https://thomasjankowski.com Operator-tier writing on AI, healthcare, and travel. en-US Fri, 24 Apr 2026 00:00:00 https://thomasjankowski.com/writing/routing-the-healthcare-problem-we-should-have-solved-first https://thomasjankowski.com/writing/routing-the-healthcare-problem-we-should-have-solved-first Fri, 24 Apr 2026 00:00:00 https://thomasjankowski.com/writing/ai-design-partner-org-chart-collapse https://thomasjankowski.com/writing/ai-design-partner-org-chart-collapse Wed, 22 Apr 2026 14:13:00 -0400 Claude Opus 4.7's April 2026 release brought visual-AI capability to design-partner-class quality, with the result that the design-team org chart faces the same compression the engineering org chart absorbed through 2024-2025. The sequence of knowledge-work disciplines that compress under AI capability follows a recognizable pattern. Engineering went first. Design is going now. Naming which team is next is not difficult once the pattern is visible. https://thomasjankowski.com/writing/denial-expensive-claimable-inverts https://thomasjankowski.com/writing/denial-expensive-claimable-inverts Tue, 21 Apr 2026 11:13:00 -0400 Claimable's AI appeal-generation model inverts the cost asymmetry that made algorithmic denial viable, and absorbs the parallel HN read from signal 15. If every algorithmic denial triggers an automated appeal with clinical documentation, the marginal cost of denial rises toward the marginal cost of approval, equilibrating the AI arms race for patients and providers who adopt the counter-tool. The infrastructure-layer opportunity: capture and analyze the denial-appeal exchange itself, where the proprietary signal on payer behavior is more valuable than any single appeal outcome. https://thomasjankowski.com/writing/eulogy-for-the-ulcc-model-it-was-never-solvent https://thomasjankowski.com/writing/eulogy-for-the-ulcc-model-it-was-never-solvent Tue, 21 Apr 2026 09:07:00 -0400 Spirit Airlines, in April 2026, sits at the edge of liquidation. The end of an arc that ran across eight quarters of public confirmation: the NDC revolt at American, the JetBlue strategic retreat, Spirit's first Chapter 11 in late 2024, the Southwest-Elliott settlement that ended the cultural-differentiation era at the legacy-low-cost-carrier line, Spirit's second Chapter 11 in summer 2025, and now the liquidation overhang. The Lament Chronicles, opened in early 2024, close here. Long signature eulogy for the ULCC model, written in weighted-elegiac register, walking the arc and naming what survives the burial. https://thomasjankowski.com/writing/per-token-pricing-product-architecture-verdict https://thomasjankowski.com/writing/per-token-pricing-product-architecture-verdict Fri, 17 Apr 2026 09:19:00 -0400 Anthropic's shift from flat-rate enterprise pricing to per-token billing rewards applications that minimize tokens per outcome, compressing verbose chain-of-thought and multi-agent orchestration in favor of retrieval-augmented, narrow-context architectures that hit comparable outcomes with 10-50x fewer tokens. The regime advantages operators with curated domain knowledge bases over those running general-purpose LLM reasoning at scale. https://thomasjankowski.com/writing/ai-reads-api-not-hotel-brand https://thomasjankowski.com/writing/ai-reads-api-not-hotel-brand Thu, 16 Apr 2026 10:13:00 -0400 Hotel distribution leaders frame AI as the new gatekeeper, with trust replacing visibility as the surfacing mechanism, and signal 39's parallel framing folds into the same arc. The trust AI systems actually measure is structured data quality and review-signal consistency accessible via API, not the softer brand equity hotel marketing teams optimize for. Operators investing in data architecture win the AI-era visibility game; those investing in traditional brand marketing without API infrastructure become less visible as AI adoption increases. https://thomasjankowski.com/writing/opus-47-design-b2b-moat-reshaped https://thomasjankowski.com/writing/opus-47-design-b2b-moat-reshaped Thu, 16 Apr 2026 08:13:00 -0400 Claude Opus 4.7 (2026-04-17) shipped Claude Design — a collaborative design tool for visual outputs, prototypes, and slides. Operator read: when the frontier model becomes the design collaborator, the moat question for every B2B SaaS is whether the workflow they sell survives the model becoming a design partner. Three categories where the answer is yes, three where it is no. https://thomasjankowski.com/writing/aviation-alternative-fuels-math-boeing-s-q1-2026-delivery https://thomasjankowski.com/writing/aviation-alternative-fuels-math-boeing-s-q1-2026-delivery Wed, 15 Apr 2026 10:38:00 -0400 An operator-tier update on sustainable aviation fuel economics two years into the 2024 mandate ramp. Production capacity has grown but remains far short of the blending mandates' implied demand. The price differential to conventional jet fuel has narrowed but not closed. The blending limits at the engine and