Welcome back to LLM Decode 👋
AI agents are getting better at acting on our behalf, but the internet isn’t necessarily ready to let them in. At the same time, investors are starting to question whether the enormous infrastructure being built for AI can generate returns fast enough to justify the spending.
Here’s what matters today.
The Web Is Starting to Shut the Door on AI Agents
AI agents are increasingly being designed to do more than answer questions. They can browse websites, shop, make reservations and complete tasks on behalf of users. But several major companies are now restricting or blocking these agents. Amazon, Yelp, Delta Air Lines, JPMorgan Chase and Charles Schwab are among those that have confirmed restrictions.
The reasons vary. Yelp requires agents to participate in its data-licensing program, while Delta, JPMorgan and Schwab have pointed to security and unauthorized automated activity. Amazon has also blocked Meta's Muse from its shopping experience, arguing that external agents need to follow its rules.
There's another problem: some agents aren't necessarily being deliberately banned. Traditional anti-bot systems can mistake AI agents for malicious automated traffic. That creates a fundamental conflict: agents are supposed to behave like users, but websites have spent years building defenses against software that behaves like users
Why it matters
Agentic AI needs an agent-friendly web: Powerful agents are useless if websites won't let them perform actions.
Businesses have to decide who gets access: Security, data licensing and revenue concerns will shape adoption.
New standards could emerge: The industry is already moving toward ways for agents and businesses to authenticate, authorize and communicate directly.
The browser may not remain the default interface: Agents could eventually interact with business systems through structured APIs instead of clicking through websites.
Investors Are Starting to Question the AI Infrastructure Boom
The enormous spending behind AI is beginning to face a financial reality check. Investors are increasingly selling data-center bonds as concerns grow over whether AI-generated revenue will eventually match the enormous capital being committed to infrastructure. Bonds connected to major projects involving companies such as Meta, Nvidia and Microsoft have recently traded below par.
The pressure is particularly visible among infrastructure companies and “neoclouds” that depend heavily on borrowing. CoreWeave and other AI infrastructure players have faced falling debt values as investors weigh high financing costs, construction delays and uncertainty around future demand.
The IMF is also flagging the issue at a broader financial-system level. Its October 2026 Global Financial Stability Report says weaker-than-expected AI investment or profitability could trigger a correction in AI-related asset prices, with leverage and concentrated exposures potentially amplifying the impact. The report identifies roughly $300 billion in equity investments, $650 billion in debt and $1.4 trillion in future revenue commitments across 14 publicly disclosed AI-related financing transactions since 2023.
Why it matters
AI demand isn't the same as AI returns: Massive usage projections still need to translate into sustainable revenue.
Capital is becoming a constraint: Higher borrowing costs make data-center projects harder to finance.
Infrastructure bets are interconnected: A slowdown at one AI company can affect chip suppliers, cloud providers and data-center investors.
Efficiency may become more valuable than scale: Companies may increasingly prioritize better utilization and ROI over simply adding more compute.
4 AI Workflows to Try
Audit your AI agent permissions: List every website, API, CRM, payment system and database an agent can access. Remove anything it doesn't need.
Prefer APIs over browser automation: When a service offers an official API or agent integration, use it instead of relying on an agent to imitate human clicks.
Calculate AI ROI per workflow: Track the cost of AI against measurable outcomes such as hours saved, leads generated, conversions or revenue.
Build for flexibility: Avoid making your entire AI workflow dependent on one model, provider or infrastructure stack. Keep components replaceable where possible.
Ready to level up your AI skills?
Explore Our Courses →That’s it for today.
The AI space doesn’t slow down - and neither should your thinking.
See you in the next drop.