Product strategy
Product decisions — launches, kills, pivots, and the bets that shaped what companies became. Every case is a real founder choice with a measurable outcome.
From the curated library
Ask the Directory -- Sign up to accessCursor: AI-native IDE — fork VS Code and go all-in on AI-first editing (2025)
Anysphere launched Cursor by forking VS Code and rebuilding it around AI-first workflows — inline editing, multi-file generation, and codebase-aware context. Rather than building a plugin for an existing editor, they bet that AI-native required owning the full IDE experience.
GitHub Copilot was the market leader in AI coding tools but operated as an autocomplete plugin within existing editors. Developers wanted more: multi-file edits, codebase-aware refactoring, and conversational coding. Building …
Cursor grew to an estimated $100M+ ARR within 18 months of launch — one of the fastest SaaS ramps in history. It attracted a devoted following among developers who found Copilot-style autocomplete insufficient. Raised at a $2.5B+ valuation. VS Code plugin alternatives couldn't match the integrated experience.
Meta: Release Llama 4 and embed AI across all social products (2025)
Meta released Llama 4 with frontier-class performance and integrated Meta AI as a core feature across WhatsApp, Instagram, Facebook, and Threads. The AI assistant was positioned as a daily utility for 3B+ users — not a standalone product but embedded in existing social workflows.
ChatGPT and Claude required users to visit a separate app or website — a significant friction point for mainstream consumers. Google had the same embedded advantage with Gemini in Android …
Meta AI reached hundreds of millions of monthly users through embedded distribution — no separate app download required. Llama 4's open release continued to build the open-source AI ecosystem. Meta's ad targeting improved significantly with AI, recovering revenue lost to Apple's ATT changes. The integrated approach proved more effective than standalone AI apps for consumer adoption.
Nvidia: Launch Blackwell GPU architecture and custom AI chips (2025)
Nvidia launched its Blackwell B200 and GB200 GPUs — purpose-built for AI training and inference at unprecedented scale. Simultaneously, Nvidia expanded into custom AI chip design for hyperscalers, directly competing with in-house chip efforts from Google (TPUs), Amazon (Trainium), and Microsoft.
The AI infrastructure buildout was accelerating — Microsoft, Google, Meta, and Amazon each planned $50B+ in capex for 2025. Nvidia's H100 had dominated 2023-2024 but competitors were closing in: AMD's …
Blackwell demand was immediate and overwhelming — all major cloud providers pre-ordered billions in chips. Nvidia's data centre revenue continued its explosive growth trajectory. The custom chip business opened a new revenue stream. However, supply constraints at TSMC limited availability, and the DeepSeek efficiency breakthrough raised questions about future demand growth rates.
Apple: Launch Apple Intelligence — on-device AI across all products (2025)
Apple integrated AI features across iOS, macOS, and its product line under the 'Apple Intelligence' brand. The approach prioritised on-device processing and privacy over cloud-based AI, with a partnership with OpenAI for complex queries users opted into.
Google had integrated Gemini deeply into Android and Search. Samsung partnered with Google for Galaxy AI. Microsoft had Copilot across Windows and Office. Apple was perceived as behind in AI …
Apple Intelligence became the largest AI deployment by device count — over 1 billion active devices. The privacy-first positioning resonated with consumers. However, early reviews noted Apple's AI features lagged behind Google and OpenAI in capability, and the slow rollout frustrated developers. The notification summary feature produced embarrassing errors, requiring patches.
Google: Launch AI Overviews in Search globally (2025)
Google rolled out AI-generated answer summaries at the top of search results globally, fundamentally changing the 25-year-old blue-links format. The AI Overviews synthesised answers directly, reducing the need to click through to websites.
Perplexity was growing rapidly with AI-native search. ChatGPT had added web browsing. Microsoft Bing had Copilot. For the first time in two decades, Google faced credible search competitors. Google's entire …
AI Overviews now appear on billions of queries. Google maintained search dominance and countered Perplexity and ChatGPT search threats. However, publishers saw traffic drops of 20-40%, triggering industry backlash, lawsuits, and regulatory scrutiny. The long-term impact on the web ecosystem remains contentious.
