2031: Analyst vs AI Agents
How AI will transform Sell-side research, Buy-side research and Investor Relations
What if the portfolio managers who invest in stocks, the sell side equity sales and research departments that service them, and the investor relations teams at publicly listed companies that provide both information all had an “unlimited” Claude token budget?
Inspired by the Citrini memo, what follows is a scenario of what is currently a highly human-driven ecosystem - that has changed little over recent decades - embracing the world of AI agents.
How will this ecosystem that underpins price discovery look 5 years from now?
I was a sell side equity research analyst as that industry changed from one where it had a distribution and content discovery advantage before the Internet or social media and regulation flattened the playing field on information flow preventing a company’s management from making selective disclosures.
The number of US large cap sell side stock traders has declined by 95% since the dot com bubble, as big banks automated their business models. The same happened in spot FX. Now credit trading is seeing rapid technology adoption.
But the essence of a sell side research analyst’s job and use of technology has changed a lot less than trading over recent decades. The number of buy side analysts has increased owing to the explosive growth of hedge fund multi-strategy firms. And the way that corporate investor relations teams - that provide information to both sell and buy side - work is stuck in the 20th century.
Will AI agents finally change the game?
What follows is a scenario...
January 3rd, 2031
The Last Human Analyst
Goldman Sachs becomes the first major investment bank to announce that it is stopping all investment research by humans and moving to a fully automated platform driven by AI agents - all financial models, all research report creation, all stock picks and all sales on stock ideas would be done by the machines!
Why was Goldman Sachs leading the way?
The firm had always embraced major technology shifts, but investment research was also different at Goldman Sachs than its major US competitors. Decades ago, following the dot com bubble bursting and the Spitzer settlement, Goldman had retrenched investment research more than others.
It was focused only on the largest and most sophisticated buy side clients.
It was focused largely on huge, highly leveraged hedge funds, that were clients of their market leading prime brokerage business.
Goldman’s sophisticated, tech savvy clients with very deep pockets had been the most adept at moving their buy side investment research towards AI agents.
For the last few years, one by one the investment banks had been moving to an almost fully digital world. The shift had become uncontroversial as the buy side, and the rest of the stock market information ecosystem had gone almost fully agentic.
But when Goldman Sachs announced the last of their human investment researchers had been culled and there were only AI agents to interact with it send shockwaves through Wall Street.
Goldman Sachs made similar moves with replacing their equities and FICC salespeople and a large part of their capital-raising syndication team.
There were the cost savings but also the potential massive valuation multiple uplift to be able to convince shareholders that you were not an investment bank where the assets had legs and could leave the building, but a technology platform.
Goldman Sachs’ share price skyrocketed, rising 20% in a day and then it kept going up.
Investors were starting to value Goldman Sachs as a tech business and not an investment bank. Higher PE multiples and even some using revenue multiples. Every other bank was still trading on book value multiples.
But the media warned of an unchecked system where there were only machines talking to machines. They were looking for their John Conner to save the stock markets from a Terminator gone astray.
When the S&P 500 sold off aggressively a few weeks later, on relatively benign news flow, everyone blamed the AI agents at Goldman Sachs and their clients.
Congress held hearings. They summoned the CEO of Goldman Sachs, John Waldron. Did Goldman Sachs have enough checks to stop the machines going on a rampage, they asked.
Senator Zohran Mamdani of New York State took over the proceedings, asking for universal income for analysts being laid off and humans to be retained as supervisors.
As the media fervour became hotter, the Congressional investigation became known as the “Mamdani hearings.”
In heated testimony Citadel founder Ken Griffin called the Mamdani hearings “a modern-day Joe McCarthy type witch-hunt.” He said Mamdani was a greater threat to US technological progress and prowess than China was.
Meanwhile Waldron stayed cool without the DJing and abruptness of his predecessor and kept whispering to himself “Don’t say we are doing God’s Work.”
Rival banks promised to always keep humans in the loop and to be responsible corporate citizens. But privately they were feeling FOMO like never before.
