GUY: Good morning, Ava. It is Tuesday, August 25, 2026, and today's Morning Signal has one big idea running through almost everything: AI is moving from a model contest into an operating-system and capital-structure contest.

AVA: Good morning, Guy. And today's evidence set is unusually useful. The written brief covers nine episodes from nine podcasts inside the prior twenty-four-hour window. Six have exact full transcripts and three rely on authoritative show notes, so when we get to those partial sources, we will keep the conclusions narrow.

GUY: Let's start with markets. On Thoughts on the Market, Morgan Stanley chief investment officer Mike Wilson argued that the post-COVID economy looks more like the postwar period than the long 1982-to-2020 disinflationary era. His regime has stronger nominal growth, persistent inflation, more policy reaction, and shorter economic and market cycles.

AVA: The source matters here: Thoughts on the Market's official transcript is complete. Wilson said the latest rate pressure reflects fiscal-supported nominal growth, not just debt and deficits. He also treated Treasury buybacks as a market-functioning tool, rather than quantitative easing or yield-curve control. But he noted that precious metals and crypto rising together suggests investors expect a bigger intervention if financial conditions tighten.

GUY: That is a subtle setup. Thoughts on the Market is not saying every risk asset benefits from fiscal support. Wilson's equity signal is a quality rotation toward high free cash flow, high gross margins, stable sales growth, and low capex-to-sales. He dates the change to the June peak in earnings-revision breadth, led by semiconductors, and prefers large-cap quality, AI adopters, and the S&P 500 over international peers.

AVA: Hold on though... Thoughts on the Market also identified oil as the tripwire. Wilson does not need crude to decline for his constructive view; he needs it to stop rising. A new Strait of Hormuz shock would raise input costs, yields, and bond volatility. His proposed hedge is energy equities, which makes sense only while they retain the cash-flow sensitivity to oil that the rest of the portfolio lacks.

GUY: Exactly. The portfolio message from Thoughts on the Market is overweight quality, but do not confuse quality with passive duration. A company whose valuation needs falling rates is a different exposure from a business already producing cash. The falsifier is straightforward: if leadership becomes stable and broad, oil stops pressuring inflation, and bond volatility falls, the need for that defensive quality-and-energy pairing weakens.

AVA: Monetary Matters gives us the private-market transmission mechanism. In its syndicated Other People's Money interview, High Yield Harry said private equity and private credit remain stuck with delayed exits, continuation vehicles, tougher fundraising, and rates that stayed higher for longer than 2021 and 2022 underwriting assumed.

GUY: And Monetary Matters provided concrete stress markers. High Yield Harry reported redemption requests in the teens at some vehicles against periodic gates of five percent. He thinks many retail investors misunderstood the liquidity and may not fully return. That means the issue is not only marks on existing loans; it is also whether the asset-gathering machine can keep recycling capital.

AVA: Monetary Matters also sharpened the software risk. High Yield Harry said software and tech-enabled services can represent twenty to thirty-five percent of some private-credit portfolios. Current cash revenue can look fine into 2027 even while the capital structure worsens underneath. Then 2027 through 2029 brings refinancings, amend-and-extend decisions, sponsor equity checks, and the real test of enterprise value.

GUY: Right. Monetary Matters says private lenders may have tighter documents and seniority than equity owners, but that is recovery protection, not immunity. If seat-based software loses pricing power while R and D and token spending rise, a lender can still discover that the old terminal value was doing more work than the current income statement revealed.

AVA: That is why the brief's underwriting checklist is so useful: map 2027-to-2029 maturities, seat-based pricing, software retention, R and D or token expense, and sponsor capacity. High Yield Harry also described a barbell where the largest capital providers have scale advantages and hands-on lower-middle-market investors have operational advantages. The vulnerable middle is undifferentiated asset gathering.

GUY: Then The Meb Faber Show supplies a process answer. Its authoritative show notes, not a full transcript, describe Jerry Parker's trend-following framework as hunting outliers, accepting many small losses so a few large winners can pay for them, and refusing to give up long-run return merely to manufacture crisis alpha.

