Quick Summary
- Market volatility is making investors more cautious and more interested in AI. Many households are prioritizing liquidity, emergency savings, and financial resilience over maximizing returns, while some are beginning to use AI for investment guidance.
- AI is improving, but it still has not proved it can consistently beat the market. Broader academic research found that AI strategies performed poorly across different market environments, often becoming overly cautious in rising markets and overly aggressive in downturns.
- Investors should be skeptical of strategies that look perfect in hindsight. Back-tested models can appear highly successful because their rules were designed using past market data. Once those strategies must operate in real time, unforeseen events can quickly undermine their apparent advantage.
- Successful retirement planning requires far more than picking investments. Asset allocation, rebalancing, tax planning, withdrawal strategies, Social Security, Medicare, estate planning, insurance, and preventing emotional decisions can matter more than finding the next winning stock. AI will increasingly help advisors perform this work, but it is not yet capable of replacing human judgment, coordination, accountability, and behavioral coaching.
It is no wonder investors are feeling uneasy. With wars, tariffs, high prices, affordability issues, Fed decisions, and mid-term elections all in the news, there is plenty to worry about. To cope, some people are turning to artificial intelligence (AI), especially large language model (LLM) chatbots, for advice and stock ideas.
PricewaterhouseCoopers’ (PwC) 2026 Market Volatility Survey found that 84 percent of consumers are managing their finances more cautiously, 76 percent are paying more attention to liquidity and emergency preparedness, and 83 percent now rank financial resilience ahead of squeezing out higher long-term returns. Only four in 10 said they felt very or extremely confident making financial decisions during volatile markets.
That last statistic rings true. When markets get rough, even confident investors often become more cautious. To manage their concerns, 24 percent said they had used or considered using AI-powered tools for financial decisions. That is less than the 50 percent who use online research and financial news or the 48 percent who seek professional advice. But AI still has ground to gain: Only 53 percent said they trust AI tools during volatile times, and 87 percent still prefer human guidance as AI improves.
Last year, I said that AI would eventually make it unnecessary for households to pick their own stocks. I also said that most people aren’t good at picking stocks. I still believe both are true.
Hendrik Bessembinder’s research explains why picking stocks is so tough: Most of the wealth in the stock market comes from just a small number of companies, while more than half of the stocks in his U.S. study lost money over time. Dalbar’s research shows that investors make things worse by buying and selling at the wrong times. We struggle to find the winners and don’t have the patience for the rest.
Large language models, such as those behind ChatGPT and Claude, power many AI tools. AI agents go a step further. They don’t just answer questions; they can be given a goal, collect information, use other tools, compare choices, and follow steps to make decisions. Researchers are already combining specialized agents, memory systems, financial news, company filings, and market data to try to automate investment decisions.
That sounds like an investor. It is not yet a particularly good one.
A recent academic study, “Can LLM-based Financial Investing Strategies Outperform the Market in the Long Run?” examined AI investment strategies over 20 years and more than 100 stocks. Earlier studies made AI look strong by focusing on short periods and a few popular winners. But when researchers included more companies, even those no longer trading, and tested different market conditions, the AI’s advantage mostly disappeared. The models were overly cautious in bull markets and overly aggressive in bear markets. That is a remarkably human way to lose money.
JPMorgan’s promising test
Then JPMorgan made things more interesting by using AI agents to backtest an investment methodology.
The bank built eight AI agents to shift money between stocks and bonds as the economy changed. The agents grouped the market into four types: Goldilocks, reflation, stagflation, and risk-off. These are simple ways to describe different combinations of growth and inflation.
In simulations spanning about 20 years, all eight agents outperformed a traditional portfolio of 60 percent stocks and 40 percent bonds when risk adjusted. This means they were judged not just by returns, but also by how much their results changed over time. The best agent beat the 60/40 portfolio by 0.7 percentage points per year, with fewer ups and downs.
To be clear, JPMorgan did not show that AI can consistently pick individual stocks better than the market. The agents were making broad choices about how much to invest in stocks or bonds, all within a system designed by experts. And again, the result came from a back test.
Back tests are great at predicting the past.
A back test asks, “How would this strategy have worked if we used it in the past?” It is a reasonable question, but it is also one that has cost investors money.
