跨会话传消息后，Codex 和 Claude 如何重构 vibe coding 工作流
Key Highlights
What frustrates people most about vibe coding has never been the inability to write code, but rather having to start over the moment the context breaks. The author of the WeChat public account Carl's AI Watts recently practiced a new idea: after Codex and Claude added the cross-session messaging feature, he used a structure of main conversation plus branch conversations to completely rebuild the vibe coding workflow, saving a large amount of handoff documents and git backups. The shift is subtle but important, because it changes the basic unit of work from a single disposable chat into a small, persistent organization that actually remembers what it was doing across sessions and can pick up where it left off. It is the kind of change that does not show up in any benchmark but quietly changes daily habits, because the friction of starting over simply disappears from the developer's day.
Specific Capabilities and What Happened
The core of this new feature is simple: when starting a new conversation, you can pick which messages from all previous conversations are available, rather than beginning from a blank slate every time. Based on this, the author designed a workflow where the main conversation is responsible for holding the project goals and key decisions, while branch conversations independently explore new ideas and experiment; once the branch exploration is done, the main conversation reads the complete record and absorbs the conclusions. In this way, no matter how far the exploration goes off track, it will not pollute the context of the main line. The result is a much cleaner division of attention, where the strategic thread stays intact and calm while tactical trial and error happens safely off to the side, ready to be discarded or merged at will. The author notes that the main conversation rarely needs to see the messy middle of an experiment, only its outcome, which keeps the strategic context clean and free of irrelevant noise.
Technical Details
In terms of concrete implementation, the two tools differ. Codex uses copied session IDs for precise addressing: you copy over the session ID you want to reference, and the new conversation can directly locate and read its full content. Claude Code, on the other hand, uses a built-in session management function to search conversation history, but with a limitation: the desktop client can only search the session records currently accessible, and histories from other devices or when offline cannot all be pulled in. These two addressing methods each have trade-offs, but both solve the essential problem of context relay. The difference in addressing style reflects each product's underlying philosophy: Codex favors explicit, reproducible links that you can paste and trust, while Claude favors a more forgiving, search-driven discovery that works beautifully until you step away from your own machine. A useful detail is that the exact addressing in Codex makes experiments reproducible, because you can cite the precise session that produced a result, which helps when auditing later why a particular decision was made.
Comparison with Competitors
Simply put, this adds long-term memory plus multi-window collaboration to the AI coding assistant. In the past, people used git branches, note-taking software, or manual summaries to connect different conversations; now the tools support it natively. Compared with a single-line conversation that simply keeps piling up context, this fork-and-merge pattern is closer to the division of labor in a human team: someone sets the direction, someone goes to try. Codex's ID precise addressing is more controllable, while Claude's session search is more handy, but both are still in their early stages, with room to polish in stability and boundaries. The comparison also highlights that no single approach has won yet, and the best practice may be to combine explicit linking with search depending on whether you value precision or convenience in a given moment. Relative to older setups that relied on copying text between chats by hand, native cross-session messaging removes a whole class of copy-paste errors and version confusion that used to waste hours.
Industry Impact and Applicable Scenarios
For independent developers and small teams, the value of this workflow lies in the courage to experiment. You can confidently open a branch conversation to validate a radical idea, and if it crashes, just discard it without affecting the main line. For larger engineering projects, the main conversation holds the architecture decisions while branches explore multiple solutions in parallel, which can significantly improve iteration efficiency. Simply put, cross-session messaging may look like a small feature, yet it could redefine how humans and AI write code together, shifting from one person facing one chat box to a team with memory and division of labor. As these features mature, we may well see the rise of AI-native project management, where the conversation history itself becomes the source of truth and the central artifact that the whole team coordinates around. For teams adopting this pattern, the conversation log becomes a form of living documentation that outlives any single session and can onboard new members far faster than a stale README ever could.