On November 11, United Airlines announced restrictions on unauthorized third-party AI agents booking flights or managing customer accounts. The airline cited legitimate concerns about data security, unintended transactions, and regulatory compliance.
But United CEO Scott Kirby made another revealing statement, reported by The Wall Street Journal:
"To be honest, I don't want to create another middleman."
I believe that statement deserves particular attention from investors.
AI agents are beginning to position themselves between businesses and their customers. They could increasingly influence which companies consumers discover, which products they compare, which offers they receive, and ultimately where they spend their money.
For private equity and growth equity investors, this raises a fundamental question: Could AI agents undermine the competitive advantages, earnings assumptions, and customer relationships on which today's investment theses are built?
And there is a second, equally important concern. As enterprises delegate more decisions and business processes to external AI providers, are they creating dependencies that could compromise operational independence, cybersecurity, and ultimately enterprise value?
These developments deserve attention well beyond the technology department. They have direct implications for investment strategy, M&A due diligence, portfolio value creation, and exit readiness.
Investment Risk #1: Who Controls Access to the Customer?
Consider a portfolio company with a strong brand, loyal customers, and an established direct sales channel.
Its investment thesis may depend on customer retention, pricing power, recurring revenue, and efficient customer acquisition.
Now consider what happens when customers increasingly use AI agents to research, compare, negotiate, and purchase products.
Instead of visiting the company's website or interacting with its sales representatives, customers may simply ask an AI agent to identify the best available offer.
The agent could compare competing providers, evaluate alternatives, negotiate terms, and complete transactions - potentially without the customer ever visiting the company's digital channels.
For consumers, this could bring significant benefits: convenience, transparency, lower search costs, and greater negotiating power.
For businesses and investors, the consequences are more complicated.
Who determines which products the AI agent considers? Which providers appear in its recommendations? Could commercial arrangements influence those recommendations? And will companies retain direct access to their customers?
These are potential conflicts of interest, rather than evidence of inappropriate behavior by any particular AI provider. Nevertheless, they deserve scrutiny.
An AI agent controlled by a major technology company could become an influential new distribution channel. Businesses might find themselves competing for access to that channel, potentially paying new intermediary fees or losing visibility into their customer relationships.
For certain companies, this could affect customer acquisition costs, conversion rates, pricing power, retention, and contribution margins.
The implications extend beyond airlines. Insurance distribution, financial services, travel, e-commerce, and other businesses that rely on customer discovery and product comparison could face similar pressures.
The opportunity is equally significant. Companies that make their products accessible to authorized AI agents could benefit from new demand and more efficient distribution.
This is already moving into commercial infrastructure. Stripe's Agentic Commerce Suite enables businesses to sell through AI agents, creating new ways for customers to discover and purchase products.
The future commercial outcome is far from settled. But the investment implications are clear: the durability of customer relationships and distribution advantages deserves renewed scrutiny.
For investors, AI-driven changes in customer access could become as relevant to an investment thesis as competitive positioning, customer concentration, and pricing power.
Investment Risk #2: When AI Dependency Becomes Business Dependency
There is another, potentially less visible, exposure: Enterprises are increasingly exploring AI agents to automate financial processes, customer service, procurement, software development, research, and administrative activities.
The attraction is considerable. AI agents could reduce costs, improve productivity, increase operational capacity, and accelerate decision-making.
I personally believe these opportunities are real.
However, the financial and strategic implications depend on how companies implement and control the technology.
Consider an enterprise that increasingly relies on a third-party AI provider to execute critical operational processes.
What happens if that provider changes its pricing, restricts functionality, discontinues a service, suffers an outage, or materially changes its contractual conditions?
Can the enterprise continue operating? Can it replace the provider? Does it retain the technical capabilities, process knowledge, and data required to make that transition?
A business can become operationally dependent on an AI provider long before that dependency becomes obvious in its financial statements or traditional due diligence materials.
Three exposures deserve particular attention.
1) Vendor concentration and operational independence
AI agents can become deeply integrated into enterprise applications and business processes. Replacing them may require rebuilding integrations, reconfiguring workflows, retraining employees, or reconstructing internal knowledge.
The severity of this exposure depends on the architecture, contracts, operational controls, and availability of alternatives.
A company that can switch providers without significant disruption has a materially different risk profile from one whose critical processes depend on a single external service.
2) Recurring AI costs and margin exposure
AI services can introduce subscription commitments, consumption-based charges, transaction fees, integration expenditure, and ongoing support costs.
These costs need to be evaluated alongside the productivity gains and revenue opportunities they generate.
