Value Moves Into the Operating Layer: Morning Brief, May 13, 2026

This Morning Brief was published for May 13, 2026. It preserves the source trail behind the day's strongest signals and frames them for public strategy readers.

The day's useful pattern is that value is migrating below the visible surface. AI, cyber, finance, logistics, and allied defence are all becoming contests over operating layers: infrastructure, governance, data, distribution, response capacity, and capital discipline.

Executive Signals

• AI value is shifting from productivity to market structure: McKinsey and PitchBook both point away from tool adoption as the main question. The real issue is whether AI changes profit pools, cost curves, liquidity paths, and control points.

• Agent infrastructure is becoming the control surface: Interaction models, agent harness debt, AI gateways, and MCP gateways all show the same pattern: the useful frontier is less about prompts and more about runtime, tools, governance, and replaceable scaffolding.

• Cyber risk is moving from theoretical AI abuse to operational response design: The AP item on AI-assisted exploit development and the critical-infrastructure coalition both suggest a transition from warnings about AI risk to concrete questions about who coordinates, detects, and owns response.

• Capital is repricing AI through proof, not narrative: AI valuations, PE value creation, and wealthtech gatekeepers all point to a tougher funding environment where returns, distribution power, and operational capability matter more than headline growth.

• Large platforms are opening back-end capabilities as products: Amazon's supply-chain launch and U.S. Bank's AWS modernization show platform economics moving into logistics and financial infrastructure rather than staying in consumer-facing apps.

• Allied defence capacity is becoming a finance and governance problem: Poland's EU SAFE loan agreement matters less as a one-off procurement story than as evidence that European rearmament is becoming a capital-allocation, industrial-policy, and sovereignty question.

Anchor Articles

01. Where AI will create value and where it won't

Why it mattersIt reframes AI from a productivity tool into a competitive-economics problem.

ActionWatch for firms that can name the profit pool, control point, or transaction-cost shift AI is expected to change.

McKinsey argues that AI productivity gains are becoming table stakes rather than durable advantage. The article says that automating existing tasks can improve speed, accuracy, and cost, but those gains tend to diffuse across industries and are often passed to customers through competition. The stronger strategic question is where AI changes offerings, business models, and market structures.

The piece is useful because it moves past the usual adoption debate. It separates three waves: productivity, differentiation, and transaction-cost reduction. The third is the most disruptive because AI agents could reduce search, comparison, negotiation, switching, and coordination costs across markets that currently depend on friction.

That matters for executives because advantage may migrate to new control points: customer interfaces, proprietary data, trusted orchestration, ecosystem position, or the ability to coordinate specialized providers. The article's banking example is especially concrete: if AI agents make deposit switching easier, banks' low-yield deposit economics could come under pressure.

This became an anchor because several other newsletter items echo the same pattern. PitchBook sees AI valuation risk, TLDR sees agent commerce and AEO visibility, and McKinsey sees profit pools shifting. The larger signal is that AI strategy is no longer about whether a firm has tools; it is about whether it understands how AI will redistribute value.

02. Private equity: Clearer view, tougher terrain

Why it mattersIt shows private equity moving from cycle timing toward operational value creation under tighter conditions.

ActionTrack whether PE firms can underwrite AI, leadership, and operational change as part of the deal thesis rather than as post-close decoration.

McKinsey's private-equity report describes a market that is no longer hidden by the fog of the post-2021 slowdown. Dealmaking and exits improved in 2025, but the terrain is more technical. The newsletter framed this as a shift from waiting for conditions to improve toward building clearer positions on value creation, leadership quality, and operating capability.

The signal is not simply that private equity is back. It is that returns are harder to win with leverage, multiple expansion, or timing alone. Slower exits, more selective fundraising, and AI uncertainty in software assets make operational improvement and portfolio discipline more important to underwriting.

AI shows up here as both opportunity and risk. PE firms can use it to improve diligence, pricing, portfolio operations, and back-office leverage, but the same technology may compress margins or weaken traditional software moats in target companies. That makes the asset selection problem more demanding.

This became an anchor because it balances the AI-heavy day with a capital-allocation lens. The report connects to PitchBook's AI return math and wealthtech gatekeepers: private markets are not only absorbing AI as a tool, they are also being forced to reprices assets, distribution, and liquidity around it.

03. As AI valuations skyrocket, who's thinking about returns?

Why it mattersIt puts hard return math against the AI funding narrative.

ActionMonitor whether AI rounds are being priced against credible exit comparables or only against scarcity and narrative pressure.