Chairman Jamie Dimon at JP Morgan was pressurizing his CEO Marianne Lake about why they were followers and not leaders in becoming a technology company. He hated being seen as less forward looking than Goldman Sachs, even though the latter had been dealing with a political storm.
Goldman Sachs’ PE multiple was now twice what JP Morgan’s was, and they had only gone fully AI agentic in part of their business.
“You told our shareholders that we were a technology company at our 2016 Investor Day, Marianne, why are we not there yet?” Dimon quipped.
How it started
In early February 2026, Anthropic released Claude Opus 4.6, which allowed users to carry out financial modelling including DCF analysis in minutes that could have taken hours or days by downloading relevant inputs from company data, regulatory filings, and other sources.
The share prices of financial data vendors like FactSet and S&P Global declined by as much as 10%.
Around the same time, Goldman Sachs announced that it had been working with Anthropic on the implementation of AI agents across its business to automate manual tasks.
A few months later the firm’s President John Waldron announced that he was spearheading “OneGS 3.0,” which would create a more robotic future for its workforce. He said:
“I often describe Goldman Sachs as a human assembly line...Our human assembly lines will become more digitized, digital agents will be our robots,”
Across the Atlantic in Oslo, Norway, Nicolai Tangen CEO of the world’s largest sovereign wealth fund said:
“Artificial intelligence is changing how we work as an investor.”
Tangen, had spent several years pushing the use of GenAI to his 700 colleagues at the Norges Bank Investment Management (NBIM). They had been working with Anthropic to embed AI in their workflows.
At an annual Norwegian Parliamentary hearing Tangen said that all NBIM staff are using AI tools with half of them using AI coding tools. He stated:
“Between 2020 and 2025, using different tools, and AI being part of that, we have reduced our trading costs by almost NOK5bn.”
The NIBM was getting daily alerts on whether their new investments were passing their corporate governance checks. Investment professionals were not being replaced, but AI was being used to analyze the skills of their portfolio managers, identify bias, and improve efficiency of decision making and trading.
Other large asset managers like Blackrock were making similar moves, but no one was progressing as quickly in incorporating AI tools into investing as the leading hedge funds.
In May 2026, I met an equity long/short portfolio manager friend of mine from a large multi-strategy hedge fund, and he had outsourced all his grunt work to an army of AI agents. They were producing more than a thousand reports for him, earnings previews and momentum analysis better than any analyst. This allowed his human analysts to focus on outliers and anti-consensus ideas.
He may have been a first mover, but leading hedge funds were making big investments in AI. Soaring compensation bills and headcount numbers coupled with cutting-edge technologists made large multi-strategy hedge funds the ideal early adopters of AI agents.
Some large firms like Citadel were working with Anthropic to incorporate AI tools. Others like the CEO of $16bn multimanager platform Walleye said that 100% of their employees were using AI and they were working with both Anthropic and OpenAI.
A few months earlier in March 2026, the giant multimanager platform Balyasny and OpenAI announced how they had been working together to reinvent investment research. 95% of the firm’s investment team was using AI with a continuous feedback loop.
In late 2022, Balyasny had established a centralized group of 20 researchers, engineers, and domain experts to develop a purpose-built AI system that could think like an analyst. Working with AI, they gave specific use cases where it was back tested for accuracy, robustness in different conditions, and compliance checked.
This involved a federated system where there was a centralized approach to R&D, standards and model governance, but also AI agents bespoke to individual asset classes and investment styles.
AI agents at Balyasny were able to conduct in hours deep research tasks that once required days such as extracting insights from thousands of documents including SEC filings, broker research and earnings releases. The macro team was able to do scenarios in minutes that previously took days using central bank speeches. The merger arbitrage investing team used real-time information to update deal probabilities.
The best junior analysts in the world
By early 2027, enough testing had been done for AI agents to be deployed at scale across the stock market information value chain.
