Agentic Web Today
“The internet is owned by robots.”
This week, two related agentic web headlines caught my eye. First, Cloudlfare did a deepdive on the nuances of AI crawler behavior and how it changes how the internet caches content. Next, I read that bots are now the the internet's primary traffic (first top story today). Taken together, we see this new, volatile and unpredictable entrant to the internet that is crawling in odd and inefficient ways now owning the lionshare of global traffic. Read both of these.
Read the article →— Jake
▸ Top Stories
Gap says it will launch checkout within Google's Gemini, in an AI first from a major fashion company
# Summary Gap launched in-app checkout functionality within Google's Gemini AI, allowing shoppers to purchase Gap products directly in the conversational AI interface without leaving the platform—the first major fashion retailer to integrate transactional commerce into an AI assistant. This matters for the agentic web because it establishes the infrastructure for AI agents to complete full purchase cycles autonomously: when an agent recommends Gap products in response to user queries (e.g., "what should I wear to a job interview?"), it can now execute the transaction directly via Google Pay, eliminating friction and creating a new distribution channel where AI platforms become shopping destinations rather than just discovery tools.
Can AI responses be influenced? The SEO industry is trying
Companies are creating self-dealing "best of" listicles (like Zendesk ranking itself #1 among service desk platforms) that Google's AI Mode pulls from and summarizes, allowing vendors to manipulate which solutions appear in AI-generated search results. This matters for the agentic web because AI agents will rely on these AI-summarized rankings to make purchasing decisions and recommend tools, creating a new attack surface where SEO manipulation directly influences autonomous agent behavior rather than just human clicks.
Google's broken promise to publishers: the ghosting of 20 web creators
# Summary Google invited 20 independent publishers to a summit in October 2024 claiming to address traffic losses from AI features and algorithm updates, but subsequently ignored their detailed feedback and technical data for 17 months while their businesses collapsed, with executives explicitly stating no recovery to previous traffic levels was possible. This matters for the agentic web because it demonstrates that AI systems consuming web content operate without accountability mechanisms or recourse for creators—when autonomous AI agents scrape, index, and monetize publisher content at scale, there's no contractual obligation, revenue-sharing requirement, or obligation to surface original sources, creating a structural problem where human content creators have no leverage over the algorithms and agents that extract value from their work.
OpenAI raises $122 billion to accelerate the next phase of AI
# Summary OpenAI closed a $122 billion funding round at an $852 billion valuation, now generating $2 billion in monthly revenue with growth four times faster than Internet and mobile-era giants. This capital enables OpenAI to scale compute infrastructure and API access globally, directly empowering AI agents to autonomously interact with enterprise systems and accomplish business tasks through integrated developer platforms and widespread deployment.
Anthropic’s Coding Agent Source Code Exposed in Leak
# Summary **What happened:** Anthropic accidentally exposed over 500,000 lines of source code and 1,000 related files for Claude Code on its NPM registry due to human error, revealing implementation details about how the AI coding agent works and unreleased features. **Why it matters for the agentic web:** Competitors and security researchers can now analyze Claude Code's architecture and integration patterns, potentially enabling faster development of rival agentic systems or identifying exploitable vulnerabilities in how AI agents interact with APIs and development environments.
OpenAI reveals updates to its agentic commerce experience for ChatGPT
# Summary OpenAI shifted its agentic commerce strategy from its Instant Checkout feature to prioritizing product discovery through its Agentic Commerce Protocol (ACP), which major retailers like Target, Sephora, and Walmart have integrated, while moving checkout experiences into merchant-controlled ChatGPT apps. This matters for the agentic web because it establishes a practical protocol enabling AI agents to autonomously browse, compare, and recommend products across multiple retailers without leaving ChatGPT, fundamentally changing how consumers shop online through AI intermediaries.
ChatGPT’s search behavior , LI Articles for AI vis , explainers still work
# Summary ChatGPT breaks user queries into 2-4 smaller searches but cites sources with minimal overlap to Google/Bing rankings—only 27% of cited sources ranked on Google for those queries, and roughly 10% pointed to error pages, indicating ChatGPT often pulls from training data rather than live search results. This matters for the agentic web because AI agents that decompose tasks into sub-queries and source information independently of traditional search rankings will fundamentally disrupt SEO-based traffic models and require websites to optimize for AI training data freshness and citation likelihood rather than search engine visibility.
ALERT: Google Quadruples-down on UCP/Agentic Commerce Introduces: Multi-item Carts, Catalog End Point, Loyalty + More! ..While ChatGPT Focuses on...Discovery?
