Steve Blank: The World Outside The Classroom Changed, The Class Didn’t 

·       Agile was replacing Waterfall with short releases, customer feedback, pivots, but teams still assembled much of the application, deployment, monitoring, analytics and integration stack themselves.

·       Startups failed by building something nobody wanted. The limiting resources were engineering capacity and cash burn.

·       Creation is now cheap and everyone has the same leverage. What is scarce: attention and distribution, proprietary data and workflow depth, production capacity, and the evidence regulators, clinicians and payers will accept.

·       In Life Sciences the lab compressed; the clinic didn’t.

·       At product/market fit: 15–30 people, most of them engineers, burning $5M – $10M a year.

·       Priced as annual seat licenses; customers tolerated long implementation cycles because everyone did.

·       Moat: the time it took to write the code, plus switching costs, sales relationships and ecosystem.

·       Seed rounds, new and small ($500K–$1.5M from the first micro-VCs), gated on a working product, reference customers and a repeatable, scalable enterprise-sales model.

·       Funding went barbell: bootstrap to revenue or raise $50M for compute and distribution. Nobody raises for engineering headcount (deep tech and biotech excepted).

·       AI redefined unit economics. Inference costs have shifted, margins are pressured, pricing playbooks are failing; founders run product-led growth and enterprise sales at once.

·       Buyers expect workflows automated and compressed; they design-partner, pilot in days and pay for outcomes rather than seats. The product is increasingly an agent doing part of a workflow.

·       Discovery now asks, What is the job, the cost per completed unit of work? Who defines “done correctly,”? What errors are tolerable, which systems and data an agent touches, which approvals stay? How are decisions audited? Who owns governance, which budget pays, how many humans stay in the loop?

·       Defensible Moat: proprietary data, workflow depth, distribution, access rights, a calibrated eval set, moats of integration, data, compliance, reliability, brand, regulatory evidence, deployment learning — not the code.

·       5–10 people could design, build and market a polished app in 3–9 months.

·       App stores, Facebook, rankings and press were new enough to earn meaningful organic distribution. Growth hacking and A/B testing were emerging crafts.

·       Monetization deferred for years while investors funded user growth. Instagram was the poster child; VC flocked to B-to-C.

·       The catch: thousands of AI-built lookalikes launch every week, users compare every interface to a frontier chatbot, retention and loyalty are zero.

·       Discovery means mass creation and testing of problem framings, value propositions, onboarding flows and distribution hypotheses — a full-stack machine to test product, acquisition and retention.

·       “Users now, money later” is no longer funded outside genuine network-effect bets; revenue is expected almost immediately.

·       What survives: network effects, brand, taste. Everything else can be cloned in a weekend and will be.

·       Simulation sat at the end as verification; truth came from physical prototypes. A failed prototype meant new parts, revised tooling and another 12-week China tooling cycle. Build, test, break, rebuild.

·       CAE tools were $30K–$50K a seat and outsourced if done at all. Nothing resembling a digital twin.

·       The physical world still imposes truth: a simulation is a hypothesis, not a shipment. In 2011 a simulation explained why a prototype broke; in 2026 the prototype tests whether the simulation was right.

·       Money flows to defense and dual use. Working out of the box is table stakes; hardware is judged on its software, improved over the air; the Department of War demands commercial iteration speed.

·       Moats: deployment data, accreditation, production capacity. Building at scale is scarce again.

·       Decisions ran on small datasets and repeated wet-lab experiments, in sequential handoffs from biology to chemistry to preclinical to clinical.

·       12-25 core employees; animal studies, manufacturing and trials outsourced to CROs and CMOs.

·       A handful of specialist life-science VCs, milestone-gated Series A. Built to license or sell — the IPO window had been shut since 2008.

·       Not compressed: wet-lab validation, clinically meaningful endpoints, representative data, safety evidence, regulatory review. Trials, regulators, payers and reimbursement run on institutional clock speed. China now owns more than 50% of drug development.

·       When candidates are cheap and trials are not, the scarce skill is choosing the endpoint, indication and evidence package a regulator will accept, a clinician will act on and a payer will pay for.

·       Discovery of stakeholders matters more than discovery in the lab.

·       Firmware embedded and deterministic; connectivity a secondary feature.

·       FDA slow and unpredictable, so startups went to Europe first; US approval followed years later.

·       Device VC was fleeing over regulatory and reimbursement risk; 5-7 years to meaningful revenue.

·       Critical transition: from a prototype that worked once to a manufacturable device that works repeatedly.

·       Raising money, hitting a 510(k), selling to a strategic all has gotten harder. Clearance is no longer the constraint, payment is. 

·       TCET covers roughly five breakthrough devices a year; automatic Medicare coverage was never implemented.

·       FDA PCCP allows AI-enabled medical devices to receive updates without requiring a new marketing resubmission.

·       You can clear an AI device relatively cheaply and still have no payment mechanism. Early-stage device funding is really hard.

·       For Lean LaunchPad: in regulated markets evidence is part of the product. Discover what each stakeholder accepts as proof before optimizing the thing being proved.