Agentic Web Today
“GPT5.4's new bias toward brand websites”
In a world where AI is taking an increasingly prominent role in telling a brands story via ChatGPT, Gemini etc., its refreshing to see that the newest model from OpenAI is putting a brand's website front and center for brand-relevant prompts. Analysis from both Profound and Writesonice found that GPT-5.4 fires queries directly at brand domains before it ever looks at third-party sources, meaning your own website is now the baseline the model measures everything else against. For brand-direct prompts, half the citations come from the brand's own domain and 71% of those queries fire in the first two positions of the search sequence.
Read the article →— Jake
▸ Top Stories
Online bot traffic will exceed human traffic by 2027, Cloudflare CEO says
# Summary Cloudflare CEO Matthew Prince stated that AI bot traffic will surpass human internet traffic by 2027, driven by generative AI agents visiting thousands of websites per task compared to the five sites humans typically visit. This matters for the agentic web because it requires building new infrastructure—like disposable sandboxes that spawn and tear down AI agents on demand—to handle millions of concurrent agent executions per second without overwhelming internet capacity.
Millions of merchants can sell in AI chats
# Summary **What happened:** Shopify launched Agentic Storefronts, enabling millions of merchants to sell directly through AI chatbots (ChatGPT, Microsoft Copilot, Google Search's AI Mode, and Gemini) with products discoverable by default and orders flowing seamlessly into merchant dashboards. **Why it matters for the agentic web:** AI agents can now autonomously browse, compare, and purchase products across a massive merchant catalog without leaving the chat interface, establishing e-commerce as a native capability of AI agents rather than requiring them to navigate external storefronts—fundamentally shifting how AI systems interact with commercial services online.
FLASHChatGPT Launches New Discovery Features
# Summary OpenAI launched new discovery features for ChatGPT that enable product comparison and shopping functionality, with Sephora among early partners demonstrating the capabilities at Shoptalk. For the agentic web, this matters because AI agents can now directly query product catalogs, compare items, and execute commerce transactions without requiring separate tool integrations—enabling autonomous shopping agents to operate natively within ChatGPT's interface.
OpenAI's first crack at online shopping stumbled. It's preparing for the next wave
# Summary OpenAI abandoned its Instant Checkout feature that allowed direct purchases within ChatGPT, pivoting instead to dedicated retail apps that redirect users to retailers' websites, after underestimating the complexity of handling transactions and losing ground to Google's more functional shopping agent. This pivot demonstrates that AI agents still struggle with critical e-commerce tasks like real-time inventory management, multi-item cart functionality, and secure payment processing—capabilities essential before autonomous agents can reliably execute transactions across the internet.
AI-driven traffic jumps 187%
# Summary **What happened:** AI-driven web traffic surged 187% year-over-year in 2025, with agentic systems and AI browsers expanding nearly 8,000%, while agentic activity increasingly targeted product pages, accounts, and checkout flows beyond traditional crawling and scraping. **Why it matters for the agentic web:** As AI agents now actively transact across the web rather than passively index it, websites must implement new detection methods (like Google's new user agent identifiers), structured data standards, and access controls to manage agent interactions with sensitive user flows and prevent abuse while capturing legitimate agent-driven commerce opportunities.
AI agents are already driving 10% of revenue for some brands. Is yours invisible to them? | Fortune
AI agents are now autonomously making purchase decisions for consumers, with some brands already attributing 10% of their revenue to agentic commerce channels, fundamentally bypassing traditional discovery mechanisms like search rankings and ads. For the agentic web to work, brands must optimize their content and product data specifically for AI agent parsing through Answer Engine Optimization and Agent Experience (AX) design, since only 12% of URLs cited by AI tools overlap with Google's top 10 results, making traditional SEO insufficient.
Agentic shopping, outcome-driven AI, and creator commerce: Takeaways from Shoptalk 2026
# Summary **What happened:** At Shoptalk 2026, retailers demonstrated three emerging trends—agentic AI agents handling shopping journeys, shift toward outcome-driven AI applications (like Gap's fit-sizing tools and Meta's post-click product details), and deeper integration of creator content with commerce—though infrastructure development is outpacing actual consumer adoption and transaction growth. **Why it matters for the agentic web:** AI agents need structured, platform-independent product data and seamless integration points across content platforms to autonomously complete shopping tasks, but the gap between technical capability and proven monetization means businesses must validate agent-driven conversion channels with data before reallocating budget away from traditional search and retail media.
OpenAI announces ChatGPT agent for web browsing
# Summary OpenAI launched ChatGPT agent, a web browsing tool that combines the autonomous clicking/typing capabilities of its Operator agent with the analytical reasoning of its Deep Research tool, enabling the AI to both interact with websites and perform complex analysis in a single system. This matters for the agentic web because it represents a critical convergence—AI agents can now not only search and analyze information but also execute actions on authenticated websites, directly competing with Anthropic's computer use and Perplexity's tools while establishing web interaction as a key competitive arena for major AI labs.
MCP's 2026 roadmap makes enterprise readiness a top priority
# Summary Anthropic's Model Context Protocol has identified enterprise deployment as its 2026 priority, with the roadmap targeting four critical gaps: standardized audit trails, dynamic authentication, gateway behavior standards, and portable configuration—issues currently blocking production deployments at scale. For the agentic web, this matters because agents executing tasks across distributed systems require verifiable audit logs and secure authentication to operate in enterprise environments; without these, AI agents cannot reliably handle sensitive operations or meet compliance requirements, limiting their real-world deployment at organizations that control significant resources and data.
