Agentic Software

Agentic software is the Vol. 164 枫言枫语 frame for software designed around agents interpreting goals, selecting capabilities, and generating or reshaping the work surface, rather than users pressing fixed buttons in a static application. In Vol. 164 从苹果聊到软件未来：Agentic Software 真的要来了？, Justin Yan and 自立 argue that adding an AI assistant, skills, or tool calls to a traditional product is only an intermediate step.

The concept overlaps with Agent Native Software, but is slightly broader. Agent-native software describes products whose core substrate is the agent; agentic software also covers the transition path where existing SaaS products, operating systems, and platforms expose Atomic Capability Services and Agent-Facing Interfaces so agents can assemble task-specific behavior.

E155.似乎没什么人再提「AI 泡沫论」了 adds a market-structure consequence: if agents operate software through Language User Interface, Model Context Protocol, and AI Skills, incumbent software vendors may lose some GUI entry value and some hidden workflow know-how even if their underlying systems remain useful.

Key Claims

• The interaction model shifts from clear input-output buttons toward ambiguous goals, tool choice, generated interfaces, and iterative human review.
• Traditional GUI still matters for inspection, trust, and presentation, but the core capability should become callable and recombinable by agents.
• Platform review systems such as App Store can become bottlenecks if they assume every app behavior is fixed before review.
• The pattern depends on Context Engineering, Persistent Agent Memory, permissions, and bounded verification; otherwise the agent can confidently assemble the wrong workflow.
• Mature agentic software may serve small groups, personal habits, and vertical workflows that fixed maximum-common-denominator software would not support economically.
• Agentic software can make workflow knowledge more explicit and portable, pressuring software products whose moat is mainly learned UI and embedded process.

Connections

• Agent Native Software — adjacent product frame where the agent is the product substrate.
• Atomic Capability Services — architectural pattern for decomposing software into recombinable abilities.
• Tencent Meeting — source example used to imagine a meeting product rebuilt as capability atoms.
• Headless Software and Agent-Facing Interfaces — interface and callable-surface requirements.
• On-Demand Apps, Generated Work Interfaces, and Token-Driven Software — downstream forms of dynamic software behavior.
• AI Coding Verification, Human Judgment Under AI, and AI Communication Ability — disciplines needed when agents generate software behavior and code.
• Language User Interface, Model Context Protocol, and AI Skills — E155’s language, connector, and process-package layer.