AI Applications

Bob Iger believes AI will enable better stories in the hands of great storytellers, improve operational efficiency, and create closer customer relationships, but he strongly rejects the idea that AI will replace human creativity, viewing it as a 'bicycle of the mind' rather than a destructive force.

OpenAI, Anthropic's Claude, and Perplexity all employ freemium models with paid tiers ($20-200/month), suggesting a standardizing business model for consumer AI applications where widespread free usage funnels toward premium subscriptions.

Ajay Banga argues AI will democratize specialist knowledge (health, climate, education) for poorer countries just as Google democratized information access, though infrastructure gaps mean near-term benefits require digital infrastructure investment.

AI-driven multi-physics simulation compresses climate computation from weeks to seconds, allowing policymakers and industries to simulate regional extreme weather, infrastructure needs, and agricultural impacts for better capital allocation.

Mastercard has launched consumer-facing generative AI assistants for shopping (natural language product discovery) and travel planning (automated itinerary booking using loyalty data), starting with a Michael Kors pilot.

Mastercard uses generative AI not just to score existing transactions but to predict the next likely fraudulent transaction by analyzing data lake patterns, and deploys AI shopping assistants that automate travel and product discovery using permissioned loyalty and payment data.

Kenichiro Yoshida positions AI as a tool to augment human creativity — streamlining game development via LLMs while preserving copyright and creator involvement — arguing the entertainment business is fundamentally people-driven and AI should not displace creators.

Iger believes AI will enable Disney to tell better stories, improve efficiency, and deepen customer relationships, viewing it as a powerful tool in the hands of great storytellers. However, he cautions that AI must not destroy human creativity and that companies should proceed responsibly, acknowledging more is unknown than known.

The emergence of AI-powered browsers will create a fragmented landscape driven by different consumer and enterprise use cases, requiring a cross-platform credential trust layer that works across all browser environments.

Stoltenberg states AI is causing a fundamental shift in warfare comparable to the Industrial Revolution, citing autonomous systems, facial recognition, drones, and AI-enabled cyber attacks. He notes NATO already uses AI for surveillance and data processing in Ukraine, but this is only the beginning.

Mastercard's decade-long AI investment now leverages generative AI to predict likely fraudulent transactions before they occur by analyzing data lake patterns, while also launching consumer-facing shopping assistants that automate complex purchase decisions using permissioned loyalty and public data.

AI born from computing will converge with entertainment, but content is creator-generated and copyrighted, so creator involvement is essential. Sony positions AI as a technology that supports creators — e.g., LLMs streamlining game development — rather than replacing them, raising ethical questions about uses like AI-generated dubbing.

Tools like Microsoft Copilot and GitHub's AI-generated code already produce 40-50% of software on GitHub, effectively doubling engineer output, and Jensen estimates a 10x productivity improvement for Nvidia's engineers — transforming the economics of the most expensive workforce in tech.

Iger views AI as a tool that will improve story quality, make distribution more relevant to consumers, and increase company efficiency, but he is adamant that AI must not destroy human creativity and that the original storyteller remains irreplaceable for decades to come.

Tangen reports that 40% of staff now contribute code and nearly half can program, with AI tools like Anthropic's Claude deployed across trading, operations, and analysis driving a 15% average efficiency improvement, demonstrating scalable AI integration in large-scale asset management.

Generative AI is being applied in payments both defensively — predicting the next likely fraudulent transaction by analyzing patterns in data lakes — and in consumer applications, where Mastercard has launched AI-powered shopping assistants that use natural language to find products (e.g., a linen suit for a beach wedding) by combining loyalty data, open data, and payment credentials.

Natural language programming eliminates the need for traditional coding languages, enabling billions of people to create software and applications, which will drive an explosion in AI-native applications similar to the iPhone's impact on mobile apps.

Bob Iger argues AI will serve as a powerful tool to improve story quality, distribution relevance, and operational efficiency at Disney, but believes human creativity remains irreplaceable for original storytelling, positioning AI as an enabler rather than a threat to creative IP value.

Mastercard is deploying generative AI for consumer-facing shopping assistants that automate complex purchases (travel, apparel) using permissioned data, while simultaneously using AI to predict fraud patterns across its network — turning data advantage into both revenue and security moats.