PitchBook's lead item argues that AI startup valuations are pulling further away from non-AI peers before the exit market has proven it can support the implied returns. The newsletter cites a median Series A pre-money valuation of $78 million for AI startups in Q1, up 31 percent from 2025 and 84 percent above non-AI peers. For Series B, the AI valuation gap is described as even wider.

The article's value is its refusal to treat high valuations as self-validating. PitchBook connects today's AI pricing to the recent software IPO drought, where private valuations and public-market multiples diverged and many companies had to accept weaker step-ups to go public.

The DPI data matters because it shows the liquidity base is fragile. PitchBook says 2019 and 2020 venture vintages have five-year DPI of 0.13x and 0.12x, the lowest at that stage since before the global financial crisis. That means current AI bets are being made while earlier capital has not yet returned much cash.

This became an anchor because it is the capital-market counterweight to AI capability stories. Thinking Machines and agent infrastructure show technology acceleration, but PitchBook asks whether funds, LPs, and public markets will tolerate the prices needed to make those rounds work.

04. Barbarians meet their new gatekeepers: wealthtech middlemen

Why it mattersIt shows private-market access becoming an infrastructure and distribution-control problem.

ActionWatch the firms that sit between asset managers and advisers; their selection power may shape which alternative products scale.

PitchBook's wealthtech piece looks at the platforms that help registered investment advisers source, evaluate, and buy private-market products for individual clients. The article describes iCapital, CAIS, Subscribe, and similar platforms as hidden distribution infrastructure behind the push to open alternatives to private wealth.

The signal is that democratizing private markets does not remove gatekeepers. It creates new ones. Platforms decide which funds are available, support due diligence, provide education, handle administration, and build white-label portals that wealth managers use to present products to clients.

This matters because private credit and private equity are already under scrutiny when retail investors want liquidity faster than fund structures can provide it. If the wealth channel becomes a major growth path for alternatives, platform incentives, diligence standards, and product curation become systemic risk questions.

This became an anchor because it adds a non-AI operating-layer signal. The same daily theme appears in a different market: power moves to infrastructure that controls access, workflow, trust, and selection. In AI that may be gateways and agents; in private markets it may be wealthtech distribution platforms.

05. Interaction Models: A Scalable Approach to Human-AI Collaboration

Why it mattersIt shifts the AI interface story from turn-based prompting to real-time multimodal collaboration.

ActionTrack whether low-latency, multimodal interaction becomes a separate product category from autonomous agent work.

Thinking Machines Lab introduced interaction models as a research preview for real-time human-AI collaboration across audio, video, and text. The TLDR summary emphasized a multi-stream design trained for continuous exchange rather than the familiar pattern of waiting for a completed prompt and then producing a response.

The primary-source page matters because it describes interactivity as something built into the model rather than bolted on through external voice, video, or agent scaffolding. That is a meaningful architectural claim: the interface is not just faster chat, it is a different way of processing human collaboration.

The trend signal is that AI products may split into two major branches. One branch optimizes autonomous task execution; another optimizes live human collaboration where timing, interruptions, visual context, and shared attention matter. That would change product design in meetings, field work, education, care delivery, manufacturing, and creative tools.

This became an anchor because it was the strongest AI capability story in the newsletter pool. It was more substantive than routine model-release chatter and connects directly to broader infrastructure signals: if the interface becomes continuous and multimodal, governance, observability, safety, and user agency all need to be redesigned.

06. Hidden Technical Debt of AI Systems: Agent Harness

Why it mattersIt names a practical failure mode for agent systems: overbuilding brittle scaffolding around fast-changing models.

ActionAsk whether an agent system's harness can be replaced quickly as models improve, tools change, or workflows move.

Lee Han Chung's post argues that agent harnesses can become hidden technical debt. The harness is the surrounding prompts, workflow code, tool wrappers, memory rules, skill files, and orchestration logic that make a model behave like a useful agent. As models improve, today's harness may become obsolete or actively constraining.

The article is useful because it avoids both hype and dismissal. It accepts that agent scaffolding is necessary today, but asks how much of it is durable. A key test is what would happen if the model became meaningfully smarter next quarter; if the team would need a week to unwind its own harness, that harness has become debt.

The signal is that AI engineering is moving from prompt craft into lifecycle design. Durable systems need thin harnesses, explicit skills, evals, and replaceable components. Otherwise teams may lock themselves into the limitations of the model generation they started with.