I spent much of my career in the financial newsletter business. Publishers would create content and offer “model portfolios” for subscribers to follow. New publishers often showed up with a back-tested model that had delivered amazing past returns. Their marketing would show how $10,000 could have grown to $100,000, or even $1 million.
The line on the chart rose to the right. It always went up and to the right.
Then, after that compelled investors to subscribe, the publisher would begin issuing recommendations in real time. Some models failed quickly. Others lasted longer, allowing publishers to keep collecting subscription fees for a while. The problem wasn’t fraud. It was overfitting. Publishers kept testing different rules, indicators, time frames, and combinations until they found one that seemed likely to make money in the future. But, unfortunately, investing doesn’t work that way.
That doesn’t mean JPMorgan’s work is useless. It is a promising experiment, but just not a finished product.
I believe that LLMs and their agents will eventually outperform the market by a wide margin. They will be able to read more filings, earnings calls, economic reports, news stories, legal documents, Reddit feeds, and industry data than any human team could manage. Specialized agents will debate, test ideas, watch for risks, and learn from experience.
But I do not expect the most meaningful benefits to land in the brokerage accounts of do-it-yourself households.
The biggest benefits will likely go to the largest firm. The largest firms will have the best private data, the fastest computers, the lowest trading costs, and the largest research teams. They will also have access to what you and everyone else are doing. These firms will also learn from public information, market trends, and social media. The biggest players watch what millions of households ask for, what they buy, when they panic, and how those choices move prices. AI is going to juice investment gains, but probably not the returns of the individual investors—do-it-yourself investors will just be a valuable dataset.
Beating the market is not a retirement plan.
But the good news is that beating the market is not the same as building a safe retirement. I mean, yes, it helps. But it’s not everything. Picking stocks is just one part of a household’s financial picture, and it might not even be the most important part.
Vanguard’s Advisor’s Alpha research says the real value of advice comes from things like picking the right asset mix, rebalancing, keeping costs down, coaching people through fear and greed, putting investments in the right accounts, making tax-smart retirement withdrawals, and harvesting tax losses. Vanguard estimates that doing these things can add up to, or even beat, an additional three percent in net returns in some cases, though the actual benefit depends on the household and the advisor’s tactics.
AI is improving at picking investments. But the LLMs that most people use are still a long way from handling the rest of the job with good judgment, accountability, and follow-through.
An LLM can explain what a Roth conversion is. But it cannot yet reliably help coordinate that with Social Security, required withdrawals, Medicare premiums, capital gains, charitable giving, estate documents, spending needs, or the chance that one spouse might live another 25 years.
An LLM can remind you to rebalance. But it cannot take the place of someone sitting with you when the market drops 25 percent, helping you avoid selling the assets your retirement plan needs for the next 30 years. Vanguard says that behavioral coaching might be the most valuable thing an advisor offers, especially when markets are tough.
An LLM can sum up an estate plan. But it cannot yet reliably spot when a beneficiary form doesn’t match the will, bring together the attorney and tax expert, organize paperwork, help the family make tough choices, or make sure everything gets done.
AI can estimate how much you might spend. But it is not ready to help someone who has saved for 40 years to learn to spend without guilt or fear. It also cannot guide a new retiree through the loss of identity, routine, and social connections after leaving work. These non-financial parts of retirement are real, and they often need as much attention as the investments.
Investing is part of the work, not the whole job. The job includes protecting assets from unnecessary taxes, excessive risk, fraud, and unforeseen life events. It includes turning a portfolio into a paycheck after employment income stops. It includes Social Security and Medicare decisions, insurance reviews, tax planning, estate planning, trustee coordination, charitable giving, business transitions, and helping heirs understand what their parents built and why.
It also means handling paperwork, creating documents, and ensuring the plan is actually implemented. A great recommendation that sits in an unread email is worth nothing.
I believe AI will become a great tool for advisors and families. It will make research faster, planning more accurate, help track tax strategies, and make routine tasks less expensive. It might even become the world’s best investor one day. But the real advantage, or as Vanguard calls it, the real “alpha,” is all the other things stock pickers have to handle. Sooner or later, AI will probably do those things too. It’s just not there yet.
