ATRIUMsearch → argument graph
Article · 2026-06-30 · 6 moments

Impressions from visiting OpenAI, Anthropic, & Cursor

A peek into where software engineering is headed from inside the sector’s leading AI labs. Agents running in the cloud are a major trend, while coding harnesses are spreading beyond the craft ✦ AI generated

Read original ↗
GUEST Katelyn LesseGUEST Peter SteinbergerGUEST OpenAIGUEST Sualeh AsifHOST Gergely Orosz
01
Mechanism

Because cloud agents have no way to surface errors to a human the way local agents do, Cursor has agents give regular 'confession' interviews so problems can be routed to the infrastructure team.

Cursor's Sualeh Asif explains that cloud agents can't 'complain' about errors the way local agents can, so Cursor built a system where agents give periodic 'confessions' that get routed to the infra team.

transcript

Sualeh Asif: Agents in the cloud don't have a way to "complain." With running an agent locally, when it gets warnings or errors, it surfaces them to a human in its response, who instructs it to do X or Y. However, there's no such loop for a long-running task on the cloud!

02
Fact

Claude Managed Agents, a six-month project at Anthropic, is a hosted service for running long-running agents across various cloud providers, and it's a major strategic focus for the company.

Katelyn Lesse, head of engineering for Claude Platform, describes Claude Managed Agents as a major six-month engineering effort to host long-running agents across multiple cloud providers.

transcript

Katelyn Lesse: While there, I met Katelyn Lesse, head of engineering for Claude Platform, who explained that Claude Managed Agents is a large, complex project which her team built over a six-month period. It's a hosted service to execute long-running agents on various cloud providers.

explains mechanism · 1

03
Mechanism◆

Cloud agents are only becoming mainstream now because coding models finally got good enough to run autonomously, agent infrastructure like MCP and skills matured, context windows grew large enough, and cloud providers built up enough GPU capacity.

Gergely Orosz hypothesizes that cloud agents are taking off now due to converging factors: sufficiently capable coding models, mature agent infrastructure (MCP/skills), bigger context windows, and abundant cloud GPU capacity.

transcript

Gergely Orosz: Coding models got 'good enough'. Before Opus 4.5 / GPT-5.4, AI models could not really code autonomously, so running them for long tasks was pointless! Infra for AI coding agents has matured. Ways of giving more context to agents have improved: things like MCP and skills became mainstream and better understood.

gives example · 2supports · 1

04
Fact

Codex's most valuable work now plays out over hours or days rather than minutes, so people should be able to delegate ambitious work without staying tied to the machine where it started.

In its announcement acquiring Ona (formerly Gitpod), OpenAI says Codex's work increasingly spans hours or days, driving the need for persistent cloud environments so users aren't tied to one machine or session.

transcript

OpenAI: As Codex becomes more capable, its most valuable work is unfolding over hours or days, rather than minutes. We believe people should be able to delegate more ambitious work without remaining tied to the machine where it began.

05
Anecdote

Running multiple AI coding agents locally overheats machines and slows systems down, which is why Peter Steinberger built Crabbox to run OpenClaw agents in the cloud instead.

Peter Steinberger got fed up with local OpenClaw agents overheating his CPU and slowing his system, so he built Crabbox to run them remotely in the cloud.

transcript

Peter Steinberger: He talked about how he has gotten really tired of having several OpenClaw agents running on his local machine, which heat up the CPU and slow down his whole system. So, he built Crabbox as a way to run OpenClaw agents in the cloud

06
Claim◆

The real excitement around Claude's Slack integration isn't the Slack mechanism itself, but that it makes it trivial to kick off AI agents that run in the cloud instead of on a local machine, with no setup required.

Gergely Orosz argues that the buzz around Anthropic's Claude Slack integration is really about the shift to easily-launched cloud agents, not the Slack mechanism itself.

transcript

Gergely Orosz: My sense is that the excitement here is less about the Slack integration itself, and more to do with the fact that it's easy to kick off one or more AIs that no longer run on a local machine. You can skip the setup entirely.

explains mechanism · 1

Highlight slides
The Real Story Behind Claude's Slack Integration✦ from: The real excitement around Claude's Slack integration isn't the Slack mechanism itself, but that it makes it trivial to kick off AI agents that run in the cloud instead of on a local machine, with no setup required.Why Cloud Agents Are Going Mainstream Now✦ from: Cloud agents are only becoming mainstream now because coding models finally got good enough to run autonomously, agent infrastructure like MCP and skills matured, context windows grew large enough, and cloud providers built up enough GPU capacity.From Local Machine to Cloud Agents✦ from: The real excitement around Claude's Slack integration isn't the Slack mechanism itself, but that it makes it trivial to kick off AI agents that run in the cloud instead of on a local machine, with no setup required.The Four Converging Factors✦ from: Cloud agents are only becoming mainstream now because coding models finally got good enough to run autonomously, agent infrastructure like MCP and skills matured, context windows grew large enough, and cloud providers built up enough GPU capacity.Before vs. After the Threshold✦ from: Cloud agents are only becoming mainstream now because coding models finally got good enough to run autonomously, agent infrastructure like MCP and skills matured, context windows grew large enough, and cloud providers built up enough GPU capacity.
Related episodes
Article · Latent Space

[AINews] OpenAI reports median internal Codex output tokens grew 56x in Research, 32x in Customer Support, 27x in Engineering, and 13x in Legal since November 2025.

→
Article · Simon Willison's Weblog

Better Models: Worse Tools

→
Article · Latent Space

[AINews] The Field Guide to Fable

→