This became an anchor because it complements the Thinking Machines story. Better interaction models and stronger agents will not automatically make current products better if the surrounding system is brittle. The operating advantage will belong to teams that can keep the harness legible, observable, and disposable.

07. Google says it disrupted an AI-driven effort to exploit a software bug

Why it mattersIt moves AI-enabled offensive security from speculation into a concrete exploitation case.

ActionWatch whether defenders shift investment from generic AI awareness to exploit-chain detection, patch velocity, and coordinated hardening.

AP reports that Google disrupted a criminal group's attempt to use artificial intelligence to exploit a previously unknown software vulnerability. TLDR framed the incident as AI being used to identify and exploit a zero-day flaw that could bypass two-factor authentication in a widely used system-administration tool.

The story matters because it is specific. AI-enabled cyber risk is often discussed as a future possibility or as generic automation. Here, the claim is tied to an attempted exploit path, a real target class, and a major defender's intervention.

The broader signal is that enterprise security risk may rise during a transition period. Attackers can use AI to accelerate reconnaissance, exploit development, and workflow automation, while defenders still depend on slow patching, fragmented asset inventories, and uneven logging.

This became an anchor because it raises the bar for AI security discussion. The key question is not whether AI can help attackers in the abstract. It is whether organizations can shorten exposure windows, coordinate vulnerability response, and harden the software layers that attackers will increasingly probe with AI assistance.

08. AI Gateways vs. MCP Gateways: What Security Teams Need to Know

Why it mattersIt separates model-traffic control from tool-traffic control, a distinction many agent security programs will need.

ActionMap where your agent traffic actually flows: model calls, tool calls, hosted agents, SaaS agents, and unmanaged developer assistants.

Noma Security's post explains that AI gateways and MCP gateways are different control points. AI gateways sit between applications or agents and model providers, handling routing, cost, observability, failover, and sometimes security plugins. MCP gateways sit between agents and tools, handling identity, authentication, access control, and approved tool reachability.

The important point is that neither layer sees the whole picture. AI gateways may see prompts, responses, and some tool-call text, but they do not govern tool-level access. MCP gateways can govern tools, but they do not necessarily understand the full session behavior or non-MCP actions.

This is a strong signal because agent security is becoming architectural. It cannot be solved by content filters alone. Security teams need to understand inference paths, tool registries, SaaS agents, developer assistants, cloud-hosted agent runtimes, and where visibility is lost.

This became an anchor because it turns agent risk into a practical operating model. Paired with the AP exploit story, it suggests the next cyber-security frontier is not just finding malicious prompts; it is correlating agent behavior across model calls, tools, identity, and runtime.

09. New cybersecurity industry coalition aims to lead US critical infrastructure protection

Why it mattersIt shows infrastructure operators trying to fill a coordination gap in national cyber resilience.

ActionTrack whether critical-infrastructure resilience shifts toward sector-led continuity planning when government coordination is uncertain.

Cybersecurity Dive reports that the Alliance for Critical Infrastructure is launching to improve how U.S. critical infrastructure prepares for major cybersecurity crises. The coalition includes major private operators and focuses on cross-sector coordination, shared dependencies, and continuity during severe events.

The article is important because it frames cyber resilience as a governance and coordination problem rather than only a technical one. Infrastructure sectors may have their own plans, but cascading failures require understanding how electricity, finance, telecom, transport, and other sectors depend on each other.

The context is a perceived retreat or weakening of federal coordination channels. The article describes private-sector actors stepping forward as agencies face budget cuts, staffing losses, and changes in public-private partnership mechanisms. That may be positive initiative, but it also raises questions about accountability, authority, and national-level decision making.

This became an anchor because it shows the same operating-layer shift outside AI. Resilience depends on who makes decisions on the bad day, what data is shared, and how cross-sector dependencies are understood before a crisis. That is a durable strategic signal.

10. U.S. Bank shifts critical apps to AWS for AI push

Why it mattersIt shows AI modernization forcing a bank to move core systems, not just adopt front-end tools.

ActionWatch whether financial institutions treat AI as a cloud and data architecture program rather than a productivity-tool rollout.

CIO Dive reports that U.S. Bank will undertake a multiyear modernization initiative with AWS, migrating hundreds of mission-critical applications and revamping payment-processing and wealth-management platforms. TLDR framed the move as a financial-services firm modernizing infrastructure for AI workloads.

The article matters because it grounds AI adoption in architecture. Banks cannot get much value from agents, real-time decisioning, fraud models, or personalized wealth platforms if the underlying data and application estate remains fragmented or slow to change.