Google Gemini Sends More Traffic To Sites Than Perplexity: Report
# Summary Google's Gemini AI more than doubled its website referral traffic between November and January 2025, surpassing Perplexity by 29% globally after the rollout of Gemini 3 models, though ChatGPT still dominates with roughly 8x more traffic than Gemini. For the agentic web, this signals that Google's integration of advanced AI models into search and its product ecosystem is beginning to materially redirect user queries away from specialized AI chatbots, meaning web developers and content creators must now optimize for multiple AI agents' crawling and citation patterns rather than relying primarily on traditional search or single chatbot platforms.
World and Coinbase Turn AI Agents Into Trusted Shoppers
# Summary World and Coinbase launched AgentKit, a developer toolkit that links verified World IDs (biometric human identity proofs) to AI agents, allowing these "human-backed agents" to make authenticated transactions and access resources that normally block automated traffic by combining World ID verification with Coinbase's x402 payment protocol. This matters for the agentic web because it solves the core problem of distinguishing productive AI agents from malicious bots—agents can now prove a real human authorized them and pay for access, transforming them from unwanted automated traffic into legitimate economic participants that websites will allow to book reservations, check prices, and conduct commerce on users' behalf.
How to build AI agents with Google-managed MCP servers
**What happened:** Google released managed MCP (Model Context Protocol) servers that allow AI agents to securely access Google and Google Cloud services like Maps, BigQuery, and Cloud Run without requiring developers to manage their own infrastructure. **Why it matters for the agentic web:** AI agents need reliable, auditable access to real-world data and enterprise tools to function effectively in production; Google's managed servers solve this by providing built-in security (Cloud IAM, VPC-SC), centralized logging for compliance, and unified discovery of available APIs, enabling agents to interact with the internet and enterprise systems at scale without single points of failure.
Agentic AI Just Went Mainstream: The Week That Changed Everything (March 16–23, 2026)
# Summary **What happened:** Shopify's Agentic Storefronts rolled out by default to all merchants in ChatGPT, Perplexity, and Microsoft Copilot simultaneously, while major enterprise platforms (Hexaware's Agentverse, Alibaba's Wukong, and Snowflake's SnowWork) launched production-ready AI agent systems for business functions during March 16–23, 2026. **Why it matters for the agentic web:** AI agents now have direct, default access to real commerce and enterprise data systems without manual integration, making the internet immediately transactional and actionable for autonomous agents—transforming AI from conversation partners into entities that can discover products, execute purchases, and complete business workflows independently across the web.
Agents, meet the Figma canvas
Figma released an MCP server that enables AI agents to directly read and write to Figma files, allowing agents like Claude to generate and modify design assets while adhering to established design systems and brand standards through a new "use_figma" tool and skills-based instructions. This matters for the agentic web because it gives AI agents access to design context and constraints, enabling them to create coherent, brand-aligned user interfaces across web products rather than producing generic designs, and allows seamless collaboration between coding agents and design systems where agents can iterate on both code and canvas simultaneously.
Exclusive | OpenAI Plans Launch of Desktop ‘Superapp’ to Refocus, Simplify User Experience
OpenAI is consolidating ChatGPT, Codex, and its browser into a single desktop "superapp" to streamline user experience while targeting engineering and business customers. This integration creates a unified entry point for AI agents to access multiple tools and services simultaneously, reducing friction for agentic workflows that require coordinated use of chat, code generation, and web browsing capabilities.
▸ Company News
Most systems today still infer human behavior from thin proxies like IP traffic and outdated enrichment. — Nyne.ai raised $5.3M in seed funding to build agent-first identity and intent infrastructure that transforms fragmented public signals into developer-ready primitives for AI agents.
Accenture Invests in DaVinci Commerce to Advance Agentic AI-Led Shopping — Accenture invested in DaVinci Commerce to expand agentic AI-led shopping capabilities across discovery, merchandising, checkout, fulfillment, and loyalty for enterprise clients.
Constant AI Launches Nia: The First AI Agent That Completes a Skip-a-Pay — From Call to Core — Constant AI launched Nia, an agentic AI skip-a-pay agent available across voice, digital banking, chat, and text, to automate end-to-end skip-a-pay requests for credit unions without staff involvement, built on infrastructure that has already processed over 1.5 million automated transactions.
Coinbase Dives Into AI Agent Payments — Coinbase is pursuing a stablecoin partnership with Cloudflare to enable AI agents to transact in cryptocurrency.
Claude can now control your Mac and run tasks remotely — Anthropic enabled Claude to control Mac computers directly through clicking, typing, and app navigation, paired with Dispatch for remote task execution, while introducing auto mode in Claude Code with classifier-based permissioning for safe autonomous shell commands.
OpenAI backs a nine-month-old startup building swarms of AI agents at a $650 million valuation — Isara raised $94 million at a $650 million valuation to build software coordinating thousands of AI agents on complex analytical tasks like financial forecasting and geopolitical analysis.
GitHub - vercel-labs/agent-browser: Browser automation CLI for AI agents — Vercel Labs launched agent-browser, a native Rust CLI for headless browser automation designed to enable AI agents to interact with web interfaces.
▸ The Feed
Get this in your inbox every Tuesday.