Labs are systematically covering high-value domains with tailored RL environments; Harvey (legal) and Cursor (coding) demonstrate rapid improvement via deployment-data flywheels; this vertical approach yields faster commercial returns than waiting for general reasoning transfer.

AI-driven multi-physics simulation accelerates weather/climate modeling by orders of magnitude, allowing hyper-local climate impact prediction (extreme weather, agriculture, water supply, insurance) and cost-effective infrastructure investment decisions — a high-value AI application with immediate economic payoff.

Mastercard has launched live AI shopping assistants that combine consumer permissioned data — loyalty accounts, payment history — with public data to automate trip planning and product discovery via natural language. This represents a concrete monetization path for AI in the payments stack beyond fraud prevention.

Generative AI can extend access to high-quality healthcare, education, and climate information to developing nations, mirroring Google's democratization of information; however, lacking digital infrastructure delays tangible benefits.

AI technologies like LLMs will streamline game development and content creation, but the content forming the basis of entertainment is creator-generated and copyrighted, so creator involvement remains essential and AI should be positioned as a support tool.

If token prices collapse to marginal cost (power, light, water) due to infrastructure overbuild, the winning investments will be boring enterprise applications — supplier reconciliation, data matching — that can waste massive token volumes for efficiency gains, not the infrastructure providers themselves.

Arora predicts agentic AI will fundamentally change product development and enterprise workflows, eliminating the need for human-centric UIs and seat-based pricing; the market will initially miss this shift then overcorrect, creating value opportunities in companies adapting to agent-first architectures.

Schwimmer describes a new AI-powered meeting preparation product built with Microsoft that synthesizes public, financial, research, corporate, and private data (email, CRM) into memos in seconds, replacing hours of analyst work.

Apple's Siri AI launch provides system-level access to 300,000 third-party app intents, enabling cross-app agent workflows (texting, calendaring, app control) that chatbot interfaces cannot, leveraging Apple's 2.5B device installed base for immediate distribution.

Gary Tan predicts a massive turnover in software as AI-native startups with ~20 engineers using agentic systems will out-execute legacy SaaS companies employing thousands of engineers using old methods.

Exxon is aligning IT with R&D to deploy AI across operations, aiming to 'highgrade' workforce tasks, shorten innovation and implementation cycles, and tackle harder problems faster, starting with internal productivity gains.

Sal Khan describes how Khan Academy became OpenAI's first education partner, gaining early access to GPT-4 to build Khanmigo — an AI tutor with guardrails for safety, privacy, and Socratic teaching. The product launched March 2023 and is scaling faster than expected, demonstrating that frontier models can deliver personalized tutoring at scale, addressing Bloom's two-sigma problem.

Tools like Lovable and Bolt.new let users generate functional websites from plain English prompts, pushing platforms like Beehiiv to build first-party AI builders to retain users who now expect zero-friction site creation as a baseline feature.

Diageo tracks 220M consumption occasions across 37 markets (who, what, where, with whom) to deploy brands precisely against high-potential moments — improving marketing ROI and enabling faster innovation cycles in volatile consumer environments.

AI will eliminate call centers and collapse demand for Indian outsourcing/coding services as AI can perform coding with half the people, increasing disruption for businesses without durable moats.

AI tools that fully automate ad campaigns — creation, targeting, placement — enable small businesses to generate more sales at lower cost, creating a powerful revenue driver for digital advertising platforms.

Chris Kempczinski identifies concrete AI use cases including personalized drive-through menus, customized employee training in local languages, and demand forecasting for food preparation (eProduction), arguing these will materially improve speed, freshness, and labor efficiency across the industry within a five-year horizon.

Trip.com applies AI to user interface (Trip Genie conversational search), engineering productivity (15-30% coding time reduction via copilot), customer service (automated need identification), and content generation (video/picture creation), creating a scalable model for travel platform efficiency.

Clients will turn to AI assistants that know them better than human advisors for wealth advice. AI can process unstructured data and complex biological systems, so it will certainly become better at investment decisions. Asset managers must put AI tools and data in front of portfolio managers.

Rentokil is deploying camera-based technology with facial recognition to identify individual rodents, enabling precise tracking and targeted treatment that reduces chemical usage while improving efficacy, demonstrating AI's potential to transform traditional service industries.

