Nassau Street Partners on AI and Private Capital

The fashionable debate about artificial intelligence in finance is usually framed as a contest of prediction. Which model can identify a market signal first? Which system can digest the largest amount of data? Which firm can turn an earnings call, filing or research report into a trade before the opportunity disappears?

At the 2026 Nassau Street Partners Family Office Summit in London, the more interesting question was about judgment. A senior software engineer at Google (NASDAQ:GOOGL) DeepMind, spoke to an audience of financial professionals and family-office representatives about the real uses and limitations of large language models. His examples suggested that AI is already capable of valuable analytical work. They also highlighted the role people may continue to play in evaluating which information is relevant to private-market decisions.

Danenberg noted that large language models are not well suited to high-frequency trading because latency is too high. That limitation is instructive. It is a reminder that the strongest use case is not always the one with the most dramatic headline. The technology may have greater value in tasks such as sentiment analysis, document review, compliance and the interrogation of investment assumptions.

An earnings call can be analysed for changes in tone or language. A large collection of compliance documents can be reviewed for potential conflicts or omissions. A portfolio team can use a model to answer routine client questions. An investment committee can ask the system to challenge a memorandum against established guidelines. None of these applications necessarily replaces the investor. Instead, they may allow investors to consider a broader range of information.

That distinction may be particularly relevant to family offices. Their influence in private markets is growing because they can provide patient, flexible capital and can often evaluate opportunities outside the constraints of a traditional fund. They may invest directly in a business, support a manager, participate in a private-credit transaction or build a portfolio around the priorities of a family rather than the timetable of outside limited partners.

However, that flexibility can also create additional responsibilities. A family office can encounter opportunities across industries, structures and jurisdictions. The documents may be inconsistent. The businesses may be difficult to compare. The most consequential risks may sit outside conventional financial metrics. A technology company can have attractive growth and weak governance. A credit opportunity can offer a high coupon while relying on fragile collateral assumptions. A direct investment can look compelling on paper but depend heavily on one founder.

AI can produce more observations about all of these situations. Human judgment can help investors interpret those observations and determine how, or whether, to act on them.

One attendee at the summit described using a model to review investment memoranda and challenge whether the analysis had addressed the investment committee’s requirements. The value was not that the system delivered a final answer. It was that it raised questions. This may be one of the most productive roles for AI in private capital: not replacing conviction, but testing it.

Danenberg offered a useful analogy from software engineering. Ask a model to inspect a large codebase and it may identify a thousand supposed bugs. Most may be fictitious or trivial. A handful may be worth investigating. The essential skill is therefore not generating the list. It is recognising the few items that could have real consequences.

Private-market investing works in much the same way. Diligence can generate an abundance of possible concerns. The challenge is to determine which one can impair value, delay an exit, weaken a covenant package or undermine the trust between an investor and management. That requires experience, context and what Danenberg called taste. It also requires the confidence to ignore noise.

The discussion became more philosophical when participants considered whether models could eventually develop something resembling human judgment. Current systems can perform the language of reasoning. They can present a balanced argument, imitate scepticism and explain a decision in polished terms. The unresolved issue is whether such performance is equivalent to understanding.

For family offices, there is no need to settle that philosophical question before taking action. The practical answer is to design systems in which AI expands analysis while people retain responsibility. A model can compare documents, summarize competing views and identify inconsistencies. It should not be allowed to create the illusion that a difficult investment has become simple.

This is also where Nassau Street Partners’ role at the summit was important. The firm did not present the event as a technology showcase divorced from the realities of capital allocation. It created a forum where the limits of the tools were discussed openly. That offers a more measured perspective amid broader claims about AI’s capabilities.

Some family offices may be able to adopt AI tools relatively quickly because their smaller structures can allow for more flexible workflows than those of larger institutions. They can test several models, build processes around their own documents and integrate the technology into the way their principals and investment teams already operate. One attendee described using multiple models and even employing another AI tool to decide which model should be used for the next quarter. The age of a single system doing everything may already be ending.

However, adopting the latest model alone may not provide a meaningful competitive advantage Frontier capabilities spread quickly, and any obvious trading edge may disappear once competitors gain access to the same tools. Potential longer-term advantages may instead depend on factors such as proprietary data, established processes and the human decisions surrounding the technology.

That conclusion may be particularly relevant to private markets. Family offices have traditionally relied on factors such as relationships, patience and individual investment judgment. AI can make those qualities more scalable, but it cannot manufacture them. Firms may benefit from using AI to broaden their research while retaining human oversight and accountability for investment decisions.

The strongest message from the Nassau Street Partners summit was therefore reassuring. As technology becomes more capable, human judgment continues to play a significant role in private-capital decisions. For family offices that can combine independence with institutional-quality analysis, that is not a threat. It is an opportunity.