# Agentic Commerce & Intermediation Disruption — Research Summary **Date**: 2026-02-23 **Source Paper**: CitriniResearch, "2028 Global Intelligence Crisis" (Feb 2026) **Analyst**: BigPic Capital Research --- ## Executive Summary The CitriniResearch paper's thesis on agentic commerce destroying friction-based business models is **directionally correct and already showing early real-world validation** — but the 2028 timeline for full-scale disruption is aggressive. The infrastructure is being built right now (protocols, payment rails, merchant integrations), and the first casualties among intermediaries have already appeared (UK price comparison sites). However, regulatory moats, trust requirements, and consumer adoption curves will likely slow the disruption beyond what the paper implies. --- ## Claim-by-Claim Assessment ### 1. Consumer AI Agents Running 24/7, Price-Matching, Managing Subscriptions **Evidence Level: STRONG — Infrastructure is live, adoption is early but accelerating.** - OpenAI launched "Buy it in ChatGPT" (Instant Checkout) in September 2025, expanded to all U.S. users including free tier in February 2026. ChatGPT now serves 900 million weekly users, with an estimated 2% of queries being shopping-related (~50 million shopping queries/day). - Perplexity launched Instant Buy with PayPal in November 2025, integrating merchants like Wayfair and Abercrombie & Fitch. - OpenAI's Agentic Commerce Protocol (ACP), built with Stripe, is open-source (Apache 2.0) and enables programmatic purchase flows within chat sessions. - Morgan Stanley projects nearly half of online shoppers will use AI shopping agents by 2030, accounting for ~25% of spending. - Shopify reported orders attributed to AI searches grew 11x since January 2025. - During the 2025 holiday season, global e-commerce traffic from AI chatbots doubled YoY, with AI credited for driving 20% of all retail sales and $262B in revenue. **Subscription management**: Consumers average $133/month in subscriptions (~$1,600/year) with ~$127/year wasted on unused ones. AI subscription management tools exist but face adoption friction — alert fatigue and decision paralysis remain challenges. This is more of a 2027-2028 maturity story than 2026. **Assessment**: The paper's vision of AI agents managing purchases is not speculative — it is live today. The gap is between protocol availability and mass consumer behavioral change. The 24/7 optimization angle is plausible on the paper's timeline for early adopters, but broad adoption will lag. --- ### 2. Travel Booking Platforms as Early Casualties **Evidence Level: STRONG — OTAs themselves acknowledge the threat.** - Expedia's 2025 10-K filing explicitly calls out AI agents as a major threat, not merely competitive pressure. Expedia outlined ~$700M in strategic reinvestments for 2026 (with ~$300M net EBITDA impact) focused on generative AI development. - Both Booking.com and Expedia partnered with OpenAI at its October 2025 DevDay, gaining access to ChatGPT's 800M+ weekly users — a defensive move acknowledging the threat of disintermediation. - The "Software Apocalypse" narrative is taken seriously: if a user can say "Book me a trip to Maui for under $5,000" and an AI handles it end-to-end, the dedicated travel app becomes redundant. - Counter-evidence: Booking.com reported Q4 2025 gross bookings of $40.2B (above expectations) with 16% revenue growth and 9% room-night growth. The incumbents are not dying yet — they are adapting. **Assessment**: Travel is correctly identified as an early stress sector. However, the paper may understate incumbent advantages: scale, supplier relationships, loyalty programs, and payment infrastructure. The more likely near-term outcome is that OTAs become AI-integrated platforms (Expedia/Booking as infrastructure layers for agents) rather than being bypassed entirely. True disintermediation requires agents to book directly with hotels/airlines, which requires payment rails, dispute resolution, and trust — areas where incumbents still lead. --- ### 3. Insurance Renewal Models Disrupted (15-20% Passive Renewal Premiums at Risk) **Evidence Level: STRONG — First market-visible casualties already occurred.** - In February 2026, UK price comparison intermediaries were hit hard: Mony Group (MoneySuperMarket) shares crashed 13% to a 13-year low; Future (GoCompare) dropped 3.2%. The trigger was Insurify launching the insurance industry's first ChatGPT app, and Spanish startup Tuio launching a home insurance comparison app on ChatGPT. - MoneySuperMarket's parent lost 144 million GBP in market value in a single session. - The STOXX 600 Insurance index fell nearly 2%, becoming Europe's worst-performing sector in mid-February 2026. - 35%+ of insurers are projected to deploy AI agents across core functions by late 2026, with processing time cuts of up to 70%. - M&A deal value involving AI in insurance surged 328% in 2025; deal volumes up 125%. **Assessment**: This is the strongest real-world validation of the paper's thesis. The passive renewal model — where insurers profit from consumer inertia — is directly threatened by AI agents that can re-shop coverage annually with zero effort. The 15-20% passive renewal premium compression is plausible on the paper's timeline, especially in markets like the UK where comparison shopping is already culturally embedded. U.S. adoption may lag due to regulatory complexity and state-by-state insurance markets. --- ### 4. Real Estate Commissions Compressed from 3% to Under 1% **Evidence Level: MODERATE — Directional pressure exists, but 1% by 2028 is aggressive.