Brodin states AI is already revolutionizing IKEA's value chain — simplifying goods flow to save costs, scaling customer service, and transforming home planning for customers. The company views AI as a broad operational lever, not a single use case.

Jane Fraser argues AI will amplify human work in banking (like an intern providing information but not decisions), with opportunities outweighing threats, though she highlights cyber fraud and social media-driven geopolitical instability as key risks.

Math falls first because every step can be validated by computer — unlike chemistry or physics which require real-world experiments. This creates a template for other formalizable domains. Current product failures (DoorDash AI) are implementation growing pains, not model limitations.

Figure skating governing body (ISU) permits AI music for routines to bypass complex multi-party licensing of contemporary songs; Czech team using Bon Jovi-inspired AI track at Winter Olympics — first widescale cultural test of AI music acceptance.

Tony Xu believes LLMs will unlock organization of physical-world data — parking, inventory, catalogues — that remains largely offline, and DoorDash is actively experimenting with AI for customer support, fraud detection, and catalogue building, though still early in the progress bar.

AI will capture the functional background music market but artist-identity-driven music will be protected through permission-based licensing with platforms, model companies, and eventually regulation like the No Fakes Act.

One million connected elevators feed a data lake that allows preventive and prognostic maintenance, reducing downtime and expanding into passenger-facing apps for traffic optimization and building integration.

Runway's CEO emphasizes building video models as creative empowerment tools for artists rather than mass-content generation, differentiating their approach from competitors focused on social media scale.

PhysicsX is moving engineering simulation from legacy numerical methods to AI inference, achieving 10,000x to 1Mx speedups while training on both simulation data and real-world measurements from wind tunnels, test dynos, and operating equipment. This enables high-fidelity simulation to be used downstream in manufacturing, testing, and operations — not just detailed design — creating a new software layer for industrial production.

The core challenge in AI-driven shopping is matching ambiguous natural-language needs to the right products across thousands of brands, not processing payments. Vertical-specific platforms with deep product understanding have an advantage over general-purpose LLMs.

AI search exposes niche merchants without big ad budgets, expanding the commerce torso and tail; however, shopping enjoyment means users may prefer AI for discovery but retain control over final purchase decisions.

Criminal organizations across Eastern Europe, Southeast Asia, and West Africa deploy generative AI for industrial-grade impersonation, achieving 9x message volume and 4x revenue versus peers, proving AI's dual-use economics favor bad actors without consumer-side defenses.

Models now matching human superforecasters on ForecastBench will plug into capital markets and corporate decision loops, making probabilistic foresight a commodity and displacing experience-based executive roles.

Netflix is scaling generative AI across its production pipeline, particularly post-production, reducing costs and enabling content that wasn't financially feasible with traditional methods.

Jon Steinback argues that current LLMs anchor thinking and restrict creativity, and that the winning approach is AI that writes with users—providing scaffolding and feedback while keeping the human in control of every word choice.

DLSS 5 introduces generative AI as a complementary scaling lever to traditional rendering compute — distilled diffusion transformer models enrich rendered frames with photorealistic detail while preserving artistic intent, enabling real-time path-traced quality at feasible frame budgets.

Even with foundational models, building production-grade CPQ requires solving complex data models, pricing engines, multi-stakeholder workflows, legal/compliance integrations, and load testing — making buy preferable for non-strategic but mission-critical systems.

By embedding AI agents into lettings workflows, Dwelly automates compliance and property‑management tasks, freeing agents to focus on high‑value client interactions, which drives scalable growth and creates a defensible model replicable across other European service sectors.

All AI products are converging to horizontal platforms covering entire workflows (like Microsoft Office or Google Drive for human-agent interaction), with differentiation coming from product sensibilities such as collaboration and multiplayer features rather than vertical depth.

Enterprise customer service AI agents prioritize speed (time-to-first-token) and cost over raw capability, giving an edge to smaller models from OpenAI, Google, Mistral, and Qwen over Anthropic's larger frontier models.

AI video tools (Midjourney, Kling, Seed Dance) now enable indie creators to produce blockbuster-quality sci-fi at fraction of traditional cost, with YouTube creators leveraging 600M+ view audiences to self-distribute theatrically, signaling a structural shift in content production and distribution economics.

