Agentes Mudam Nuvens: Cada Terceira Solicitação à Documentação da Timeweb Cloud Vem da IA

Mikhail Shpakov, head of development at Timeweb Cloud, discusses the impact of AI agents on the cloud services market. The company has implemented an MCP server that allows managing the cloud through assistants (Cursor, Claude Code, TimewebGPT). In two weeks, the endpoint received 112,000 requests, with volume growing 1.5 times weekly. Log analysis showed that out of 237,000 requests to the documentation, 76,000 (32%) were made by bots from AI vendors. ChatGPT-User (requests on behalf of users) made 61,475 calls — 23 times more than Googlebot (2,691). The training crawler GPTBot only accessed it 10 times. Agents more often read pages related to actions (creating a server, managing DNS) rather than settings. Shpakov predicts that provider recommendations will depend on documentation clarity for models, not on advertising. Manual labor costs are decreasing, but responsibility remains: an agent cannot provide SLAs or guarantees. Everything recoverable (web layer, CI) will shift to an "agent + cheap VM" bundle, while clients will continue to entrust data and backups to the provider. Managed services will need to prove their usefulness to agents through API maturity, MCP, and machine readability. A layer of oversight for autonomous systems will emerge: independent monitoring, logs, confirmations, and limits. Shpakov concludes that basic IaaS will become a commodity, "services with a panel at triple the price" will lose their meaning, and the ones that will remain sustainable are those bearing responsibility and providers of agent-friendly building blocks.