** - The NAR settlement (2024) already decoupled buyer/seller agent commissions, creating structural pressure. - AI can automate up to 70% of tasks performed by junior real estate staff, per investor estimates. - A sell-off of real estate service companies occurred in February 2026 on AI disruption fears. - Zillow integrated with ChatGPT in October 2025, forcing industry reckoning. - However, critical infrastructure gaps remain: MLSs do not operate MCP (Model Context Protocol) servers allowing AI systems to securely access listing data in real time. Without this, AI agents cannot fully execute on real estate transactions. - The industry is debating data access rules rather than building open infrastructure — a significant adoption blocker. **Assessment**: Commission compression is real and was already underway before AI (via Redfin, the NAR settlement, etc.). AI will accelerate the trend. However, reaching sub-1% by 2028 requires solving MLS data access, regulatory compliance in 50 states, and the trust problem for the largest financial transaction most consumers ever make. A more realistic estimate might be 1.5-2% by 2028 for the median transaction, with sub-1% possible in tech-forward markets. --- ### 5. DoorDash/Uber Eats Delivery Moats Destroyed **Evidence Level: MODERATE-WEAK — Incumbents are adapting faster than the thesis implies.** - Both DoorDash and Uber Eats have proactively integrated with ChatGPT, allowing users to browse via the chatbot (though checkout still occurs in-app). - DoorDash CEO Tony Xu's counter-argument is substantive: AI agents may handle the first steps (discovery, ordering), but DoorDash orchestrates the physical delivery logistics — which is the hard part. - DoorDash reported record DashPass signups in 2025. Uber Eats delivery bookings rose 26% YoY in Q4. - The multi-app aggregator/dashboard threat exists (ChowNow advocates against 30% commission models) but hasn't materially dented market share yet. - AI coding tools lowering barriers to entry for new delivery apps is speculative — the barrier isn't coding, it's building a two-sided marketplace of restaurants and drivers at scale. **Assessment**: The paper overstates the near-term threat to food delivery. The physical logistics moat (driver networks, restaurant relationships, real-time routing) is far harder to disintermediate than pure information intermediaries like insurance comparison sites. The "checking 20+ platforms" argument is valid for price comparison but less relevant when the core value is same-hour physical delivery. This is more of a margin compression story (commissions declining from ~30% toward ~15-20%) than an existential threat on a 2028 timeline. --- ### 6. Payment Rails Disruption: Stablecoins Bypassing 2-3% Interchange Fees **Evidence Level: MODERATE — Infrastructure is growing fast, but adoption remains niche for consumer commerce.** Key data points on stablecoin infrastructure: - Stablecoin circulation is projected to exceed $1 trillion by late 2026. - Actual stablecoin payments volume is ~$390B/year, but that is only ~0.02% of global payments volume (McKinsey). - B2B stablecoin payments surged 733% YoY in 2025. - Stablecoin payments grew from under $100M monthly in early 2023 to over $6B monthly by mid-2025. - Visa's stablecoin-linked card spend reached $4.5B annualized by January 2026 (460% YoY growth). Agent-to-agent payment infrastructure: - Google launched Agent Payments Protocol (AP2) in September 2025 — a universal standard for AI agent payments, with Solana as the primary settlement layer and Base (Coinbase L2) as an alternative. - OpenAI/Stripe's ACP and Google's Universal Commerce Protocol (UCP, announced January 2026) are competing standards. - Transaction flow: agent wallet signs payment on-chain (USDC on Base or Solana), verified in ~2 seconds. **Assessment**: The infrastructure for agents to route payments via stablecoins exists and is being standardized. However, the paper's implied scenario of mass consumer abandonment of card rails is premature. Key obstacles: (1) Consumer protection — card networks offer chargebacks, fraud protection, and dispute resolution that stablecoins lack; (2) Regulatory uncertainty — U.S. stablecoin legislation is still pending; (3) Visa and Mastercard are integrating stablecoins rather than being bypassed (Visa settling $4.5B in stablecoins); (4) OpenAI charges merchants 4% on ChatGPT Instant Checkout — *more* than interchange. The most likely near-term path is card networks absorbing stablecoin rails, not being replaced by them. --- ### 7. "Habitual Intermediation" as a Moat Category That Ceases to Exist **Evidence Level: STRONG CONCEPTUALLY — Real-world evidence is emerging.