Claude 4.5 / Sonnet coding agents triggered step-function adoption: SemiAnalysis went from <10 to 90 users (10x) and per-user token spend 10x, driving 100x company AI spend; suggests enterprise coding is first killer app with massive latent demand across knowledge work.

Real-time voice agents prioritize latency (time-to-first-token) and cost over raw intelligence, creating a segment where small models from OpenAI, Google, Qwen, Mistral outperform larger frontier models; Anthropic's focus on coding benchmarks has left it under-optimized for this growing but smaller TAM versus AI coding tools.

An engineer with today's best models is approximately 1000x more effective than one without, and there is no single programming task where a human outperforms the model — pure software development is no longer a durable moat.

Applications solving N-squared coordination problems across multiple organizational constituents (e.g., Legora in legal with corporate counsel, law firms, clients) build defensible moats that horizontal foundation models cannot easily replicate; expansion into adjacent verticals (tax, compliance) compounds value.

Management signaled enterprise focus with internal coding tools potentially offered externally via API; if Meta's frontier models become state-of-the-art, excess compute capacity could be sold to enterprises, creating a new revenue line in 2-3 years.

General models improve broadly but leave durable gaps: (1) products with access to personal/private data the base model cannot see, and (2) extremely hard problems where a niche model trained on specific data (like AlphaFold for protein folding) achieves high accuracy affordably — material science and chip design are cited as similar opportunities.

Frontier labs (OpenAI, Anthropic) are prioritizing healthcare but cannot deploy in isolation; the investable opportunity lies in application-layer startups that connect models to real-world data (EHR, wearables) and workflows through partnerships.

Synthetic panels will surpass human panels by unlocking the 95% of hypotheses currently untested due to cost and complexity, with high-stakes simulations costing $10-20M to run but preventing billion-dollar failures, enabling value-based pricing.

Sea Limited has deployed AI chatbots across millions of daily orders, achieving over 80% full automation of customer inquiries with satisfaction rates exceeding human agents, proving AI can scale high-quality support in high-volume consumer platforms.

Sea's Free Fire game (100M+ daily players) now uses AI to generate game art assets previously outsourced, and deploys sophisticated AI bots that are indistinguishable from human players to improve new user onboarding and team-play engagement.

As open-weight models become sufficiently capable and runnable on efficient hardware, the marginal cost of AI delivery approaches zero, allowing application-layer software to revert to predictable SaaS pricing rather than usage-based models.

Conveo's founder argues that enterprises need 'AI plus people' to unlock value, as AI raises the floor on research volume but requires change management to route insights to decision-makers across functions.

Every great company of the last 20 years has had exceptional design; AI-generated 'vibe coded' designs with telltale patterns (excessive bold, purple gradients, card overload) make startups look generic and untrustworthy. Founders must invest in intentional design to stand out and attract talent.

AI music tools will serve as powerful assistants for talented producers and artists (breaking logjams, enabling previsualization), but cannot replicate the lived experience and emotional provenance that define great art; licensing frameworks will evolve to compensate training data sources.

Founders with lifelong vertical exposure (HVAC, pharmacy, audio) possess unrecognized edge; AI automates grunt work while vertical knowledge drives sales and trust.

High-stakes decisions (World Cup sponsorship, defense, supply chain) will drive massive per-query spend on foundation models; new enterprise budget category emerging.

Despite hype, consumer AI impact limited beyond voice interfaces; Suno's pivot from creation to consumption tool could unlock Spotify-scale opportunity.

AI-powered computer vision enables plant-level precision — distinguishing weeds from crops and targeting inputs only where needed — reducing herbicide application by 60% and nitrogen fertilizer costs by two-thirds, creating a win-win for farmer profitability and sustainability.

As model providers like OpenAI and Anthropic build their own applications to capture value if API access is restricted, vertical software companies face existential pressure from both rising inference costs and direct competition from their model suppliers.

Winning AI applications (Cursor, Sierra) embrace building disposable 'sand castles' — products architected around current model jagged edges, with teams that re-evaluate every assumption every 4 weeks as capabilities shift, inverting traditional product management.

Investors view Suno as potential Instagram-like platform for AI music creation, but lawsuits from Universal and Sony create existential risk; Warner and BMG settlements show path to resolution but timeline uncertain.

