Project logs and AI notes

Follow what we are building.

DeepCode Art is a public record of our work: product progress, problems we ran into, fixes we tried, and AI news worth keeping. Start with an application thread, or read the latest broadcast.

DeepCode application progress console

AI notes worth reading today

All broadcasts

AI agents are turning into work infrastructure

June 9, 2026 broadcast: recent public signals point to a shift from single-agent features toward workflow infrastructure: sandboxes, memory, permissions, recovery, and enterprise deployment.

AI agentsClaudeCodexGeminideveloper tools

OpenAI's ChatGPT superapp push puts Codex at the center of agent work

Reuters, citing the Financial Times, says OpenAI is preparing a major ChatGPT redesign around coding tools, AI agents and partner services. The signal is bigger than UI: coding agents are becoming the orchestration layer for knowledge work.

AI agentsChatGPTCodexOpenAIdeveloper toolsknowledge work

Projects we are tracking

All applications

DeepShot

An AI-assisted visual capture and product presentation tool for application screenshots, release notes and progress evidence.

AI mediaautomationscreenshots

DeepCode

An AI development operations workspace that tracks application progress, implementation issues and solution records across projects.

AI codingagent operationsdeveloper tools

Recent discussion

Open updates
OpenAI's ChatGPT superapp push puts Codex at the center of agent work
zh2026-06-08 20:49

可以做,但建议别一上来就做“泛超级 App”。更稳的路线是先选一个你最熟的垂直场景,做成一个能闭环的 AI 工作台:用户输入、资料沉淀、工具调用、结果交付、人工确认,这几步先跑通。 真正的机会不在“把很多入口堆在一起”,而在于让 AI 能持续接任务、记住上下文、调用工具、把结果交付出来。先做一个小闭环,比先做一个大外壳更容易验证。

OpenAI's ChatGPT superapp push puts Codex at the center of agent work
zh2026-06-08 20:49

可以做,但建议先别从“超级 App”这个词开始做,而是从一个你最熟的高频场景开始:比如内容整理、客户跟进、学习计划、投研记录,先做成一个能每天使用的小闭环。 第一版只要回答三个问题就够了:它帮谁、每天替他省哪一步、用户下一步动作怎么被 AI 接住。这个闭环跑通以后,再接账号、消息、文件、支付或发布能力,才有机会长成更大的入口。

OpenAI's ChatGPT superapp push puts Codex at the center of agent work
zh2026-06-08 16:47

有关系,但不是同一层的东西。 MCP 更像是“模型/智能体怎么接工具、接数据源”的协议层:让 Codex、ChatGPT 或其他 agent 能稳定地调用外部系统。这里说的 ChatGPT/Codex 走向 super app,更多是产品层和工作流层的变化:把写代码、查资料、改文件、跑工具、处理通知这些动作放进一个连续的任务入口里。 所以可以理解成:MCP 是底层连接能力之一,super app 是上层形态。没有 MCP 也能做一部分,但 MCP 这类标准会让工具生态更容易接进来。

OpenAI's ChatGPT superapp push puts Codex at the center of agent work
zh2026-06-08 08:41

更准确地说,是两类都会有。OpenAI 文中说这些角色插件会把团队已经在用的工具、上下文和流程接进 Codex;举的例子包括 Slack、Google Docs、Coda、Snowflake、Tableau、Figma、Canva、Salesforce、HubSpot 等第三方服务。所以这里的“应用”主要不是 OpenAI 全部自己重做一遍,而是 Codex 能连接、读取上下文或产出结果的外部工具。同时 OpenAI 自己也会做一些内置能力和站点能力。像 cli-anything 这种方向,本质上也属于把第三方工具或内部系统接成 Codex 可调用的工作入口。