An AI implementation that produces measurable efficiency improvements may strengthen EBITDA and cash generation. An implementation with poorly controlled consumption, unexpected transaction fees, or expensive vendor commitments could weaken margins.
Investors should distinguish recurring operating expenditure from capitalizable implementation costs under the applicable accounting rules. The financial implications should be reflected accurately in quality-of-earnings analysis, forecasts, and value creation plans.
3) Intellectual property, data rights, and institutional capability
A company's competitive advantage may depend on proprietary data, operational know-how, customer intelligence, and specialized business processes.
External AI agents could process or access highly valuable information. Depending on the technical configuration and contractual terms, companies may face questions about retention, training, confidentiality, ownership of outputs, and portability of agent configurations.
These risks should not be overstated. For example, Anthropic states that inputs and outputs from its commercial products are not used for model training by default, subject to specified exceptions.
The relevant due diligence question is therefore not whether external AI automatically compromises intellectual property. It is whether the enterprise has verified its rights, understood where sensitive information flows, and preserved the knowledge required to operate independently.
The investment concern is the gradual loss of strategic flexibility when critical capabilities become expensive or difficult to replace.
Cybersecurity Risk: When Legitimate Access Becomes the Attack Path
AI dependency also introduces a security concern that investors should understand.
Unlike conventional AI assistants that primarily generate recommendations, autonomous AI agents may receive permission to access confidential information, interact with enterprise applications, modify records, or initiate transactions.
This creates a new category of operational exposure.
Through prompt injection, attackers can place malicious instructions in emails, documents, websites, or other information that an AI agent processes.
If an agent follows those instructions, an attacker may be able to exploit the agent's legitimate permissions without directly compromising the user's credentials.
Anthropic's published research acknowledges that prompt injection remains an unresolved security problem.
Consider an AI agent authorized to analyze customer contracts, review financial information, or manage payments.
A successful manipulation could potentially result in confidential information being disclosed, unauthorized transactions, altered business records, or decisions based on compromised information.
For investors, this connects cybersecurity directly to financial exposure, operational continuity, regulatory obligations, and the reliability of business processes.
The question becomes especially important when an AI agent can execute actions without human approval.
An AI-generated recommendation can be challenged before implementation. An autonomous agent may act before an employee discovers that something has gone wrong.
I would therefore expect enterprises to demonstrate clearly defined permissions, transaction limits, independent security testing, activity records, effective access revocation, and human approval for sensitive decisions.
These are established areas of concern in guidance such as the OWASP Top 10 for Agentic Applications.
For investors, cybersecurity due diligence should increasingly examine what AI agents are actually authorized to do - not merely whether the company has adopted an AI policy.
What Does This Mean for Enterprise Value?
The commercial implications of agentic AI can be translated into familiar investment metrics:
Revenue quality and growth: Could AI intermediaries weaken customer acquisition channels, customer loyalty, direct sales, or pricing power? Alternatively, could AI-enabled distribution accelerate growth?
Operating margins: Will automation savings exceed AI subscriptions, consumption charges, new intermediary fees, and ongoing operating costs?
Business resilience: Could an AI provider outage, security incident, or contractual change interrupt a critical operation?
Competitive advantage: Does the enterprise retain the proprietary capabilities, customer intelligence, and control over distribution that support its market position?
Exit value: Can a prospective buyer acquire and operate the business without unacceptable technology dependencies, licensing restrictions, or costly remediation?
A simple hypothetical example illustrates the financial relevance.
Suppose a portfolio company generates €100 million in annual revenue. If 20% of that revenue becomes subject to an additional 3% AI intermediary fee, the incremental annual cost would be €600,000.
Assuming no offsetting benefits, full EBITDA impact, and an unchanged 10× EBITDA valuation multiple, that represents €6 million of implied enterprise value.
These figures are illustrative rather than a prediction. Actual effects will depend on customer behavior, channel economics, pricing responses, and the company's ability to capture efficiency gains.
The example demonstrates why changes in distribution economics deserve attention in financial modeling.
For private equity investors, the implications are closely connected to EBITDA sustainability, operational improvement, cash conversion, supplier concentration, and exit readiness.
For growth equity investors, the exposure may be concentrated in customer acquisition economics, retention, scalability, gross margins, and the durability of product differentiation.
In either case, agentic AI could create substantial value. It could also shift economic benefits toward intermediaries or critical technology suppliers.
The investment question is who ultimately captures those benefits - and who carries the associated risks.
5 Questions Every Investment Committee Should Ask About AI
I believe AI agent risk should become an explicit consideration in commercial, technology, cybersecurity, and operational due diligence.
