Skip to content

Case Studies

Companies and case studies referenced in Blueprint for an AI-First Company, organized by category. Each entry includes what the book discusses and which chapters reference them.

AI-First Companies

Companies built from the ground up with AI at their core. If you remove the AI, the product doesn't exist. Central examples in Chapter 1 (The AI-First Imperative) and the AI-First vs AI-Enabled framework.

Company What the Book Discusses Chapters / Frameworks
Harvey Legal AI platform that reached $8B valuation and $100M ARR. Uses "citation-first output" with 74% answer quality on BigLaw Bench. Expert seeding strategy -- hired lawyers from major firms to define workflows. LexisNexis partnership for exclusive content. Compliance-native architecture anticipating EU AI Act, ABA 2024, UK SRA standards. Ch 1 (AI-First Imperative), Ch 2 (Mindset), Ch 9 (Data Strategy), Ch 11 (Ethics & Governance)
Glean Enterprise knowledge platform with $7.2B valuation and $100M+ ARR. Knowledge graph takes 12-18 months to mature. Integration depth across enterprise data sources creates switching costs. $30/month per user, claims 2-4 hours per week savings. Ch 1 (AI-First Imperative), Ch 9 (Data Strategy)
Perplexity AI-native search. 312M queries in May 2024, 780M by May 2025, growing to 1.4B by June 2025. DAU/MAU ratio of 53% far exceeds benchmarks. Network learning effect -- each query improves accuracy for future queries. Demonstrates cold start breakthrough via network effects. Ch 1 (AI-First Imperative), Ch 9 (Data Strategy)
Midjourney AI image generation. Launched 2022 with 11 employees, hit $200M annual revenue by 2023. No separate "AI team" -- the entire company is the AI team. Example of AI expertise distributed throughout the organization. Ch 1 (AI-First Imperative)
Cursor AI-first code editor. Revenue grew from $1M (2023) to $100M (2024), projected $200M in 2025. Implements explicit PermissionOptions with allowlists, denylists, and "YOLO mode." Tab completion model achieved 28% higher accept rates with 21% fewer suggestions. OpenAI engineering teams adopted it. Ch 1, Ch 2 (Mindset), Ch 5 (Building with AI), Ch 9 (Data Strategy)
Mistral European AI lab. Founded 2023, reached $6B valuation within 18 months. Fastest time-to-first-token at 0.30 seconds. Offers both open-weight and commercial models. Ch 1, Ch 3 (AI Landscape)

AI-Enabled Companies

Established companies that added AI capabilities to existing products. The core product survives without AI, but AI makes it significantly better.

Company What the Book Discusses Chapters / Frameworks
Salesforce Marketing says "Now the world's number one generative AI CRM" -- classic AI-enabled positioning. Agentforce platform allows banking agents to retrieve transactions autonomously but requires human approval for credits and merchant notifications. Built CRM since 1999. Ch 1, Ch 11 (Ethics & Governance)
Notion Turn off Notion AI and the workspace still functions for notes, docs, and wikis. Product existed since 2016. AI priced as $10/month add-on per member. Example of AI as enhancement, not foundation. Ch 1 (AI-First Imperative)
Adobe Describes designers becoming "creative directors for an incredibly fast, versatile, but literal-minded AI assistant." Illustrates the Creative Director Model for human-AI collaboration. Ch 2 (Mindset)
Shopify CEO Tobi Lutke issued mandatory AI policy in April 2025 -- before requesting headcount, demonstrate why AI can't handle the task. Runs 40-60M LLaVA inferences daily using fine-tuned open models. Ch 2 (Mindset), Ch 3 (AI Landscape)
Canva Magic Studio serves 220M+ monthly users. Magic Write feature saw 8B uses since launch. Demonstrates the compound iteration mental model -- multiplication effect of AI across a large user base. Ch 2 (Mindset)
Grammarly Shows multiple suggestions instead of one "right answer." Targets 95% user-generated accuracy. 10 suggestions yield 98% accuracy, 44% activation rate. Example of designing for probabilistic outputs. Ch 2 (Mindset)
Figma Data shows 84% of designers collaborate with developers weekly as AI compresses the gap between concept and prototype. But fewer than half feel AI makes them better at their jobs -- the efficiency trap. Ch 2 (Mindset)

Infrastructure & Operations Cases

Companies whose AI infrastructure and operational decisions provide key lessons for the book.

