May 23, 2026
AI Super App Update
The first update in the series: Claude adds multi-agent terminal workflows, Anthropic pulls in major talent, Codex gets /goal and AppShots, Google shows Spark, and Cursor starts looking like a super app contender.
This was the first Agent Native update. The goal was simple: make sense of the updates that actually matter across models, platforms, and the new super apps for AI work.
The theme was that every frontier company is trying to own the same surface: one app where agents can chat, code, do knowledge work, use tools, and show interfaces while they work.
The first super app map
| Company | Surface | What to watch |
|---|---|---|
| Anthropic | Claude Desktop and Claude Code | Whether they unify coding and knowledge work. |
| OpenAI | Codex | Long-running goals, plugins, annotations, and app context. |
| Gemini / Spark | Whether they pick one clear agent surface. | |
| Cursor | IDE plus browser | Whether coding expands into broader knowledge work. |
Claude got better for agent workflows
Claude Code added a multi-agent terminal view, which makes it easier to run several research or coding tasks at the same time instead of staying in one linear terminal chat.
Anthropic also pulled in Andrej Karpathy, which was a major talent signal. The broader pattern is that top technical people are moving toward the frontier labs as individual contributors.
Claude Desktop is close to a super app
The desktop app is what makes Claude interesting to me. A true super app should let you chat, code, do knowledge work, connect integrations, run automations, and open an interface on the side.
I still do not love the split between Claude Code and the knowledge-work product, because skills and workflows should feel unified across the whole app.
Codex added the pieces power users need
/goal gives Codex a longer-running objective so it can stay oriented around the outcome instead of just answering one command. That is a big deal for tasks that need planning and persistence.
Plugin sharing, design annotations, and AppShots all make Codex feel less like a coding assistant and more like an operating layer for work. AppShots especially matters because it lets you bring context from whatever app you are using into Codex immediately.
Google had potential but too many surfaces
Google talked about agents everywhere, but the confusing part is the product surface. Is the agent app Gemini, Spark, AI Studio, or Antigravity?
Gemini Spark looked interesting because it points toward an agentic mode inside Gemini, but Google needs one clear place where power users know to go.
Cursor became a serious contender
Cursor's newer composer model felt fast, cheap, and good enough that it was fun to use again. The in-app browser also makes it feel closer to Codex than people might expect.
It still does not do documents, spreadsheets, and presentations the way Codex or Claude can, but it is clearly moving toward the same super app category.