Outset's persona-core digital twins (grounded in 75-min paid interviews + operational data) let any employee simulate buyer reactions to pricing, features, messaging — unlocking research for segments too expensive/slow to survey continuously.

The competitive advantage in AI video is moving from raw model quality to opinionated product harnesses that handle character consistency, timing, and editing workflows, with cost per clip remaining a key constraint.

Metaview's Fillmore acts as an AI co-worker that autonomously handles sourcing, research, outreach, and scheduling, delivering hiring outcomes in half the time with better candidate experience than human recruiters.

Management highlighted AI tools that increase platform engagement while reducing human doctor/provider involvement on the backend, creating operating leverage as the subscriber base scales toward 10-20M+ users.

Claire integrates with major payroll providers (Gusto, QuickBooks, TriNet) to offer instant wage access, reducing credit risk via employment verification and expanding into auto loans and broader workplace-connected credit for the 50% of US workers living paycheck-to-paycheck.

AI agents combining semantic search with multi-step reasoning can discover causal patterns across disparate datasets (weather, 911 calls, terrain) that are invisible to keyword search, enabling proactive resource allocation for events like rip currents or hate-crime threats.

AI agents will replace human wedding planners and travel coordinators by instantly matching preferences to vendors (florists, bands, hotels) across complex multi-leg trips, with Booking already deploying early tools and iterating rapidly.

The industry is shifting from a few generalist frontier models to millions of domain-specific models (legal, finance, healthcare, coding, security), each post-trained on proprietary product data and reward signals, making post-training the new standard after product-market fit.

Canva, Notion, Airtable represent 'no-code era' tools now threatened because agents bypass them entirely — users generate flyers, databases, apps directly in ChatGPT/Claude without needing specialized SaaS.

Free frontier models (ChatGPT, Gemini) provide superior homework assistance at zero marginal cost, destroying the value proposition of paid EdTech platforms like Chegg; revenue down 39% YoY and market cap collapsed to ~$85M.

Paid acquisition fueled by massive creative velocity (400-500 UGC assets/month), affiliate programs paying revenue share, and referral credits tied to usage limits is the winning growth model for AI SaaS, replacing pure PLG. Meta's Andromeda update made creative the targeting, requiring constant creative diversity to avoid fatigue; token economics allow higher CACs that justify the spend.

The highest ROI for continual learning is on tasks at the edge of model capabilities where users push boundaries; models learn from these interactions to master previously impossible tasks, expanding the product's frontier.

Small businesses are 'perfect AI hedge' — last disrupted by AI/robotics (physical work); AI should drive revenue growth not cost cutting because employees wear multiple hats (automating one hat doesn't reduce headcount); concrete deployments: Claude for outbound sales in painting business, AI social media marketing for 7-generation peanut manufacturer enabling D2C channel.

Frontier labs (OpenAI, Anthropic) could aggregate ride-hailing/delivery via agents, but managed transactions (driver-rider coordination, pickup experience, lost items, payment) require operational infrastructure that pure software players lack. Uber will participate selectively with APIs but retain the consumer front-end where 200M users start their journey.

The next major AI value creation will come from applying frontier models to scientific discovery—material science, physics, mathematics—producing breakthroughs that connect with mainstream society more profoundly than chatbots.

Zeta Global is expanding from marketing ROAS optimization into enterprise business intelligence — ingesting sales, inventory, and production data to drive operational decisions — via a co-marketing partnership with Palantir, significantly expanding its TAM beyond advertising.

Empirical cross-lingual transfer matrices reveal asymmetric synergies (script matters more than language family) and model-size-dependent interference, enabling compute-optimal data mixtures for low-resource languages without linguist intuition.

AI enables selling outcomes directly (autopilot) vs tools (copilot). Customer support already $1B+ ARR in autopilot (Sierra). Typical services:software spend ratio is 6:1 — capturing the $6 services budget with software-like margins. But human judgment remains in loop: start with lots of humans/little AI, end with lots of AI/little humans. Not full replacement.

Robinhood is embedding AI (Cortex) into the advisor dashboard to answer client questions like 'what's going on with Tesla stock today,' differentiating its platform and increasing advisor productivity, which could accelerate advisor adoption and asset growth on the platform.