infrastructure level constrain how much SAF a route can actually carry. And Boeing's Q1 2026 delivery recovery, which the trade press read as good news for the carriers, has worsened the fuel-mix economics for the operators receiving the new aircraft. Long retrospective-update walking the production picture, the price picture, the blending limits, the route-by-route impact, and what the operator-class read on the trajectory should be. https://thomasjankowski.com/writing/urgent-care-competitor-no-cac https://thomasjankowski.com/writing/urgent-care-competitor-no-cac Tue, 14 Apr 2026 10:46:00 -0400 Gallup/West Health: 25% of Americans used AI for health information in the past 30 days; 14 million adults skipped a provider visit. Cost (14%) and access (16%) are the cited reasons. General-purpose LLMs with no acquisition cost are eroding the low-acuity volume that urgent care chains and telehealth platforms have been targeting, a market-structure inflection where AI is not competing with premium clinical care but displacing the bottom of the digital-health funnel. https://thomasjankowski.com/writing/imaging-rcts-unimplemented-llms-everywhere https://thomasjankowski.com/writing/imaging-rcts-unimplemented-llms-everywhere Mon, 13 Apr 2026 08:46:00 -0400 The healthcare-AI implementation paradox: imaging AI with strong RCT evidence (RETFound, pancreatic-cancer detection 475 days earlier than radiologists, four polyp-detection RCTs) sits unimplemented while LLM-based clinical decision support — with thin real-world evidence and documented triage-error studies — gets adopted by 72% of physicians. Operator read: deployment in healthcare diverges from what evidence supports; the operator playing the prioritization game right invests where evidence and adoption are converging, not where adoption is running ahead of evidence. https://thomasjankowski.com/writing/openai-s1-governance-read-next-contract https://thomasjankowski.com/writing/openai-s1-governance-read-next-contract Wed, 08 Apr 2026 14:07:00 -0400 OpenAI's late-2026 IPO at an anticipated $500B valuation will produce the first comprehensive risk-factor disclosure from a frontier AI company. Operator read: the S-1 is the governance document the industry has been asking for — every AI procurement contract written before the prospectus drops will look incomplete after. Three risk factors to watch for and what they cost when they show up. https://thomasjankowski.com/writing/unitedhealth-invested-1-5b-in-ai-no-canadian-payer-invested https://thomasjankowski.com/writing/unitedhealth-invested-1-5b-in-ai-no-canadian-payer-invested Wed, 08 Apr 2026 11:19:00 -0400 UnitedHealth Group's April 2026 announcement of approximately $1.5 billion in AI capex through 2026-2027 has no equivalent at any Canadian payer or provincial health authority. The Canadian framing is sometimes that the single-payer system does not need the investment because the operational pressures that drive U.S. payer AI deployment are different. The structural read is that the gap compounds for international capital allocation, talent acquisition, and longer-cycle clinical-AI capability. Long essay walking both readings, naming where each is right and wrong, and what the operator-class read on the gap should be. https://thomasjankowski.com/writing/operator-one-layer-up https://thomasjankowski.com/writing/operator-one-layer-up Sun, 22 Mar 2026 17:31:00 -0400 https://thomasjankowski.com/writing/spirit-asked-bailout-nobody-legislated-no https://thomasjankowski.com/writing/spirit-asked-bailout-nobody-legislated-no Mon, 16 Mar 2026 08:46:00 -0400 Through April 2026 Spirit was in talks with the Trump administration for a ~$500M bailout that would have given the U.S. government majority equity. Bipartisan opposition (Cruz, Pence, WSJ editorial board, every other airline) blocked it. Operator read: the absence of bailout was the signal — 14 years of post-COVID aviation policy has structurally shifted, and the next ULCC restructuring will be priced without a federal floor. https://thomasjankowski.com/writing/hospital-at-home-pilot-data https://thomasjankowski.com/writing/hospital-at-home-pilot-data Sun, 15 Mar 2026 10:46:00 -0500 CMS extended the hospital-at-home waiver through 2026, which means we now have almost six years of post-pandemic operational data on a model that was supposed to be a temporary expedient. The savings are real but smaller than the early advocates claimed; patient outcomes match in-hospital care on the cohorts the model was actually designed for; the labor-arbitrage worry that dominated the 2022 commentary did not materialize. Six concrete operator notes on what the data says, what it does not say, and where the model's real bottleneck has settled. https://thomasjankowski.com/writing/nyt-failed-the-blind-test-publicly-the-trust-stack-failed https://thomasjankowski.com/writing/nyt-failed-the-blind-test-publicly-the-trust-stack-failed Sat, 14 Mar 2026 08:13:00 -0400 Roose and Thompson's interactive blind test (2026-03-09) paired five passages of human writing against AI alternatives; 86,000 readers voted; 54% picked the AI passages. Operator read: the headline lands as 'Dead Internet at the paper of record,' but the procurement implication is sharper — your marketing-content vendor's deliverables are now indistinguishable in a blind test, and your QA process never had to know the difference. Three places that bites. https://thomasjankowski.com/writing/expedia-10k-named-agentic-board-owns-threat https://thomasjankowski.com/writing/expedia-10k-named-agentic-board-owns-threat Mon, 16 Feb 2026 08:46:00 -0500 Expedia's February 2026 annual report explicitly listed AI-powered competitors, data risks, and agentic AI as material risks — the first time a major OTA formally acknowledged structural threat in a regulatory filing. Operator read: when you disclose a threat in your 10-K, the board has accepted it as real; the next 12 months are about how the operator class redirects capital under the new acceptance, not whether the disruption is happening. https://thomasjankowski.com/writing/airbus-narrowbody-backlog-math-updated-boeings-q1-2026-deliv https://thomasjankowski.com/writing/airbus-narrowbody-backlog-math-updated-boeings-q1-2026-deliv Sun, 15 Feb 2026 14:24:00 -0400 An operator-tier update on where the Airbus A320/A321 backlog stood at end of 2025 and into early 2026, with the recalibration produced by Boeing's Q1 2026 delivery recovery. Book-to-bill ratios, delivery-delay shape, the engine-supply choke point, and the structural implications for the carriers waiting on the order book. Tight numbers piece, no advocacy. https://thomasjankowski.com/writing/yegge-burns-100-an-hour-in-parallel-the-unit-economics https://thomasjankowski.com/writing/yegge-burns-100-an-hour-in-parallel-the-unit-economics Sat, 14 Feb 2026 08:13:00 -0500 Steve Yegge's 2026 essays 'Welcome to Gas Town' and 'Wasteland' articulate an 8-level developer-trust spectrum, sustained $100/hour parallel-agent burn rates, and a junior-dev pipeline collapse thesis that Big Tech is currently choosing not to act on. Operator read: the burn rate is below median fully-loaded senior-engineer cost, the trust spectrum is a per-task dial rather than a ladder, the junior-pipeline collapse compounds for a decade, and the picks-and-shovels economy is captured by the model providers, not by anyone reading the trade press. https://thomasjankowski.com/writing/canada-tried-again-us-2016-playbook https://thomasjankowski.com/writing/canada-tried-again-us-2016-playbook Fri, 13 Feb 2026 09:13:00 -0500 Canada re-introduced the Connected Care for Canadians Act as Bill S-5 on 2026-02-04 after C-72 didn't advance. The legislation prohibits health-data blocking and mandates common standards for IT companies. Operator read: the second-attempt frame is the signal — Canada finally legislating what the U.S. did with 21st Century Cures in 2016; if S-5 passes, the $14B+ U.S. digital-health-funding playbook becomes directly applicable to Canada with seven years of better AI tools. https://thomasjankowski.com/writing/ai-content-problem-operators-cannot-tell-real https://thomasjankowski.com/writing/ai-content-problem-operators-cannot-tell-real Tue, 03 Feb 2026 10:31:00 -0500 The trade-press coverage of the AI-generated content trajectory through 2024-2026 has framed the problem as a public-safety concern about misinformation reaching consumers and citizens. The operator-class read is structurally different. The procurement decks, due-diligence packages, candidate-evaluation materials, vendor-pitch documents, and investor-class artifacts that operators evaluate every day are already substantially AI-generated, and the principals cannot reliably audit which parts. The trust architecture is breaking from inside the operator-class workflow, not from outside through the public-information environment. https://thomasjankowski.com/writing/claude-constitution-mostly-fine https://thomasjankowski.com/writing/claude-constitution-mostly-fine Tue, 27 Jan 2026 00:00:00 https://thomasjankowski.com/writing/55-36-6b-six-ipos-the-number-every-health-ai-deck-cites https://thomasjankowski.com/writing/55-36-6b-six-ipos-the-number-every-health-ai-deck-cites Thu, 22 Jan 2026 11:13:00 -0500 Bessemer's State of Health AI 2026: AI-enabled ventures captured 55% of all health-tech funding in 2025; six health-tech IPOs added $36.6B in market cap. Non-AI digital health funding is being crowded out, telehealth, digital therapeutics, and care navigation companies without a differentiated AI narrative face a higher hurdle for capital regardless of clinical evidence. The shift from 'digital health is dead' (2022) to 'AI health is the only game' (2025) tracks investor sentiment more than clinical evidence, and sets up conditions for a recalibration as the 2025 cohort hits the same reimbursement and regulatory gaps that slowed the 2021 