Nate's core claim across this cluster: AI failure is organizational, not technical. Model capability was never the bottleneck; data readiness, unclear KPIs, missing change management, no human-in-the-loop design, and treating "AI strategy" as separate from business strategy are what tank the 80-95% of initiatives that stall. That thesis holds from his earliest SaaS pieces through his newest executive briefings, but what he tells leaders to do about it evolved considerably.
Late 2024 into early 2025, the question was whether AI kills SaaS. Nate's answer: no single force does, but three levers reshape it — AI-driven customization at scale, the falling cost of intelligence compressing margins, and "workflow breakage" where AI erases the process a tool was built around rather than just adding features to it. He argued vertical specialists survive platform consolidation by going deeper into messy, regulated, or specialist niches the generalist model-makers won't bother with.
By spring 2025, with model quality converging, his framing shifted from "which model" to "where is the leverage": his Distribution-Tokenization-Reward (D-T-R) framework and later "distribution, middleware, workflow" model told operators to stop chasing capability and start diagnosing which lever they actually control, then build a guiding policy (borrowing Rumelt's strategy kernel) instead of a feature list.
From late 2025 on, the frame got structural. As building got cheap (Lovable, Cursor, agent wrappers), Nate argued durable value moved to five things AI can't replicate on its own: trust, proprietary context, distribution/curation, taste, and liability. He introduced the "middleware trap" — companies sitting between model providers and customers get squeezed unless they own one of four defensible positions (proprietary context, infrastructure agents depend on, deep workflow lock-in, or trust/governance). His "coordination tax" argument went further: most knowledge work (60-70% of hours) is coordination overhead that agents dissolve entirely, not incrementally automate, which reorganizes headcount rather than just tasks. The weekly Executive Briefings apply this same "diagnose your actual exposure, don't chase the headline" discipline to live news — OpenAI's capacity bind, agent security holes, EU AI Act enforcement, data-readiness gaps — always ending in a prompt or diagnostic leaders can run that week.
Key moves
Run a diagnosis before you build anything: name the actual constraint (data quality, unclear KPI, missing owner), don't default to "we need a better model."
Score your position on Distribution, Tokenization, and Reward-clarity (D-T-R) before betting on a vertical or negotiating a vendor contract; a "5" on any single lever is where a platform will eventually squeeze you.
Pick one guiding policy (distribution-led, middleware-led, or workflow-led) and let it filter what you say no to — a list of AI initiatives is not a strategy.
Treat 80%+ of AI project failure as a people problem: fix data readiness, set one measurable business KPI, and build the human-in-the-loop escalation path before launch, not after a Klarna-style reversal.
Assume model quality keeps commoditizing; put your durable bets on trust, proprietary context, distribution/curation, taste, or liability — the five things a better model can't hand you for free.
Audit every vendor dependency with the "middleware trap" test: does this provider have both the incentive and the ability to absorb what you do, and what do you still own if they do?
Separate commodity engineering (sandboxing, linting, codebase hygiene — your team can do this in days) from genuine domain expertise (regulatory, compliance, jurisdiction-specific judgment — worth paying consultants for). Most SOWs blur the two on purpose.
Audit your calendar for the "coordination tax": meetings and translation artifacts that exist only because humans can't share context, not because they create value — agents collapse this layer before they collapse individual tasks.
Before adopting a new model or platform, ask what happens to your organizational context (accumulated institutional memory, workflows, CLAUDE.md-style instructions) if you have to switch — that's your real switching cost, not the subscription price.
Never let the CEO be AI-illiterate; if leadership can't use the tools themselves, transformation stalls regardless of budget.
Route sensitive documents to local/offline models (e.g., LM Studio) rather than uploading them to a hosted provider when the data can't leave the building.
Re-run your competitive/strategic diagnosis quarterly — a single keynote or model release can flip a weak lever to a strong one overnight.
2026-05-06 — The next AI platform winner won't have the best model — Computer-use gives agents reach into old software, but the durable moat is semantic control — owning the layer that tells an agent what an action actually means.
2026-03-24 — Accenture booked $2.2 billion in AI consulting last quarter — Scores the five hardest agent-deployment problems and finds a 4:1 ratio of solvable engineering work to genuine domain-expertise work, meaning most consulting spend is avoidable.
2026-03-19 — Perplexity shipped its best product and it might not matter — Even excellent execution on the middleware layer is fragile; durable positions are proprietary context, agent-dependent infrastructure, deep workflow lock-in, or trust/verification.
2026-03-05 — Grab the prompt kit I built to audit your AI platform lock-in — Argues the real prize is enterprise-scale context and retrieval, not model quality, and that switching platforms after months of accumulated organizational understanding may become nearly impossible.
2025-12-31 — The Composability Era: What Happens When Anyone Can Build a UI — No-code tools finally let non-engineers prototype real interfaces, while front-end engineering itself shifts from hand-building screens to defining the schemas and contracts that keep AI-generated UI consistent.