The risk side is also material. Moving critical workloads changes concentration risk, vendor dependency, resilience requirements, regulatory expectations, and operational control. The AI promise has to be weighed against the governance burden of using hyperscale cloud for more of the bank's operating core.

This became an anchor because it connects McKinsey's AI value argument to a concrete enterprise action. The signal is not that another company signed a cloud deal; it is that the AI operating model is pulling legacy firms into deeper infrastructure modernization.

11. Amazon Opens Its Supply Chain Network to All Businesses

Why it mattersIt turns Amazon's internal logistics capability into a broader infrastructure product.

ActionWatch whether logistics advantage shifts from owning demand to renting a mature operating network.

Amazon launched Amazon Supply Chain Services for businesses of all sizes, offering access to its freight, warehousing, fulfillment, and delivery network. TLDR noted that brands including Procter & Gamble, 3M, and Lands' End are already using the service to move and deliver products faster.

The article matters because Amazon is packaging a hard-won operating capability as an external service. That changes the competitive frame: logistics, inventory placement, fulfillment speed, and delivery reliability become available to firms that could not build Amazon-scale infrastructure themselves.

The strategic question is whether Amazon becomes a neutral utility for commerce operations or a deeper dependency layer for brands. The service could reduce barriers for smaller firms, but it may also concentrate more supply-chain visibility and operational leverage inside Amazon's platform.

This became an anchor because it broadens the brief beyond AI while reinforcing the same pattern. Durable advantage is moving into operating systems, data, routing, automation, and infrastructure that other firms plug into. The visible storefront is less important than the network underneath it.

12. Poland becomes first nation to sign EU SAFE loans, expects billions for defense

Why it mattersIt turns European rearmament into a financing, industrial-policy, and sovereignty signal.

ActionMonitor whether EU defence financing translates into faster delivery capacity or becomes another political bottleneck.

Breaking Defense reports that Poland became the first country to sign the EU's Security Action for Europe instrument agreements, clearing the way for Warsaw to receive roughly 43.7 billion euros, or about $51.6 billion, in defence funding. The story followed the Breaking Defense Europe newsletter's broader signal that Europe is gearing up for significantly higher defence spending.

The article matters because it is not just a procurement item. SAFE is a financing mechanism that turns European defence modernization into long-horizon capital allocation, with political debates about sovereignty, industrial sourcing, and EU-level borrowing attached.

The industrial-policy angle is important. A related Breaking Defense source-page noted that SAFE funding includes categories such as artillery, air and missile defence, unmanned and anti-drone systems, ammunition, space resources, cybersecurity, and AI solutions. That makes the program a map of capability demand through 2030.

This became an anchor because it is the day's strongest allied defence signal. It has clear NATO and Canadian relevance without forcing a U.S. defence frame: European allies are trying to turn urgency into money, money into capacity, and capacity into sovereign resilience.

Related Links

• The AI paradox in Europe's consumer industries: More spending, elusive impact- Supported the McKinsey AI value theme by showing the gap between spending and measurable EBIT impact.

• The Inference Shift- Connected low-latency interaction models to inference hardware and memory-bandwidth economics.

• Foundation Model Scaling- Added technical context on scaling beyond pretraining into post-training and test-time compute.

• AI Used to Build a Working Zero-Day Exploit- Same AP source as the anchor; retained as a related signal for the offensive-AI risk cluster.

• AI tool poisoning exposes a major flaw in enterprise agent security- Expanded the agent-security cluster beyond gateways into runtime tool integrity.

• Yum Builds an AI Backbone Across 35,000 Restaurants- Showed AI modernization as data-model standardization across a large operating footprint.

• Critical Infrastructure Gets a New Cyber Coalition- Related to the ACI anchor and useful for follow-up on public-private resilience coordination.

• NATO needs policies, standards for sharing AI-enhanced geospatial intel- Connected allied AI advantage to governance and data standards rather than more tools alone.

• Latvian defense minister resigns, following lagging response to drone incursions- Illustrated how drone detection and alerting failures can become political accountability issues.

• Havelsan unveils Barkan 3 unmanned ground vehicle- Added a concrete autonomy and swarm-capability example from the defence-industrial cluster.

• PayPal and Google executives: AI agent commerce will rely on crypto payment infrastructure- Kept agentic commerce as a supporting signal without letting crypto dominate the issue.

• How to See Aging as a Leading Indicator to See Where Work Hides- Supported the product-delivery theme from TLDR Product without becoming a central anchor.