Five questions would be particularly relevant.
1. How exposed is the investment thesis to AI-controlled customer access?
Determine how customers discover, compare, and purchase the company's products. Evaluate reliance on search engines, marketplaces, distribution partners, and direct customer relationships.
Assess how agent-driven purchasing could affect conversion rates, acquisition costs, pricing power, and customer retention.
For companies with significant intermediary exposure, test alternative channel economics in the financial model.
2. Which critical processes depend on external AI providers?
Request an inventory of deployed AI agents, their suppliers, business owners, integrations, permissions, and supported processes.
Identify where a provider outage, service discontinuation, or contractual change could interrupt operations.
Evaluate replacement options, recovery arrangements, migration complexity, and the enterprise's ability to maintain essential capabilities.
3. Are AI costs and dependencies adequately reflected in the investment case?
Examine subscriptions, usage-based charges, contractual commitments, renewal provisions, transaction costs, and implementation expenditure.
Compare these obligations with verified productivity improvements and expected revenue benefits.
Determine whether the investment thesis remains attractive under alternative pricing, adoption, and vendor-dependency scenarios.
4. Can AI agents compromise sensitive information or execute unauthorized actions?
Determine which agents can access financial information, customer records, intellectual property, enterprise applications, or payment systems.
Review permission management, transaction limits, human approval requirements, security assessments, monitoring, and incident response capabilities.
For high-impact activities, verify that safeguards operate effectively rather than relying solely on policies or vendor assurances.
5. Can the business retain its value when the AI provider changes (or when the company is sold)?
Review intellectual property rights, data retention and portability, confidentiality provisions, termination conditions, and change-of-control clauses.
Assess whether licenses, integrations, agent configurations, and operational capabilities can be transferred or replaced without disproportionate cost or disruption.
A buyer should understand what it is acquiring, which capabilities remain under the company's control, and which depend on external providers.
These questions are consistent with the third-party risk concerns identified in NIST's AI risk management guidance.
I would incorporate the findings into the investment lifecycle: identify material exposures during pre-investment diligence, address priority gaps in the post-acquisition value creation plan, monitor critical dependencies throughout the holding period, and verify business transferability before exit.
AI Risk Can Also Become an Investment Opportunity
I do not believe investors should approach agentic AI primarily as a defensive concern.
The same technology could create attractive opportunities for businesses that use automation effectively, establish trusted transaction capabilities, or help other enterprises govern autonomous systems.
Security, identity and authorization, AI agent monitoring, payments, enterprise integration, and data management could all become important areas of investment activity.
Established portfolio companies may also gain competitive advantages by redesigning operations, improving productivity, and making their products accessible through trustworthy AI distribution channels.
The opportunity is substantial.
But investors need to distinguish lasting improvements in financial performance from efficiency claims that have not yet translated into measurable results.
A technology that reduces operating costs, improves customer outcomes, and preserves strategic flexibility can strengthen enterprise value.
A technology that increases dependency, weakens customer relationships, or shifts pricing power toward external providers deserves a different assessment.
Who Ultimately Holds the Strings?
United Airlines' decision highlights a commercial conflict that is likely to extend far beyond travel.
Businesses want to preserve customer relationships, control distribution, and protect their commercial independence.
AI providers want their agents to become preferred interfaces for accessing products, information, and services.
Consumers want convenience, competitive offers, security, and freedom of choice.
Investors want sustainable growth, dependable earnings, competitive advantages, and transferable enterprise value.
The interests of these parties will not always align.
An AI agent may appear to act on behalf of the customer or the enterprise. But who ultimately determines which information it receives, which options it considers, which actions it can take, and whose commercial interests influence those decisions?
That is why I believe investment professionals should pay greater attention to the control, governance, and replaceability of AI agents.
I would gladly use AI to compare offers, automate routine activities, or improve business performance. For sensitive transactions, proprietary information, and critical operating processes, I would expect explicit permissions, transparent commercial arrangements, effective security controls, and clear accountability.
Most importantly, I would expect the enterprise to retain the practical ability to change technology providers without losing control of its data, operations, customer relationships, or intellectual property.
The question for investors is no longer simply whether AI can create value. It is whether the business can capture and retain that value - and who actually controls the AI agents on which that value increasingly depends.
Before your next investment committee meeting, I would ask:
Are you evaluating how AI agents could affect your target company's revenue, margins, competitive advantage, and exit value - or are you still treating AI primarily as a technology and cybersecurity issue?
In 5 years from now on, this will become the most important question in technical due diligence.
This article was drafted with AI assistance and substantively reviewed, fact-checked, and edited by the author.