Company What the Book Discusses Chapters / Frameworks
Klarna Bought OpenAI models for customer service. Month one: 2.3M conversations, resolution time dropped from 11 to 2 minutes, equivalent to 700 agents, projected $40M profit improvement. By mid-2025, began rebalancing toward human agents -- AI excelled at routine but couldn't handle fraud claims, disputes, or emotional scenarios. Ch 2 (Mindset), Ch 9 (Data Strategy), Ch 10 (Operations & GTM), Ch 11 (Ethics)
Morgan Stanley Took GPT-4 and trained it on 70,000+ proprietary research reports. 98% of advisor teams actively use the tool. "Makes you as smart as the smartest person in the organization." Classic Boost path -- vendor model plus proprietary data. Ch 2 (Mindset)
Bloomberg Spent $3.5-8M training BloombergGPT (50B parameter model). 9-person team. Data privacy drove the Build decision -- "Using an API like OpenAI's isn't suitable for us." Serves clients paying $25K+ annually. Ch 2 (Mindset)
OpenAI Spent $9B to generate $4B revenue in 2024. Multiple major outages (June 2025: 12 hours, 21 components; December 2024: 9 hours from Azure datacenter power failure). Illustrates single-point-of-failure risks and platform dependency. Ch 4 (Infrastructure)
Vercel (v0) Iterates on prompts "almost daily" using automated evaluations. Each edge case becomes a test case preventing regression. Example of compound iteration at speed. Ch 2 (Mindset)

Data Strategy Cases

Companies whose data strategies -- both successes and failures -- illustrate core data principles in the book.

Company What the Book Discusses Chapters / Frameworks
Tesla 2M+ vehicles capture "Autopilot Snapshot" clips of edge cases automatically. Automatically surfaces the 0.01% of cases that train networks. Earns $7K per vehicle as a data collector vs. Waymo spending $150K per vehicle. However, FSD "hasn't improved all year" based on 2025 data -- data collection without proper curation creates noise. Ch 9 (Data Strategy)
Spotify 1.4 trillion events daily from 678M users. 520 experiments on mobile home screen alone each year. Multi-task training shows transferable learning across podcasts and music. Users engaging AI recommendations show 40% higher retention, 140 vs. 99 minutes daily usage. Ch 9 (Data Strategy)
Duolingo Birdbrain AI estimates probability of correct answers. When learners struggle, difficulty updates in real-time for all current and future learners (network learning). 59% DAU growth (21M to 34M users), 80%+ organic acquisition, near-zero customer acquisition costs. Rewrote Session Generator from 750ms to 14ms. Ch 9 (Data Strategy)
Netflix Handles 4,000+ daily deployments with automated canary rollouts. Deploys in under 15 minutes vs. traditional enterprises at 8-90 days. Illustrates the velocity gap in data-driven iteration. Ch 9 (Data Strategy)
Zillow Shut down Zillow Offers in 2021 after AI property valuation failed. Wrote down $500M+, $304M Q3 losses, 25% workforce reduction. Two-thirds of purchased homes valued below purchase price. Models relied on data 30+ days old for near real-time decisions. Ch 10 (Operations & GTM)
Stitch Fix Experienced client declines, recovered with hybrid AI-human model. ML generates recommendations, human stylists add nuance. Result: 40% increase in average order value, 40% increase in repeat purchases, 30% reduction in returns. Recovery took 12-18 months. Ch 9 (Data Strategy)

Governance & Risk Cases

Companies and incidents that illustrate AI governance, ethics, and risk management lessons.

Company / Incident What the Book Discusses Chapters / Frameworks
IBM AI Ethics Board established 2019. Policy Advisory Committee, distributed accountability with ethics focal points in every business unit. AI Risk Atlas embedded in watsonx. Five years of governance refinement. Ch 11 (Ethics & Governance)
JPMorgan Elevated AI governance to 14-member Operating Committee in 2025. CDAO at the table -- one of few Fortune 1000 CDAOs at that level. AI innovation as operating committee mandate. Ch 11 (Ethics & Governance)
Air Canada Chatbot invented a "bereavement fare" policy. Canadian tribunal ruled the airline liable ($812.02). Established precedent: companies are liable for what their AI agents say. Ch 6 (Agent Architecture), Ch 11 (Ethics)
Chevrolet Dealership ChatGPT-powered chatbot with insufficient guardrails. Users manipulated it to agree to sell a Tahoe for $1 and recommend Tesla. Went viral. Illustrates scope creep failure mode. Ch 6 (Agent Architecture)
Samsung Three engineers entered proprietary source code into ChatGPT within 20 days. Samsung banned all generative AI tools company-wide. Illustrates privacy leakage risk. Ch 11 (Ethics & Governance)
X (Grok) Grok chatbot started injecting political claims into unrelated conversations in May 2025. Hardcoded administrative instructions overrode evidence-based programming. No audit trails. Ch 4 (Infrastructure)
Clearview AI 30.5M euro GDPR fine for facial recognition database built from 30B+ images collected without consent. Ch 9 (Data Strategy), Ch 10 (Operations & GTM)