Anthropic: Launch Claude Code — AI-native CLI and agentic coding (2025)
Anthropic released Claude Code, an AI-powered CLI tool that could autonomously navigate codebases, write code, run tests, and manage git workflows. It was positioned as an agentic coding assistant rather than a simple autocomplete — a direct challenge to GitHub Copilot and Cursor.
GitHub Copilot had 1.8M+ subscribers but was primarily autocomplete — suggesting next lines of code. Cursor had shown that AI-native IDEs could capture developer attention. But a growing segment of …
Claude Code rapidly gained traction among senior engineers who preferred terminal workflows over IDE plugins. The agentic approach — where Claude could independently explore, plan, and execute multi-step tasks — differentiated it from autocomplete-style tools. It became a key driver of Anthropic's API revenue growth.
Anthropic: Launch Model Context Protocol (MCP) as open standard (2025)
Anthropic released MCP — Model Context Protocol — as an open standard for connecting AI models to external tools, data sources, and APIs. Rather than keeping it proprietary, they published the spec and SDKs for anyone to implement, including competitors.
The AI ecosystem was fragmenting — every model provider had proprietary function-calling formats, every tool had custom integrations, and developers were building N×M connectors. OpenAI had plugins (failed) and function …
MCP was adopted by major players within months — Cursor, Replit, Sourcegraph, and dozens of AI tooling companies integrated it. It became the de facto standard for AI-tool interop, similar to how USB standardised hardware connections. Anthropic gained ecosystem influence disproportionate to its market share.
DeepSeek: Release R1 reasoning model as fully open-source (2025)
Chinese AI lab DeepSeek released its R1 reasoning model with fully open weights, matching OpenAI o1 performance at a fraction of the training cost. They published their training methodology openly, shattering the assumption that frontier AI required billions in compute.
OpenAI, Google, and Anthropic had spent billions training frontier models, creating a narrative that only well-funded Western labs could compete. Export controls restricted China's access to top Nvidia GPUs. DeepSeek, …
DeepSeek R1 triggered a global market shock — Nvidia lost $600B in market cap in a single day as investors questioned whether AI infrastructure spend was overblown. The model was downloaded millions of times and became the foundation for dozens of derivative models. It proved frontier AI could be built for ~$5.6M vs billions, fundamentally reshaping the AI cost curve.
Perplexity: AI-native search engine challenging Google (2023-2024)
Perplexity launched an AI-powered search engine that provides direct answers with citations instead of blue links. Rather than competing on index size or ad revenue, they bet that LLMs would fundamentally change how people find information.
Google Search had been essentially unchanged for 20 years — type keywords, get blue links, click through to websites. SEO spam had degraded result quality. Reddit and forums became the …
Perplexity grew to 15M+ monthly active users and $35M+ ARR by mid-2024. Raised at a $3B+ valuation. Became the go-to example of AI disrupting an entrenched market. However, publishers filed lawsuits over content use, and Google launched AI Overviews as a competitive response.
Nvidia: All-in on AI/data centre GPUs over gaming (2022-2023)
Nvidia shifted its strategic focus and production capacity toward AI data centre GPUs (A100, H100) as AI training demand exploded. This meant deprioritising gaming GPU supply — their historical core business — in favour of $30-40K enterprise chips.
ChatGPT's November 2022 launch triggered an unprecedented AI infrastructure gold rush. Every tech company — Microsoft, Google, Meta, Amazon — needed thousands of GPUs to train and run AI models. …
Nvidia's data centre revenue grew from $15B to $47.5B in a single year (FY2024). The company's market cap surpassed $3T, briefly becoming the world's most valuable company. H100 GPUs had 6-12 month wait lists. Jensen Huang's bet on AI infrastructure proved to be the most valuable strategic call in recent tech history.
