I’ve recently been trying – and (so far) failing! – to automate the booking of group cycling (“spin”) classes at my local leisure centre. Competition for places is intense, so you have to get in quick!
If the class you want is full, you can go on a waiting list to be alerted by email if a place becomes available. That requires manual attention and action too.
But currently there are too many barriers to automation, including:
- The booking software blocks bots (i.e. non-humans)
- My AI agent isn’t comfortable storing my username and password
- The method I was using required my PC to be powered on
And yet…. AI agents promise something simple and powerful: your own smart personal assistant that can do online tasks for you, 24/7, under limits you set. The appeal is obvious in everyday moments many people care about:
-
Buy concert tickets the instant they go on sale.
-
Snag scarce product drops (new phones, GPUs, sneakers) before stock vanishes.
-
Grab time-sensitive appointments (visas, medical slots, government services) when they appear.
-
Rebook disrupted travel while you sleep, then ask for quick approval.
-
Reserve popular restaurants or attractions that release limited daily slots.
- Buying/selling stocks, shares, bitcoin, currency, etc. (if the software you use doesn’t already offer this)
This article isn’t a how-to. It’s a plain-English tour of the hurdles on both sides—your agent and the websites—plus what’s being built to make an agent-friendly web real.
The roadblocks
Traffic spikes and queues. Ticketing and drop sites meter surges with virtual waiting rooms to keep pages up and make access fair. Your agent can’t skip the line; it must queue like everyone else, survive auto-refreshes, and cope with randomized positions. Queue-it
Bot defenses. Many sites use modern checks such as Cloudflare Turnstile, which silently evaluates whether a visitor looks human. If your traffic appears automated, the site can challenge or block it. An agent needs an approved path, not evasive tricks. Cloudflare
Rules and laws. In ticketing, automation done the wrong way can be illegal. The US BOTS Act prohibits circumventing issuer security or purchase limits, and bars selling tickets acquired that way. Platforms also mirror these restrictions in their terms. Federal Trade Commission
Pages built for people, not software. Agents thrive on machine-readable structure. Standards like Schema.org help software understand what’s being sold, but they don’t guarantee a full, agent-safe checkout path. Most flows still assume a person clicking through a human UI.
Green shoots that enable agentic commerce
Big moves are creating official lanes for agents to transact.
-
OpenAI’s Agentic Commerce. Instant Checkout and the Agentic Commerce Protocol let merchants present standardized purchase flows to agents inside ChatGPT, with payments on Stripe. Early coverage confirms the push toward in-chat shopping and fair merchant ranking. This replaces brittle page-clicking with sanctioned, auditable transactions. OpenAI
-
Google’s AP2 (Agent Payments Protocol). Google’s AP2 is an open, shared protocol for secure, compliant payments among agents, merchants, and financial institutions. It supports cards, real-time bank transfers, and stablecoins, aiming to prevent a fragmented ecosystem and give banks clear risk controls. If widely adopted, AP2 provides a common payments “language” agents and merchants can trust. Google Cloud
Standards bodies are moving too. The W3C AI Agent Protocol Community Group is working on open, interoperable mechanisms for agents to identify themselves, request permissions, and interact across the web predictably—laying foundations for an “agentic web.” W3C
In sector verticals, the pattern is similar. For sport and leisure, OpenActive defines consistent availability and an Open Booking API, which reduces guesswork and enables fair, auditable automation where providers adopt it. OpenActive
Why “always on” matters (and the rise of hosted agents)
For agents to be fully effective, they must be available at all hours. A desktop browser may sleep at the worst moment. A smartphone is better, but the ideal is hosted agents in the cloud with secure secrets, durable state, and strong observability.
Recent developments point the way:
-
OpenAI Operator / Computer-Using Agent. Research previews show a hosted agent that can operate a browser and real GUIs on your behalf, demonstrating how agents can act even when your devices are offline. OpenAI
-
AWS Amazon Bedrock AgentCore. A managed platform for long-running, production-grade agents with observability, identity, and support for extended execution windows—built for durable, always-on workloads. Amazon Web Services
-
Microsoft Copilot Studio. A cloud service to build and host custom agents across Microsoft 365 and other channels, with governance and enterprise integrations. Microsoft
-
LangGraph Platform. A hosted runtime focused on long-running, stateful agents, with deployment, checkpointing, retries, and monitoring. LangChain
-
Agentic browsers. Even browsers are becoming agent-aware. Opera’s new Neon markets local, task-executing automations as part of the agentic browsing race. Reuters
Maximizing upside, minimizing downside
-
Mandates and limits. Give agents narrow, revocable permissions—which sites they may act on, spending caps, frequency limits. Protocols from OpenAI and Google move this way with explicit authorization, identity, and audit trails. OpenAI
-
Human-in-the-loop for costly or irreversible actions. A quick phone confirm avoids expensive mistakes and refund hassles.
-
Prefer official lanes over scraping. Choose providers that expose structured data and sanctioned checkout or booking APIs. It’s fairer, more reliable, and less likely to break. Google for Developers
-
Respect defenses and queues. They exist to ensure scale and fairness. Agents should work with them, not against them.
-
Log everything. Hosted platforms with good observability make it easier to audit actions, explain outcomes, and correct mistakes.
The bottom line
Empowering agents isn’t about beating websites. It’s about creating trusted, standardized paths so agents can participate safely and fairly. Virtual waiting rooms, bot checks, and purchase rules evolved to protect users and systems.
New payments rails like Google’s AP2 and OpenAI’s Agentic Commerce, paired with web standards and hosted agent platforms, are building the missing infrastructure.
As more sites adopt structured data, open booking and payment endpoints, and agent-aware APIs, your personal internet assistant will move from brittle page clicking to first-class, accountable participation at any hour.