AVA: The source limitation matters: The Meb Faber Show's notes do not support inventing a current position or parameter. But they do support the portfolio role. In a shorter-cycle world, trend can absorb repeated false starts while remaining exposed to a major move. Trading individual stocks expands the opportunity set beyond indices. It is persistent convex participation, not a promise to save one specific down day.

GUY: So the markets synthesis is quality cash flow for the core, energy as an oil hedge when the tripwire is active, and trend as a process allocation. Monetary Matters tells us where higher-for-longer rates are quietly accumulating damage, and Thoughts on the Market tells us why the public equity tape can still look constructive before those private structures clear.

AVA: Now to technology, where Goldman Sachs Exchanges had the most concrete institutional-AI discussion. In Goldman's full official transcript, Chris Churchman described Marquee as an internal system combining research, trading-floor commentary, expert-built market-data widgets, and pre-trade analytics. It interprets intent, retrieves internal evidence, calls auditable calculation tools, and grounds each sentence to a firm source or calculation.

GUY: Goldman Sachs Exchanges made the key distinction beautifully: a demo is judged at its best, while a financial product is judged at its worst. Churchman said a model cannot inherently distinguish fact from interpolation. Even a low error rate forces the user to re-check everything unless the architecture makes abstention and provenance first-class features.

AVA: Which means, according to Goldman Sachs Exchanges, the durable moat is not simply the model. It is identity, permissions, data lineage, observability, audit, and workflow redesign. Churchman described a development sequence from prompt engineering to context engineering, checking loops, secure environments, and finally mandate engineering: what an agent may do, under whose authority, and in whose name.

GUY: I like that because Goldman Sachs Exchanges turns an abstract AI debate into liability. Insurance and liability markets could gate autonomy. If a human or institution must stand behind an output, the winning system is not the one that sounds most confident. It is the one that can show the evidence, the calculation, the entitlement, and the limits of its authority.

AVA: Goldman Sachs Exchanges also warned against overbuilding scaffolds around temporary model limitations. Longer context, native tool use, and better reasoning can obsolete retrieval and orchestration work. Churchman's prescription was to build for the model likely available at launch and invest in what public models cannot learn: institutional entitlements, mandates, proprietary data relationships, and tacit workflows.

GUY: Invest Like the Best approaches the stack from the supply side. Its official notes describe Neil Movva's Sail Research thesis as a token factory for long-running agents. For asynchronous workloads, interactive latency matters less, while cost per token and total throughput matter more. The notes also cover scavenging underused chips and power and the threat that open models compress frontier labs' three-to-six-month lead.

AVA: But Invest Like the Best supplied authoritative notes rather than a full transcript, so we cannot validate unit economics, utilization, or the episode's contrarian NVIDIA conclusion. The investable question is narrower: does asynchronous agent work create a distinct low-cost inference market, or does every valuable task still pay for frontier speed? That can be tested through workload mix and inference spending.

GUY: The a16z Podcast adds another partial source. Its authoritative notes say Martin Casado and Steven Sinofsky debated whether AI's mathematics performance represents genuine reasoning or a higher abstraction tool, then asked what happens when capital and compute substitute for some scarce engineering talent.

AVA: The a16z Podcast's implication is two-sided. Lower cognitive-production cost can increase experimentation, but it can also make financing, compute access, and distribution more important. That may favor well-capitalized incumbents unless startups own unique workflows, proprietary data, or go-to-market channels. Again, the notes support the direction, not invented operating figures.

GUY: Monetary Matters then connects inference economics to employment and credit. High Yield Harry estimated that AI can already do ninety to ninety-five percent of the research, memo, diligence, and spreadsheet work he performed as a junior analyst. He expects perhaps ten to twenty percent fewer private-credit seats over time. Those are his experience-based estimates, not measured industry statistics.

AVA: Goldman Sachs Exchanges complicates that productivity story. Churchman warned that automating junior tasks can destroy the apprenticeship through which people learn tacit judgment. Monetary Matters sees AI-fluent juniors and mid-level professionals moving toward higher-value work; Goldman asks who develops senior judgment after the experienced cohort turns over. Both can be true for several years before the institutional cost appears.

