Agent Development Pattern: Patch Weaknesses, Then Absorb Them into Models
The takeaway is a systems view: agent reliability today depends on scaffolding that future model versions will gradually internalise, so over-engineering external layers carries a hidden tax. Builders designing agent stacks can use this lens to decide what to build versus what to wait for. The trade-off framing is a useful counterpart to tool-heavy frameworks like LangChain when evaluating thin versus thick orchestration.
A widely shared analysis traces agent development as a series of fixes: adding retrieval and data APIs for missing facts, state saving and context management for long tasks, retries and recovery for failed tool calls, permission isolation and approval for risky actions, and testing for substandard results.
These support systems decide whether a model's raw capability works in real scenarios.They also add costs: latency, maintenance, error propagation and coordination overhead.
Model progress then absorbs some of these capabilities, simplifying the structure again, so the boundary between model-internal and external systems keeps shifting.
回看 Agent 的发展,基本是一路补短板: - 缺少当前事实,就加检索和数据接口 - 长任务容易丢信息,就加状态保存和上下文管理 - 工具调用会失败,就加重试、检查和恢复 - 行动可能越权,就加权限隔离和审批 - 结果看着对但不达标,就加测试和验收 这些「支撑技术」决定了模型的原始能力能不能真正落进现实场景。 但它们也带来自己的代价:延迟、维护成本、错误传播、协调开销。 于是形成两股力量:外围不断加支撑系统来兜底,模型进步又把其中一部分能力吸收进去,让结构重新变简单。 哪些留在模型里,哪些留在外面,会一直变。📊 1 ⚡
This report is categorized under AITOP's "How-to" desk, focusing on China's AI industry and domestic foundation model ecosystem. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows. Key entities and technical topics under continuous tracking include Agent, context management, tool use.
Source attribution: curated from 李继刚 (@lijigang_com).