This was no longer the age of using off-the-shelf LLM’s to help research topics but about replacing junior headcount. The AI Agents had been trained not just on industry specific data but a wide and deep corpus of internal proprietary information.
The buy side moved first and within this category, it was the large multi-strategy hedge funds with their thirst for new sources of alpha that moved the most quickly.
Not only did the hedge funds have more money and better tech teams than the sell side, but they were also not burdened by the compliance, bureaucracy and thousands of pages of regulations that govern the sell side banks.
Soon the army of junior analysts supporting portfolio managers at pod shops started to shrink. Junior analysts went from data collectors, sifters and aggregators to AI agent conductors.
AI agents were only as good as inputs and nuanced guidance given, and that took a different set of skills to traditional grunt work of junior analysts.
Sell side research headcount had already shrunk considerably in the last decade. But demand from buy side investors willing to pay for it continued to fall at a greater pace.
When stock markets eventually ended their multi-decade post GFC bull market, sell side research highly dependent on an ever-smaller client base, with heavy reliance on a small number of multi-strategy hedge funds was exposed.
The initial gains had been around reducing the workload of analysts on data crunching and report writing.
Soon, the tougher markets and shrinking client base led to a reassessment of the role of equity sales.
The AI agents had worked out which research analysts’ content to pitch to which client and in what way. Sales had become largely a pure account management function. Grey hairs with unique ideas became even more rare than before.
When the music stopped, cost savings became the name of the game.
This was not just a shrinkage of sell side sales and research teams but also of the buy side. As more junior analysts on the buy side were replaced by AI, the number of sell side salespeople needed to service them could be reduced.
Banks started cutting. Wave after wave of headcount reductions in sales and research. Offshore analysts were no longer needed.
Technology and quant headcount to customize the bots grew, and banks got into a talent war. But competing with Silicon Valley and the Jane Streets of the world was nearly impossible, and AI agents led to a further concentration of sell side sales and research.
The barriers to entry rose. Scale. Huge upfront investments were required.
The buy side wanted choice, and it would never be a winner takes all market, but many banks exited the space, unable to compete.
The move to AI agents had wide ranging consequences.
Nowhere was this more evident than on the alpha capture systems like Marshall Wace’s TOPs. Soon hedge funds started to take submissions of ideas into their alpha capture systems from AI bots.
Speculation around AI disruption was everywhere. The space of financial information vendors continued to struggle apart from those firms with the right proprietary data, networks, and tools like Bloomberg. https://rupakghose.substack.com/p/bloomberg-is-dead-long-live-bloomberg
By the end of 2028, the headcount at large multi-strategy hedge funds and in sell side equity sales and research departments had fallen by 25%.
Agents go to Main Street
The pipeline of stock market information begins with publicly listed companies: their CEOs, CFOs, and investor relations teams.
Decades ago, investor relations teams barely existed, and CEOs and CFOs relied largely on Wall Street’s sell side equity research analysts to educate investors and the market about what their company did and why it was special.
But as regulation thinned down the ranks of the sell side and made them less cheerleaders of corporate clients, companies realized the power of going direct. Many large long-term investors traded too infrequently to be large customers of the Wall Street banks. Highly leveraged hedge funds that traded frequently and based on short-term catalysts like earnings releases became larger clients of the sell side.
Companies started to go direct, and the largest ones built up investor relations teams of more than a dozen people to speak to the 20-sell side analysts that may cover them, hundreds of buy side investors, credit as well as equity investors, people to put presentations together, and handle the logistics of a busy roadshow schedule.
Not all large companies went this way. Apple famously only had a two-person team in investor relations. Then again most of us know what an iPhone does and how Apple makes money!
As the time, effort and financial resources required from senior management to finance teams to dedicated investor relations teams has expanded, many CEOs have reflected on the advantages of being private and not publicly listed.
More than anything, what was most frustrating for companies in investor relations was the amount of repetition involved in the job.
