Reading Morningstar’s 2026 Report: AI Is Rewriting How Active Funds Research, Decide, and Survive

Can AI manage a fund independently and outperform the market? That is often the first question in discussions of AI investing. Frontwise’s view is that the question can obscure the more important changes already under way.

Once AI continually influences what to research, which evidence to trust, when to revise a view, and how to allocate capital, it has already entered the core investment process. Who gives the final sign-off is only one part of that change.

This article examines Morningstar’s June 2026 report, AI in Active Fund Management: The State of Adoption in 2026. The report uses “active fund management” broadly, covering hedge funds, active equity, and systematic strategies.

Key takeaways

• Case: Morgan Stanley Counterpoint Global uses 56 AI agents in parallel to produce segment-revenue forecasts, margin forecasts, price targets, and a full research report in about an hour. Since January 2026, the team has also run an AI-managed shadow portfolio alongside a live human-managed portfolio.
• Speed: A rigorous fundamental analyst may cover only 20–30 issuers in depth. When AI handles initial screening and analysis, both research coverage and the pace of validation change.
• Limit: Robeco tested a GPT-like process for extracting return signals. The prototype looked impressive, but benchmark tests did not show incremental alpha; “rear-view mirror bias” remains a material risk.
• Division of labor: People set risk budgets, handle exceptions, and remain accountable to limited partners; analysts increasingly coordinate multiple research agents.
• Moat: Frontier models caught up with Man AHL’s internally developed tools within roughly 18 months. The durable assets are proprietary data, process records, and validation systems.
• Path: Choose one narrow use case, run it in shadow mode, and expand authority only as evidence accumulates.

---

1. An Efficiency Arms Race Is Taking Hold on Wall Street

Fifty-six AI agents work in parallel to produce segment-revenue forecasts, margin forecasts, price targets, and full research reports in about an hour—work that could previously occupy a team of analysts for weeks.

This is Morningstar’s account of Morgan Stanley’s Counterpoint Global. More experimentally, since January 2026 the team has run an AI-managed shadow portfolio alongside a human PM’s live portfolio. That creates a forward-looking comparison rather than relying only on historical backtests.

Capabilities have advanced sharply, but validation of investment alpha is still catching up. That tension defines the current state of adoption at leading asset managers.

2. What Counts as “Core”? Do Not Reduce It to Who Signs the Trade

It is easy to frame the debate as a binary question: can AI completely replace a human fund manager and consistently generate alpha?

That end-state framing misses the changes already taking place. Which companies a fund studies, which hypotheses it tests first, and when it adjusts a position in response to new evidence all shape the eventual capital allocation. If AI influences those decisions continuously, the investment process has already changed—even if a human PM still signs off.

The shift has two layers:

• Traditional machine learning has supported systematic equity platforms at firms such as BlackRock for years, dynamically adjusting security-level signals while investment managers calibrate principles and boundaries.
• Generative AI combines unstructured-data analysis, coding, hypothesis generation, tool use, and cross-checking. It is moving from a one-off analytical aid toward a system that participates continuously in research.

3. Faster Research Is Redrawing the Opportunity Set

Human research capacity is finite. Based on its manager-research experience, Morningstar notes that a rigorous fundamental analyst may cover only 20–30 issuers in depth. Many events, market segments, and peripheral opportunities therefore receive little attention.

When AI takes on initial screening, financial-statement analysis, earnings-call review, and comparable-company research, three things change at once:

• Wider coverage: Opportunities once ignored because research costs were too high become economical to investigate.
• Faster validation: A thesis previously revisited once a quarter can be checked continuously as new evidence arrives.
• Resource reallocation: Analysts spend less time on repetitive processing and more on ambiguous, contested, and high-judgment questions.

The deeper change is that the research feedback loop is being compressed. Morningstar cites Robeco as an example: reproducing an academic strategy that once took an intern weeks can now be tested initially in minutes through an agent-based workflow.

In active management, a core competitive capability will be the speed and quality of the full cycle: discovering opportunities, testing judgments, deploying strategies, and recognizing failure.

4. Alpha Is Not Yet Proven—and That Is the Next Frontier

Morningstar’s report does not avoid the failures.

Robeco, for example, tested a GPT-like process for generating earnings forecasts and extracting return signals. The prototype looked impressive, but benchmark tests did not show incremental alpha. Large models are also non-deterministic, and “rear-view mirror bias”—inflated backtest results caused by future information leaking into training data—remains a serious problem.

These limits do not make AI irrelevant. They make a new validation discipline necessary.

Counterpoint’s shadow portfolio is one attempt. Rather than relying on an attractive backtest, the team records what the AI saw in the live market, what it recommended, how its view differed from the human portfolio, and what return pattern followed.

How deeply AI can enter the investment process depends on how well the institution can verify its incremental contribution.

5. A New Division of Labor: People Own Boundaries and Exceptions; AI Accelerates Iteration

Investment judgment can be decomposed into pattern recognition, hypothesis generation, context retrieval, cross-validation, scenario comparison, and evidence weighting. More of the review process can also be automated: models can cross-check one another, identify missing assumptions, and surface contradictory evidence.

The human role does not disappear. It moves toward the highest-consequence work: major disagreements, unusual deviations, and decisions that exceed the system’s authority.

• Fund manager: Sets risk budgets, defines objectives, handles exceptions, and remains accountable to limited partners.
• Analyst: Becomes a coordinator of research agents, directing several systems working in parallel.

6. Tools Turn Over Quickly—So What Is the Durable Moat?

Man AHL offers a warning to institutions that mistake tools for durable advantage. The team developed Alpha Assistant and AlphaTrend internally, then moved rapidly to a Claude Code environment in 2026 and built more than 90 internal skills. Within roughly 18 months, frontier-model progress had overtaken capabilities that previously required bespoke software.

Tools change quickly. The organizational assets that can survive model cycles are different:

• proprietary data that has been cleaned and made usable;
• traceable research workflows and historical decision records;
• robust evaluation benchmarks, risk controls, and permission boundaries.

As general-purpose models make public information easier to process, the advantage from generic tooling will decay. Differentiation will depend on data infrastructure, better research questions, and more rigorous validation.

7. An Institutional Playbook: Start With One Real, Testable Workflow

Financial institutions should not attempt transformation on paper. Frontwise recommends three steps:

1. Choose a narrow workflow. Select a real and important investment process with measurable outcomes—for example, monitoring the failure conditions of a thesis across a set of holdings—and let AI participate continuously.
2. Run in shadow mode. Record AI’s real-time recommendations without executing them. Measure error rates, missed information, review cost, and any genuine improvement in judgment.
3. Expand authority gradually. Grant limited authority only as evidence and risk tolerance allow, while preserving monitoring, one-click pause, and rollback mechanisms.

Conclusion

Active managers do not need to wait for a fully autonomous AI fund manager before they begin.

When AI continually influences what is researched, which evidence is trusted, when a judgment changes, and how capital is allocated, it is already at the core of investing. The organizational and engineering work done today will determine whether that capability becomes a sustainable advantage in the next decade.

For the original report: Use the “Read the original” link at the end of the page. The cases and figures in this article should be checked against Morningstar’s report when cited elsewhere.

---

Source: Morningstar Manager Research, AI in Active Fund Management: The State of Adoption in 2026, June 8, 2026. The cases and figures come from the report; the analysis of how active managers may evolve and how institutions should implement AI is Frontwise’s interpretation. This article does not constitute investment advice or imply future performance.

Full report: See “Read the original” at the end of the article. Verify all figures against the source report before reuse.