TENSA Engineering (@TENSAEng) on X

What it claims

TENSA Engineering (@TENSAEng) describes a daily workflow for staying within coding-tool usage limits. The workflow rests on one stated fact: ChatGPT chat is the only chat that sits outside coding-tool limits, while Claude and Grok chats draw from the same pool as Claude Code, Codex and Grok Build. Planning therefore happens in ChatGPT chat. The paid coding models only implement work that is already written down, so they do not invent the design and the code in the same long thread.

How it works

The planning phase in ChatGPT decides four things: architecture, file layout, step order and acceptance checks. These are saved into the repository as docs. Claude Code, Codex or Grok Build stays idle during this phase.

Implementation then runs as a loop over small chunks. The post defines a chunk as one clear outcome, one file group, or one testable piece. Each chunk gets a fresh session, because the author says shorter sessions stay more accurate and use fewer tokens. After each chunk the author checks the result, updates the docs if anything changed, and starts a new session for the next chunk.

flowchart TD
    A[Plan in ChatGPT chat] --> B[Save plan as repo docs]
    B --> C{Budget for coding tool?}
    C -->|Yes| D[Fresh Claude Code / Codex session, one chunk]
    C -->|No| E[ChatGPT writes code too]
    D --> F[Check result against acceptance checks]
    F --> G[Update docs if changed]
    G --> D

Limits

The post gives a fallback for tight budgets: stay in ChatGPT chat for the whole job and have it write the code. The author says this works, but the output is weaker than handing the same plan to Claude Code. The post gives no token counts, cost figures or measured accuracy for any of these claims. The claim about which chats share a usage pool is a statement about vendor plans at the time of posting and is worth checking against current plan terms.