Apple: App Tracking Transparency — ATT privacy change (2021)
Apple required all iOS apps to ask user permission before tracking across apps and websites. This was framed as a privacy feature but had massive strategic implications — it devastated Meta's ad targeting while strengthening Apple's own growing ad business.
Facebook had built the most powerful ad targeting system in history by tracking users across the entire internet via the Facebook Pixel. This made Meta the second-largest ad platform after …
ATT cost Meta ~$10B in annual revenue and reshaped the entire digital advertising industry. Apple's own ad business (Search Ads) grew 200%+ as advertisers shifted spend to Apple's walled garden. Users opted out of tracking at 75-95% rates, validating the privacy thesis.
Meta: Open-source LLaMA AI models (2023)
Meta released its LLaMA large language models as open weights, allowing anyone to download and modify them. This was the opposite of OpenAI's closed approach. The strategic logic: commoditise the model layer so Meta's advantage in data and distribution would matter more.
OpenAI had gone from open to closed, charging for API access and keeping GPT-4's architecture secret. Google kept Gemini closed. Anthropic kept Claude closed. Meta had the largest social media …
LLaMA became the foundation of the open-source AI ecosystem. LLaMA 2 and 3 were downloaded millions of times. The open approach attracted top AI researchers to Meta, accelerated innovation, and prevented OpenAI/Google from creating a closed AI duopoly. Meta AI was integrated into WhatsApp, Instagram, and Facebook with 3B+ potential users.
Figma: Browser-based design tool with real-time collaboration (2016)
Figma bet that design tools should be browser-based with real-time multiplayer collaboration, like Google Docs for design. This was contrarian — designers were loyal to Sketch (Mac-native) and the browser was seen as too slow for design work.
Design was becoming a team sport. Product teams with designers, engineers, and PMs needed to collaborate on design files, but Sketch was Mac-only and desktop-first — sharing meant exporting PNGs …
Figma overtook Sketch as the dominant design tool. Browser-native collaboration proved transformative for design teams. Adobe attempted to acquire Figma for $20B in 2022 (blocked by regulators). ARR exceeded $600M by 2023.
Netflix: Invest $100M in House of Cards original content (2013)
Netflix committed $100M to produce House of Cards, their first original series. No streaming platform had produced premium original content before. The decision was based on viewing data showing users loved David Fincher and Kevin Spacey.
Content licensing costs were rising steeply as studios realised streaming was cannibalising their businesses. Disney pulled content from Netflix for its own platform (announced 2012). Starz didn't renew its deal, …
House of Cards was a massive hit that proved streaming platforms could produce prestige TV. It drove subscriber growth and led to Netflix spending $17B/year on content by 2023. The data-driven commissioning approach became Netflix's core competitive advantage.
Salesforce: Pioneer SaaS CRM with no software (1999)
Marc Benioff bet that enterprise software could be delivered via the internet, eliminating on-premise installations. At the time, Siebel Systems dominated CRM with traditional licensing. 'No Software' became Salesforce's rallying cry.
Enterprise software was broken. CRM implementations from Siebel and SAP cost $5-50M, took 12-18 months, and failed 60% of the time. Marc Benioff had just left Oracle where he was …
Salesforce proved the SaaS model viable and grew to $31B+ annual revenue by 2023. They effectively created the SaaS industry category. Siebel was acquired by Oracle in 2006 for a fraction of its peak value.
Amazon: Launch the Fire Phone (2014)
Amazon designed a smartphone with 3D 'Dynamic Perspective' and Firefly visual recognition. Priced at $199 with contract (matching iPhone), it was a direct hardware play to compete with Apple and Samsung in the premium segment.
Mobile commerce was exploding — over 50% of Amazon traffic was coming from mobile devices by 2014. The Kindle Fire tablet had shown Amazon could build affordable hardware, and the …
The Fire Phone was a catastrophic failure. Amazon took a $170M write-down on unsold inventory. The 3D gimmick added cost without value. It was discontinued within a year. Amazon redirected hardware efforts to Echo/Alexa, which succeeded.