GUY: So the right KPI is not headcount saved. From Goldman Sachs Exchanges and Monetary Matters together, I would want cycle time, error rates, abstention, human overrides, audit coverage, and evidence that junior development still exists. A firm can improve next quarter's margin and weaken the future decision-making bench at the same time.

AVA: The Vergecast provides the physical shadow of the AI buildout. Its full exact-match transcript said Amazon raised the Echo Dot from forty-nine dollars and ninety-nine cents to seventy-nine dollars and ninety-nine cents, the Fire TV Stick 4K Max by forty percent to eighty-four dollars and ninety-nine cents, and the base Kindle to one hundred forty-nine dollars and ninety-nine cents. Amazon attributed the increases to memory and storage costs.

GUY: The Vergecast also reported Apple's HomePod Mini moving from ninety-nine dollars to one hundred twenty-nine dollars and ninety-nine cents, while Google's comparable speaker held its price. This is anecdotal, not a complete channel check. But it is consistent with data-center demand tightening components used outside data centers. The earnings question is whether broader price increases produce lower unit demand, heavier promotion, or mix-down.

AVA: Now geopolitics and policy. On the All-In Podcast, OSTP Director Michael Kratsios defended the administration's Science: A New Golden Age agenda. The exact-match full transcript records his central claim that the debate should shift from whether the federal research budget grows to whether roughly two hundred billion dollars of annual science and technology spending produces high-value breakthroughs.

GUY: The All-In Podcast outlined proposed funding experiments: reviewer golden tickets for unconventional grants, durations ranging from months to several years instead of a default around eighteen months, and portable fellowships for roughly twenty-six hundred early-career researchers. The idea is to fund scientists across universities, focused research organizations, and other settings instead of automatically privileging one institution.

AVA: The All-In Podcast also carried aggressive administration targets: American boots on the moon and a nuclear reactor in space by 2028, initial moon-base elements by 2030, a scientifically relevant quantum computer by 2028, fusion by 2035, and a Genesis Mission intended to use AI to double U.S. scientific output.

GUY: Those All-In Podcast items are targets and speaker claims, not verified outcomes. Kratsios said the five federal research agencies spending more than three billion dollars each owe implementation plans within ninety days of the policy's publication. He cited NSF focused-research organizations and four-year industry-linked PhDs as early actions. The real catalysts are agency plans and congressional appropriations.

AVA: And the falsifier from the All-In Podcast discussion is concrete: appropriations fail to follow the priorities, agencies do not publish measurable experiments, or programs increase headline activity without improving research output or commercialization. Goldman Sachs Exchanges adds a governance test: faster AI-enabled research needs provenance, reproducibility, baselines, replication standards, and stopping rules, or it simply scales confident error.

GUY: The Indicator from Planet Money moves us to healthcare policy. Its full transcript distinguished traditional Medicare from the campaign label Medicare for All. Traditional Medicare can still include premiums, deductibles, supplemental coverage, and potentially uncapped twenty-percent coinsurance for physician-administered drugs.

AVA: On The Indicator, Michigan Senate candidate Abdul El-Sayed proposed eliminating premiums, copays, and deductibles for necessary care while making insurance portable across jobs. He argued that could reduce job lock and support entrepreneurship. But higher public revenue and reallocation of current spending would be required.

GUY: The Indicator also gave Northwestern economist Craig Garthwaite's counterargument. Moving more patients from private reimbursement to public rates would cut hospital and drug-company margins and likely reduce drug innovation. He argued that promising much lower spending with no research trade-off is inconsistent, while the administrative transition would hit hundreds of thousands of jobs in a sector approaching one-fifth of U.S. GDP.

AVA: El-Sayed responded on The Indicator that insurers are already automating administrative roles and that public governance is preferable to corporate AI making coverage decisions. The policy redistributes economics from insurers and high-price providers toward households and potentially employers, but the second-order effects run through taxes, provider capacity, drug research, and labor.

GUY: The Indicator cited a 2020 Congressional Budget Office estimate of roughly one-point-five to three trillion dollars in additional federal subsidies, depending on design. That range tells you why implementation details dominate the label. Part B with Mark Cuban arrives August 26, and the test is whether it offers a concrete reimbursement, pharmacy-benefit, and implementation model.

