Automating Web Workflows with LLMs
summary
Apify demonstrates an LLM-driven web-agent that uses screenshot with set of marks plus HTML or accessibility trees, exposes predefined actions (go to URL, click element, fill form) instead of generating code, and integrates via Apify SDKs for Python/JavaScript. Live demos extract pricing pages into structured fields (plan name, monthly price, RAM), complete flows in ~25 seconds, and cost cents per run, but stability issues remain: context window limits, stochastic outputs, cookie modals, captchas and Cloudflare block some sites.
AI-written summary · about
chapters (8)
- 0:00Automating web workflows
- 5:27Manual automation fails
- 7:58LLM automation agent
- 16:33Web automation challenges
- 22:40Form filling demo
- 25:52Data extraction demo
- 27:17AI web agent demo
- 38:13Question and answer session
related talks
01h 10mTeach an LLM to Navigate the Web
01h 28mBuilding LLM Agents Made Simple a
01mSpeed Up Your Code With LLM's!
41mBuilding LLM Agents – Challenges and Learnings
19mEveryday AI: Leveraging LLMs for Simple...
21mBuilding Agents (the hard parts!)
22mThe agent-ready web: Simplify user actions with WebMCP
35mBuilding LLM-based Agents