2025-12-21 — Executive Briefing: The Bubble Test for OpenAI — OpenAI is structurally caught between consumer simplicity and enterprise-agent friction; the real 2026 product is compute governance, not the model, and that gap is what leaders need to build around.
2025-11-23 — Executive Briefing: Daily Deep AI Adoption is a Silent Crisis — The strategic question has flipped from "does AI work" to "where do we still matter once it's everywhere," and most leaders haven't updated their thinking accordingly.
2025-11-16 — Executive Briefing: The Biggest AI Reset Since 2022 — As frontier-model leadership starts rotating every 6-12 months, durable value migrates away from the model itself toward whoever owns the intent-surface and workflow layer.
2025-10-31 — The Open Web is Over — As AI search replaces clicks with citations, there's a temporary 12-18 month window where small brands and individuals can out-rank incumbents in AI answers by structuring content for extraction rather than SEO.
2025-10-26 — Executive Briefing: What I Tell Leaders Stuck in AI Hell — Catalogs nine recurring, diagnosable failure patterns — from integration tar pits to premature scaling — that trap AI initiatives in permanent pilot purgatory.
2025-09-28 — Executive Briefing: Your 2026 AI Roadmap in Minutes — Lays out five inflection points and cross-cutting themes — compliance, memory, outcomes, on-device, premium AI — to anchor 2026 build/buy/budget planning.
2025-08-24 — Executive Briefing: Fire Your AI Strategy, Build an AI-Native Business — With 95% of generative AI pilots failing to move revenue, the fix isn't a better AI strategy — it's rebuilding organizational information flow itself, since hierarchies built for intelligence scarcity can't use intelligence abundance.
2025-08-06 — ChatGPT-5 Won't Save You: 10 Reasons Why Your AI Strategy is Failing — Walks through ten organizational failure modes — magic-wand thinking about data and models, vague KPIs, no human-in-the-loop, ignored total cost of ownership — that no model upgrade fixes.
2025-07-06 — The Executive Guide to Not Bullsh*tting About AI — Uses real regulatory penalty cases to show the cost of AI-washing claims and how to protect the business from both fines and customer abandonment.
2025-07-03 — The Complete Microsoft AI Copilot Roadmap 2025 — A comprehensive, 49-page-equivalent guide to Microsoft's fragmented Copilot product family, covering setup, workflows, org rollout strategy, and the Vodafone 68,000-employee case study.
2025-06-23 — Software 3.0 vs AI Agentic Mesh: Why McKinsey Got It Wrong — Contrasts Karpathy's builder-grounded "Software 3.0" (natural language as programming interface, partial autonomy) against McKinsey's consultant-crafted "agentic mesh," and shows why practitioners like Cognition and Anthropic reject distributed multi-agent architecture as fragile.
2025-06-04 — Mary Meeker AI Trends 2025 Deep Dive: 7 Unexpected Truths — A second-pass walkthrough of Meeker's deck surfaces points she glossed over: 2019 (not ChatGPT) as the real inflection point, AI proliferation over centralization, and the unresolved cost-vs-usage paradox.
2025-06-02 — I Summarized Mary Meeker's Incredible 340 Page 2025 AI Trends Deck — Summarizes Meeker's landmark report and responds point by point: adoption headlines mask organizational distrust, value concentrates in a few B2B niches with data/compliance moats, and durable human edge lives in judgment, taste, and persistence.
2025-04-18 — Why the Pipes Suddenly Matter — Whoever owns the data-ingest pipe owns the margin; the D-T-R scorecard (Distribution, Tokenization, Reward clarity) predicts which vendors will vertically integrate and where lock-in risk hides in a contract.
2025-04-15 — You Don't Need Better Models—You Need a Better Strategy — Applies Rumelt's diagnosis-guiding policy-coherent action strategy kernel to AI, arguing leverage now lives in distribution, middleware, or workflow depth — not model access.
2025-04-07 — The Best of Times, The Worst of Times—A Revised 2025 AI Outlook — A capital-markets shock collided with accelerating model capability, widening the gap between what AI can do and what organizations can actually deploy; middleware becomes the year's real battleground.
2025-03-21 — AI Means SaaS Doesn't Taste Like (Bland) Chicken Anymore — SaaS's predictable per-seat ARR model is cracking as AI enables outcome-based and usage-based pricing, using Klarna's swing from loss to profit as the clearest example.
2025-03-15 — Dead as a Dinosaur: 2010's Software Laws Are Gone — Contrasts the software-era laws (free-then-monetize, hyperscale via network effects, ecosystems) against new AI-era laws (compute is king, data network effects, distribution dominance, responsible innovation as competitive advantage).
2024-12-17 — When AI Agents Devour the App Layer — Responds to Satya Nadella's claim that AI agents dissolve the SaaS app layer by arguing vertical specialists thrive by owning the 20% of tasks too niche, regulated, or precision-critical for generalist agents.
2024-12-05 — SaaS in the Age of AI: The Great Software Overhang — Introduces three levers reshaping SaaS value — AI-driven customization at scale, the falling cost of intelligence, and workflow-breaking discontinuities — as a more useful frame than the bulls-vs-bears debate.