Gartner: Over 40% of agentic AI projects face 2027 cancellation

AI agent development is a discipline of scope control rather than raw technical ambition. It draws a clear line between chatbots, which stop at an answer, and agents, which execute the next operation. It also distinguishes agents from RPA and workflow automation, where a person defines the steps in advance. The recommended stance is to look for the simplest workable solution and add complexity only when needed.

Anthropic's guidance is threaded throughout. Its 'Building effective agents' essay separates predefined workflows from agents, and its MCP standard for connecting AI models to tools was donated to the Linux Foundation's Agentic AI Foundation in December 2025. The ReAct pattern, proposed in a 2022 arXiv paper, is presented as a natural starting point for inquiries and research, where the next move depends on what the previous step returns.

Cost discipline runs alongside the technical advice. Vendor-published reference figures put PoC work in a wide band, from 500,000 to 3,000,000 yen at one firm and 1,500,000 to 5,000,000 yen at another. Production development is quoted at several million to tens of millions of yen, with department-level builds at 3,000,000 to 15,000,000 yen and cross-department or core-system work at 15,000,000 to 50,000,000 yen or more. Monthly API and maintenance costs are noted separately, since human review of AI output remains an internal labor cost that rarely appears in a quote.