OpenClaw author practical sharing: Writing code with ACP Codex sub-threads

Headline

OpenClaw author Peter Steinberger codes with ACP Codex subthreads

Summary

Peter Steinberger (an old-timer in the iOS community and the author of OpenClaw) tweeted that he now likes to code using “Acp codex subthreads”. This refers to OpenClaw’s ACP protocol—running OpenAI Codex in a separate chat subthread, so the conversation stays continuously active. OpenClaw is an open-source personal AI assistant that connects to more than 20 platforms like WhatsApp and Telegram. This tweet shows how developers are actually plugging AI agents into everyday work.

Key points:

• Put Codex into its own subthread, separating code generation from everyday chatting so they don’t interfere with each other.
• Supports multiple platforms, so migration costs are low if a team wants to use it.

Analysis

OpenClaw documentation confirms this approach. The ACP framework (see docs.openclaw.ai/tools/acp-agents and docs.openclaw.ai/cli/acp) supports binding external coding tools like Codex to specific threads to run continuously. The benefits are:

• separating code and discussion, so the main conversation doesn’t get spammed by code;
• keeping context available, making it easier to iterate on code in multiple rounds and debug;
• isolating different tasks so team collaboration and review are clearer.

Steinberger’s background helps us understand why he chose this:

• He is the primary maintainer of OpenClaw (repo: github.com/openclaw/openclaw). In 2026, his blog (steipete.me/posts/2026/openclaw) announced that he joined OpenAI, and the project was transferred to an independent foundation;
• From his tweets, it’s clear he prefers proprietary AI models working alongside open-source tools rather than everyone doing their own thing.

In real use, OpenClaw can run on platforms like Discord and Telegram—it can be embedded directly into existing team communication workflows without being locked into a single ecosystem. However, since there are currently no publicly disclosed user numbers or retention data, how big it can get still remains to be seen.

Impact Assessment

• Importance: Medium
• Category: Developer tools, technical insights, open source

Conclusion: For teams and tool builders looking to integrate AI coding into daily collaboration, it might be worth trying sooner; in the short term, it’s not very meaningful for short-line traders; long-term holders and foundations can watch penetration progress, but there’s no need to bet immediately.