# Summary Google announced four major updates to its Universal Commerce Protocol (UCP) including multi-item cart capability, enhanced catalog endpoints, loyalty program integration, and a self-service merchant center—establishing standardized infrastructure for AI agents to conduct multi-item transactions across merchants. This matters for the agentic web because multi-item carts enable AI agents to unlock retail economics (free shipping thresholds, bulk discounts) that make agent-driven commerce viable for both merchants and consumers, while establishing protocol standards that prevent fragmentation across different agent builders.
AI Is Eating the Internet and PR Needs a New Playbook
# Summary AI search tools and chatbots have captured traffic that previously flowed to traditional publishers, with major tech publications losing 65M combined monthly visits since 2024 as users get answers directly from AI instead of clicking through to articles. This matters for the agentic web because AI agents will increasingly retrieve information, make decisions, and complete tasks by querying AI systems directly rather than visiting websites, making traditional web traffic and content discovery mechanisms obsolete as the primary path for reaching audiences.
Anthropic's entire agent architecture just leaked, and competitors are taking notes
# Summary Anthropic accidentally leaked 512,000 lines of Claude Code's source code via an npm package, exposing its multi-agent orchestration system, 40+ permission-gated tools, persistent memory, sub-agent parallelization, and other architectural details that competitors can now study. This matters for the agentic web because it reveals the technical blueprint for how Claude orchestrates multiple specialized agents working in parallel with persistent context—a key competitive advantage that now allows other AI companies to understand and potentially replicate the coordination mechanisms needed for complex, multi-step autonomous tasks on the internet.
48% of Google searches now trigger an AI Overview.
# Summary Google AI Overviews now trigger on 48% of searches (up from 31% a year ago), placing AI-generated summaries above the fold and eliminating the assumption that users click through to websites, with LinkedIn data showing B2B awareness traffic down 60% as a result. This matters for the agentic web because AI agents will increasingly consume these overviews and citations rather than visiting source websites, making visibility within AI-generated summaries the new discovery mechanism—meaning companies must shift from traditional SEO to optimization for agent-readable content and structured data, or risk losing pipeline visibility as agents become the primary interface between users and information.
AI Bot Traffic: Why Cloud Costs Are Skyrocketing in 2026
# Summary Generative AI bots now account for over half of global web traffic and are rapidly growing (LLM indexing bots quadrupled their share in eight months), forcing publishers to absorb massive infrastructure costs by bypassing caching protocols and overloading servers while generating zero revenue. This matters for the agentic web because as AI agents become primary consumers of internet content rather than humans, the economic model breaks down—publishers must choose between blocking agents (losing AI visibility) or funding expensive infrastructure for unpaying bot traffic, creating pressure for paid API access that could fragment the internet into exclusive corporate tiers versus open web access.
ChatGPT Search Is Citing Fewer Sites, Data Shows
# Summary ChatGPT Search began citing approximately 20% fewer websites per response after switching to GPT-5.3 Instant as its default model in early March, dropping from an average of 19 unique domains to 15 per response, while maintaining the same crawl depth within each cited domain. This matters for the agentic web because AI agents will now have access to fewer source websites per response, concentrating visibility and traffic among a smaller set of domains—meaning agents delegating tasks through ChatGPT will receive answers drawing from a more limited information pool, reducing discovery of niche or specialized sources and amplifying the dominance of high-authority sites that meet citation thresholds.
Inside Google and Shopify's sprint to build the future of AI shopping
# Summary Google and Shopify launched the Universal Commerce Protocol in January to create standardized infrastructure enabling AI agents across the web to handle shopping interactions and potentially make purchases on behalf of users, partly in response to legal battles like Amazon's injunction against Perplexity's shopping agent. This matters for the agentic web because it establishes agreed-upon protocols for how independent AI agents can access and transact on retail sites—solving the technical and legal fragmentation that currently requires agents to either disguise themselves as humans (risking legal action) or negotiate individual access agreements with each retailer.
▸ Company News
Definity scales agentic AI with enterprise-wide Gemini rollout — Definity Financial rolled out Google Cloud's Gemini Enterprise agentic AI platform enterprise-wide to enhance underwriting, claims processing, and customer servicing across its Canadian P&C operations.
Domo Launches AI Agent Builder and MCP Server to Connect Enterprise Data to the AI Ecosystem | Domo — Domo launched AI Agent Builder, AI Toolkits, a centralized AI Library, and an MCP Server to enable enterprises to build and deploy custom AI agents integrated with governed business data and external AI platforms like Gemini and Claude.
Exclusive | OpenAI Backs New AI Startup Seeking Bot Army Breakthroughs — OpenAI backed Isara, a startup founded by Eddie Zhang and Henry Gasztowtt, to launch software enabling AI agents to communicate and solve complex problems in finance and biotech.
▸ The Feed
Get this in your inbox every Tuesday.