Descriptions:
Ondra Urban, a developer advocate at Apify, walks through the practical mechanics of enabling AI agents to autonomously pay for tools and services—a capability he calls agentic payments. Presented at AI Dev 26 in San Francisco, the talk opens with a concrete scenario: an agent orchestrating a birthday party by booking a restaurant, buying concert tickets, and purchasing running shoes. The punchline is that while agents can negotiate and communicate, they still can’t close the transaction.
Urban surveys the options developers currently reach for—giving agents a credit card (no spending controls), or pre-subscribing to every API (no dynamic discovery)—and explains why both break down at scale. He then introduces the x402 payment protocol, one of several competing standards alongside L402, AP2, UCP, ACP, and Visa Intelligent Commerce, and explains why x402 is particularly suited to API-level tool interactions. The protocol flow is covered in detail: a payment facilitator verifies available funds, the server completes the work, then settlement occurs—avoiding blockchain latency on the critical path.
The central technical challenge Urban addresses is deploying x402 across Apify’s marketplace of 25,000 tools (called actors), which paid out $1.2 million to its global developer community in a single month. The protocol’s original “exact” pricing scheme works for fixed-cost API calls but breaks for usage-based pricing like LLM token generation or variable-duration database queries. The talk covers how Apify worked around this limitation and what a more flexible dynamic pricing scheme looks like in practice. Developers building agent marketplaces or multi-tool autonomous systems will find this a useful technical reference for the current state of agent payment infrastructure.
📺 Source: DeepLearningAI · Published May 21, 2026
🏷️ Format: Hands On Build
