A lot of work around AI in 2023 was spent on building picks and shovels. I would know this because t...
TL;DR · AI 摘要
Jerry Liu指出2023年AI基建（如LlamaIndex）聚焦于底层工具建设，而当前重心已转向稳固的智能体抽象与编码能力；真正持久的护城河在于构建支持智能体行动的上下文层，尤其是复杂文档解析与结构化。
核心要点
- 2023年AI工作重心是打造‘铲子和镐’式基础设施，LlamaIndex早期即为此定位。
- 核心智能体抽象已趋稳定，编码智能体正推动更高阶的智能体工程范式。
- 智能体的长期价值壁垒在于上下文层建设——尤其文档解析、工具集成与系统对接能力。
结构提纲
按章节快速跳转。
回顾2023年大量投入集中在底层工具链，LlamaIndex最初使命即为此类‘铲子和镐’建设。
当前核心智能体抽象已固化，编码智能体使开发者能以更高抽象层级构建智能体。
智能体需依赖上下文才能执行任务，其关键组件包括Web搜索、系统记录、SaaS工具等。
专注将复杂PDF等非结构化文档精准解析为高质量语义上下文，并扩展配套工具链。
思维导图
用一张图看清主题之间的关系。
查看大纲文本（无障碍 / 无 JS 友好）
- AI智能体演进与上下文基建
- 历史阶段：2023基建期
- ‘铲子和镐’：LlamaIndex等框架定位
- 聚焦数据接入、索引、检索等底层能力
- 当前阶段：智能体抽象成熟
- 编码智能体降低开发门槛
- Agent as first-class programming construct
- 未来护城河：上下文层
- 文档解析（PDF/扫描件/多模态）
- 工具集成（Web/SaaS/内部系统）
- LlamaIndex：专注文档基础设施
金句 / Highlights
值得收藏与分享的关键句。
A lot of work around AI in 2023 was spent on building picks and shovels.
The core agent abstractions have solidified, and coding agents are letting everyone 'engineer' agents at a higher level of abstraction.
What continues to have a durable moat is building the context layer for agents.
We're building the best-in-class capabilities for agents to parse the most complex documents into clean context.
Today, a lot of that is no longer relevant. The core agent abstractions have solidified, and coding agents https://t.co/7WtTVGzMyJ" / X
Jerry Liu on X: "A lot of work around AI in 2023 was spent on building picks and shovels. I would know this because that was basically the core goal of the original @llama_index framework. Today, a lot of that is no longer relevant. The core agent abstractions have solidified, and coding agents https://t.co/7WtTVGzMyJ" / X
Don’t miss what’s happening
A lot of work around AI in 2023 was spent on building picks and shovels. I would know this because that was basically the core goal of the original
framework. Today, a lot of that is no longer relevant. The core agent abstractions have solidified, and coding agents are letting everyone "engineer" agents at a higher level of abstraction. What continues to have a durable moat is building the context layer for agents. Agents don't inherently have the ability to do things without access to the right tools and data. The context layer can take many forms: - Web search and automation - systems of record - tools to SaaS software Every piece of software needs to become agent native, since agents are the new workers, and they need context to do things. For
today, we're building a slice of that context layer: the document infrastructure for agents. We are building the best-in-class capabilities for agents to parse the most complex documents into clean context, and also building expanded tooling on top of that. Big shoutout to
for a fantastic interview + article!
Quote
LlamaIndex 
@llama_index
·
14h
Our CEO @jerryjliu0 in @VentureBeat , on what's actually changing in the LLM stack: "We've really identified that there's a core set of data that has been locked up in all these file format containers. Ultimately, whether you use OpenAI Codex or Claude Code doesn't really
·
9
4
29
16
Read 9 replies