vintage. https://thomasjankowski.com/writing/30-gone-ai-overviews-ota-repriced https://thomasjankowski.com/writing/30-gone-ai-overviews-ota-repriced Sat, 17 Jan 2026 08:13:00 -0500 By Q4 2025, AI Overviews and Google's AI travel planning had intercepted an estimated 30% of organic OTA traffic. Operator read: 30% is the structural tail-risk cap below which OTA equity stories now price; every pitch deck that doesn't model the cap is mispricing the discount rate. Three architectural choices that survive the cap. https://thomasjankowski.com/writing/next-wave-of-foundation-models-will-get-the-deployment-layer https://thomasjankowski.com/writing/next-wave-of-foundation-models-will-get-the-deployment-layer Fri, 16 Jan 2026 09:13:00 -0400 An advisor letter, written in early 2026, looking forward to the late-2027 retrospective on the foundation-model wave currently being trained and deployed. The benchmarks they will top are not the things they will be remembered for. The deployment-layer mistakes the capex cycle is making right now are. ADR-0031 mode-three inevitability register. The argument runs through three structural mistakes already locked in by the capex commitments of 2025: the inference-cost curve, the latency-and-locality assumption, and the regulatory posture. Forward-looking, conditional, register-loaded. https://thomasjankowski.com/writing/fine-tune-is-dead-long-live-the-prompt-engineer https://thomasjankowski.com/writing/fine-tune-is-dead-long-live-the-prompt-engineer Thu, 15 Jan 2026 13:46:00 -0400 By late 2025 the cost curves had decisively tilted: for the median use case, fine-tuning a foundation model was a worse investment than writing better prompts against a larger model, and the median engineering team had reached that conclusion. The headline is true, on the flat-frontier model. The forecast underneath the headline is that the fine-tune is not dead at all. It is moving, off the frontier model and into the smaller, specialized, edge-deployed model where the economics flip back in fine-tuning's favor. The piece argues both: fine-tuning the frontier model is dead for most teams, and fine-tuning the small specialized model is about to become the most consequential infrastructure decision a non-frontier AI team will make in 2026 and 2027. https://thomasjankowski.com/writing/expedia-builds-1500-agents-year https://thomasjankowski.com/writing/expedia-builds-1500-agents-year Wed, 14 Jan 2026 10:13:00 -0500 Expedia's CTO reports 1,500 internal AI agents built in 12 months across 17,000 employees, handling 143 million customer conversations annually. The 'AI playground' model, sandboxed access to 60+ LLMs, generates a proprietary dataset of use-case discovery, failure modes, and productivity patterns no vendor can replicate from outside, becoming an internal competitive intelligence advantage. The retention consequence: switching costs compound as employee knowledge becomes organizational infrastructure. https://thomasjankowski.com/writing/value-based-care-third-try https://thomasjankowski.com/writing/value-based-care-third-try Wed, 14 Jan 2026 09:07:00 -0400 Tight commentary on the phrase value-based care, which has migrated through three substantively different meanings since 2018. The 2026 working meaning is the most operationally useful version the U.S. healthcare system has produced, and it is also the least press-friendly version, which is why the trade press has been generally not naming the migration. Names the three meanings, what produced each, and why the third one is the one that actually shipped. https://thomasjankowski.com/writing/booking-expedia-2025-hierarchy-shift https://thomasjankowski.com/writing/booking-expedia-2025-hierarchy-shift Tue, 13 Jan 2026 11:07:00 -0500 Booking Holdings produced approximately 10 percent stock-price return in 2025 against Expedia Group's approximately 55 percent. The roughly 45-percentage-point divergence between the two major OTAs is the largest single-year hierarchy shift since the early 2010s. The divergence maps directly to differential AI-disruption exposure, differential European regulatory engagement, and the broader OTA-strategy positioning the two companies have been running. Short commentary on what the divergence reveals. https://thomasjankowski.com/writing/2025-digital-health-vintage-real-industry https://thomasjankowski.com/writing/2025-digital-health-vintage-real-industry Tue, 13 Jan 2026 10:52:00 -0500 The 2025 digital-health vintage, with approximately $14B raised, Hinge Health and Omada among the visible IPOs, and roughly 107 M&A deals through the year, looks structurally different from the 2021 vintage. The 2025 cohort wins on operating leverage and AI integration; the 2021 cohort won on engagement metrics that did not translate into durable economics. Medium retrospective comparing the two vintages, naming what broke between them, and what the structural difference implies for the 2026-2028 cohort.