** The concept that consumer habits (defaulting to the same platform, not bothering to comparison shop) constitute a moat that AI agents eliminate is well-supported: - The MoneySuperMarket crash demonstrates the market pricing this risk. - McKinsey projects agentic commerce could reach $1.7 trillion by 2030, up from $136B in 2025. - PayPal CEO Alex Chriss called agentic commerce the biggest transformation since e-commerce's advent. - The "attention tax" that intermediaries charge (making it too tedious for humans to optimize) is exactly what AI agents are designed to eliminate. **Assessment**: This is the paper's strongest conceptual contribution. The distinction between "structural moats" (physical logistics, regulatory licenses, network effects) and "habitual moats" (consumer inertia, complexity barriers, attention costs) is analytically useful. Businesses relying primarily on habitual intermediation are genuinely at risk. The question is timeline — behavioral change even with AI assistance takes time, and many consumers may not adopt AI agents for years. --- ### 8. Credit Card Companies Facing Combined Headwinds **Evidence Level: MODERATE — Threats are real but incumbents are actively adapting.** Headwinds identified: - AI agents could target the 2-3% interchange fee as a cost to eliminate (per Citrini's own modeling, covered by Benzinga). - Stablecoin payment rails offer near-zero transaction costs. - Workforce reduction (the paper's broader thesis) would shrink the card-spending consumer base. Counter-evidence (incumbent adaptation): - Visa launched the Trusted Agent Protocol in October 2025 — an open framework for safe agent-driven checkout, with 10+ partners. - Visa is positioning AI and tokenization at the heart of its 2026 strategy. - Visa launched USDC settlement in the U.S. in 2025, settling via Solana — integrating stablecoins rather than fighting them. - Mastercard partnered with MoonPay for stablecoin-funded digital wallets linked to Mastercard accounts. - Both networks are building the rails that agents will use, potentially maintaining their position as infrastructure. **Assessment**: The paper correctly identifies the theoretical threat vector. However, it underestimates Visa/Mastercard's ability to adapt. These companies have successfully navigated every payments transition for decades (cash to card, card to digital, digital to mobile). Their moat is not just interchange — it is fraud prevention, dispute resolution, merchant acceptance, regulatory compliance, and global reach. The more likely outcome by 2028 is margin compression (interchange declining 0.3-0.5 percentage points) rather than existential threat. The workforce reduction angle (fewer consumers = fewer transactions) is a separate macroeconomic thesis that depends on the broader labor displacement claims. --- ## Key Data Points Summary | Metric | Value | Source | |--------|-------|--------| | ChatGPT weekly active users | 900M | OpenAI (Feb 2026) | | ChatGPT shopping queries/day | ~50M (est. 2% of sessions) | OpenAI | | Shopify AI-attributed order growth | 11x since Jan 2025 | Shopify | | Stablecoin payments volume (annual) | ~$390B (0.02% of global payments) | McKinsey | | B2B stablecoin payments growth | +733% YoY (2025) | Industry reports | | Visa stablecoin settlement volume | $4.5B annualized (Jan 2026) | Visa | | Mony Group (MoneySuperMarket) stock decline | -13% in one session (Feb 2026) | Market data | | AI-driven commerce projected value (2030) | $1.7 trillion | Consulting estimate | | AI-driven commerce current value (2025) | $136B | Consulting estimate | | Morgan Stanley AI shopping agent adoption (2030) | ~50% of online shoppers, ~25% of spend | Morgan Stanley | | Insurance AI M&A deal value growth (2025) | +328% | Industry data | | Booking.com AI reinvestment budget (2026) | ~$700M total, ~$300M net EBITDA impact | Booking 10-K | | OpenAI ChatGPT checkout merchant fee | 4% per transaction | OpenAI | --- ## Overall Plausibility Assessment **Rating: 7/10 — Directionally correct, timeline is aggressive.** The paper's core insight — that AI agents eliminate the friction that intermediaries monetize — is analytically sound and already manifesting in markets. The strongest evidence is in: 1. **Insurance comparison** (MoneySuperMarket crash, ChatGPT insurance apps) 2. **Travel booking** (OTAs explicitly flagging AI agents as existential risk in SEC filings) 3. **General commerce** (protocols live, merchant integration accelerating, usage data showing traction) The thesis is weakest for: 1. **Food delivery** (physical logistics moat is underestimated) 2. **Payment rails** (Visa/Mastercard are integrating stablecoins, not being bypassed) 3. **Real estate** (MLS data access barriers, regulatory fragmentation, trust requirements for high-value transactions) The 2028 timeline for "full disruption" is aggressive. A more realistic timeline: - **2026-2027**: Infrastructure buildout complete, early adopter wave (10-15% of digitally native consumers regularly using AI agents for commerce) - **2028-2029**: Mass-market adoption begins for simple intermediation (insurance, travel, subscription management) - **2030+**: Complex intermediation disruption (real estate, financial advisory, enterprise B2B) --- ## Early Warning Indicators for Investors 1. **ChatGPT/Perplexity shopping conversion rates** — Watch for OpenAI disclosing actual purchase volumes, not just query counts. 