026. The authors call this a descriptive pattern, not proof of cause. Anthropic's conditional 2030 model scenarios, raised by the host, put overall unemployment between 3.9% and 11.9%. One guest argued that steady AI gains can cross the point where AI beats humans at some jobs, but offered no unemployment forecast.

8 min read