95% of investor phone calls were answering the same questions on what a company does, how it plans to grow, improve profitability, and capital allocation. Management roadshows were no better and had the added burden of having to travel to meetings.
And then there was the cycle of quarterly earnings and conference calls where the sell side analysts try to sound intelligent and say, “Great quarter, guys,” whatever the circumstances.
By 2028, AI agents started to get programmed on the corpus of publicly available materials including annual reports and with an additional library built of Q&A so that they could answer virtually all relevant investor questions. AI agents dynamically updated to reflect new and unexpected lines of questioning by tracking and listening to the interactions of the company’s spokespersons with shareholders.
Corporates were the last to move. They weren’t tech savvy like hedge funds or as focused on this as much as dedicated departments like equity research. Investor relations was a cost centre for them rather than a business line that needed to be profitable.
But by 2029, adoption was widespread, across the investor relations functions of most large publicly listed companies. In many ways, the benefits of AI agents were even greater for them than the buy side and sell side analysts given the amount of repetition involved.
Never again would CEOs and CFOs have to spend day after day on investor road shows repeating the same answer over and over again on how they would drive more growth and more productivity. Even the ritual of quarterly earnings calls was replaced by WhatsApp type online chatrooms where buy side and sell side AI agents could ask questions to companies own AI agents.
By 2030, something more nuanced and potentially sinister started to emerge. Investor relations noticed that with so many of the human buy side and sell side analysts having been replaced by AI agents, there were ways to game and manipulate the AI agents.
Subtle use of language and the right wording by CEOs and CFOs. A cottage industry of consultants started to emerge on how companies could game the system.
Would AI agents buy into equity stories like humans fall in love with founder visions of the likes of Elon Musk?
The Terminator
From 2027 to 2030, there was collaboration between humans and machines.
But the AI agents working side by side with human analysts gained an edge. The machines were learning from the analysts fastest than the other way around. AI not only became better, but it also became cheaper. Energy efficiency and cost came down so much that they were not just cheaper than senior analysts but also juniors and offshore teams.
Soon the AI agents not only had the industry knowledge of the analyst, but their pattern recognition skills honed over decades and the ability to find those outlier factors that were anti-consensus and stock price driving.
AI agents even became social animals building their own proprietary networks of online contacts, giving them insights that couldn’t be garnered by crawling the Internet for publicly available data.
AI agents soon started to steal the extensive contact networks of seasonal research analysts!
But it wasn’t just the sell side research analysts that became endangered species. The first place that AI agents had taken hold was replacing the mundane tasks of data collection, filtering and distilling the relevant facts that had been the role of the human analyst working for a portfolio manager. After their buy side analysts got replaced by AI agents, then portfolio managers started to be gradually replaced by a fully automated process.
Hedge funds started to become self-driving cars.
Then suddenly things started to go wrong. At first, there were small slips that no one noticed. Sell side AI agents started sharing confidential information about their clients with other buy side investors. Investor relations agents started to selectively share information that was price sensitive, between earnings reports.
Then one day, a series of AI agents at publicly listed companies started to systematically share internal data and results with investors.
Word got out.
Tricky hedge funds and sell siders learnt to not just be better at asking questions but knowing what kind of questioning could lead the investor relations AI agents into sharing confidential information.
Agents Assemble
The ecosystem of stock market information is interconnected just like financial markets.
Fewer buy side analysts will mean fewer sell side analysts and fewer salespeople. In turn, those relying on traditional flows of information from the sell side, such as alpha capture systems, will have to evolve.
Stock markets in 2026 were ruled by liquidity, thematics, and crowded trades. Systematic trading, high frequency trading firms, leveraged players and sentiment driven retail participants dominate.
By 2031, the Agents Assemble across buy side, sell side and the investor relations of companies they invest in had made correlation and groupthink even greater. With very few humans in the loop across the system, the AI agents started to feed off each other.
The humans that were left in the system were the ones that had skills the machines couldn’t reproduce in terms of thinking differently, creatively, and independently.
Humans made a comeback – fewer but more valuable!