2. **Stablecoin payment volume growth** — Track McKinsey's filtered "actual payments" metric, not raw on-chain volume. 3. **UK price comparison site revenues** — MoneySuperMarket and GoCompare are the canary in the coal mine for habitual intermediation disruption. 4. **OTA customer acquisition costs** — Rising CAC would indicate agents are fragmenting the funnel. 5. **Visa/Mastercard payment volume growth rates** — Deceleration relative to e-commerce growth would signal early bypass. 6. **Insurance renewal retention rates** — Industry data on automatic renewal rates declining would validate the passive premium thesis. 7. **Stripe/PayPal agentic commerce volume disclosures** — Both companies are positioning as agent commerce infrastructure; their volumes will be leading indicators. 8. **Google's UCP adoption** — Google Search AI Mode + Gemini integration with Universal Commerce Protocol could be the mass-market catalyst. --- ## Counter-Arguments and Reasons the Thesis Could Be Wrong ### 1. Trust and Consumer Behavior Are Sticky Most consumers have never used an AI agent to make a purchase. Behavioral change is slow even when technology is available. The paper assumes rational optimization, but consumers are not rational — brand loyalty, habit, and comfort matter. Many people will continue using Expedia or their existing insurance broker for years, even if an AI agent is theoretically better. ### 2. Incumbents Are Adapting, Not Dying Visa, Mastercard, Booking.com, Expedia, and PayPal are all building agent commerce infrastructure. They have the capital, merchant relationships, and regulatory expertise to remain central. The most likely outcome may be that intermediaries evolve into "agent infrastructure providers" — the AI uses Booking.com's inventory and Visa's payment rails, and the intermediary takes a smaller but still substantial cut. ### 3. Fraud, Liability, and Consumer Protection AI agents making autonomous purchases create new fraud vectors. Who is liable when an AI agent buys the wrong thing? What happens when an agent is tricked by a fake merchant? Card networks provide chargebacks and dispute resolution. Stablecoin transactions are generally irreversible. Until consumer protection frameworks for agent commerce are established, adoption will be limited by trust. ### 4. Regulatory Barriers Are Real Insurance is regulated state-by-state in the U.S. Real estate has licensing requirements. Financial services have KYC/AML obligations. These create friction that AI agents cannot simply bypass — they require legal and regulatory infrastructure that takes years to build. ### 5. The "Agent Tax" May Replace the Intermediary Tax OpenAI charges 4% on ChatGPT Instant Checkout — more than Visa's interchange. If AI platforms become the new gatekeepers, the intermediation fee may shift rather than disappear. Merchants may trade one intermediary (Expedia) for another (OpenAI/ChatGPT). The net economic benefit to consumers could be smaller than the paper implies. ### 6. Physical Logistics Cannot Be Disintermediated by Software The paper's claim about DoorDash/Uber Eats conflates information intermediation with physical logistics. You can compare prices with AI, but someone still has to pick up the food and drive it to your house. The logistics moat is real and requires capital-intensive infrastructure, not just software. ### 7. Network Effects in Two-Sided Marketplaces Platforms like DoorDash, Uber, and Airbnb have strong network effects: more users attract more suppliers, which attracts more users. An AI agent can compare across platforms, but it cannot create a new platform's supply-side network overnight. The barrier to competing with DoorDash is not the code — it is the millions of drivers and hundreds of thousands of restaurants. ### 8. Income and Digital Divide AI agent commerce assumes smartphone ownership, AI service subscriptions, digital literacy, and comfort with autonomous purchasing. A significant portion of consumers (older demographics, lower income, less tech-savvy) will not adopt agent commerce for years, limiting the addressable market for disruption. --- ## Conclusion The CitriniResearch paper identifies a real and accelerating trend. The infrastructure for agentic commerce is being built by the largest technology and payments companies in the world. The first market-visible casualties (UK price comparison sites) have already appeared. However, the paper's implicit timeline of comprehensive intermediation destruction by 2028 underestimates incumbent adaptation, regulatory friction, consumer behavioral inertia, and the distinction between information intermediation (easily disrupted) and physical/regulatory intermediation (much harder to disrupt). **For investors, the key framework is**: Rank intermediary businesses by the ratio of "friction value" (what they charge for reducing complexity) to "structural value" (logistics, regulation, trust, network effects). Businesses high on friction and low on structural value are at immediate risk. Businesses with deep structural moats have time to adapt. --- *Disclaimer: This is educational research only — not investment advice. Markets involve risk.*