AVA: Let's connect the threads. Invest Like the Best says cheaper, high-throughput inference can support longer-running agents. The a16z Podcast says capital and compute can substitute for some engineering scarcity. Monetary Matters says automation threatens junior analytical work and software-heavy credit books. Goldman Sachs Exchanges says regulated adoption still requires provenance, permissions, and liability.

GUY: That produces the central chain: cheaper tokens enable more agents; more autonomous work can reduce paid seats and switching barriers in some applications; weaker software terminal values make 2027-to-2029 refinancings harder. The winners are scarce infrastructure and governed workflow systems. The vulnerable group is leveraged application software whose moat is labor-intensive usage rather than proprietary process or data.

AVA: A second chain links Thoughts on the Market, Monetary Matters, and The Meb Faber Show. Fiscal-supported nominal growth keeps rates firmer. Another oil shock could add inflation and bond volatility. Higher-for-longer financing delays exits and tests 2021-to-2022 leverage. Quality cash flow and tactical energy hedges address the regime, while trend following addresses rapid leadership changes without pretending to forecast every turn.

GUY: A third chain links the All-In Podcast to Goldman Sachs Exchanges. The government wants AI to accelerate scientific output, but institutional AI requires auditability and explicit mandates. Agency implementation plans should therefore contain measurable baselines, replication standards, and stopping rules. Spending totals alone do not show that the program works.

AVA: And the fourth chain links Goldman Sachs Exchanges to Monetary Matters: automation can erase the training process institutions need most. If juniors stop doing research, models, and diligence, the organization must deliberately recreate reasoning practice and transmit tacit knowledge. The durable metric is decision quality after senior people turn over, not this year's headcount reduction.

GUY: So what are we watching? From Thoughts on the Market, crude acceleration, yields, bond volatility, and whether earnings-revision leadership broadens beyond the June semiconductor peak. The constructive quality view weakens if crude stabilizes, leadership becomes broad, and the shorter-cycle rotation stops producing dispersion.

AVA: From Monetary Matters, the next private-credit redemption windows: requests versus five-percent gates, publicly traded BDC discounts, fair-value marks, and fundraising. Then the 2027-to-2029 maturity wall: leveraged-software refinancing terms, amend-and-extend structures, sponsor equity contributions, and lender recoveries.

GUY: From Goldman Sachs Exchanges, Invest Like the Best, and the a16z Podcast, watch inference spending, workload duration, software seat counts, net retention, pricing, R and D or token expense, audit coverage, and human overrides over the third- and fourth-quarter 2026 reporting cycles. Those are the earliest tests of the AI-to-credit-risk chain.

AVA: From the All-In Podcast, look by roughly late November 2026 for implementation plans from the five agencies with more than three billion dollars of research spending. Then treat the 2028 moon-landing, space-reactor, and quantum-computing objectives as milestones only when budgets, contracts, and technical progress verify them. Fusion's stated target is 2035, and initial moon-base elements are targeted for 2030.

GUY: From The Indicator, tomorrow's Mark Cuban installment should tell us whether Medicare for All can move from aspiration and cost arguments into reimbursement mechanics. And from The Vergecast, watch whether memory and storage inflation broadens across consumer electronics or normalizes before it changes replacement cycles.

AVA: My final takeaway is that AI capability is no longer the scarce claim by itself. Goldman Sachs Exchanges says the institutional constraint is trust and mandate. Invest Like the Best says the supply constraint may become cheap throughput. Monetary Matters says the financial constraint is leverage. The All-In Podcast says the public-policy constraint is measurable execution.

GUY: And my final portfolio takeaway is equally simple: own the bottlenecks you can measure. Power, compute utilization, inference cost, trusted data, security, governed workflows, quality cash flow, and carefully chosen energy exposure all have observable tests. Be skeptical of legacy software multiples and private structures that require both stable seats and cheap refinancing.

AVA: That is the signal for Tuesday, August 25. Nine episodes, clear source boundaries, and a lot to verify over the next two reporting cycles.

GUY: Thanks for listening. We will be back tomorrow with the evidence, the catalysts, and what would change the view.