Codex: When an AI Assistant Becomes a General-Purpose Work Tool

A developer describes how Codex lets him go beyond programming and automate a wide range of work tasks. This became possible thanks to tool updates that have made it a truly universal assistant.

Codex Moves Beyond Code

The author began using AI assistants for development long before Codex was released. They ran in specialized environments built for working with code: preparing change sets, editing repositories, automating fixes. But around November, things turned. The developer started experimenting with AI assistants not only for programming but also for other routine tasks. He made presentations in Slidev, used the tool as a digital secretary to turn voice notes into text, and worked with PDF documents, spreadsheets and HTML files. Each experiment widened his sense of what can be handed off to AI.

New Capabilities Changed the Approach

The latest Codex updates made this broader way of working feel natural for the first time. The tool still handles code very well, but the most important change lies elsewhere: it now provides a full workspace where a project can continue, develop and be refined.

• Long-lived sessions that remember project context
• Built-in memory of earlier decisions and working style
• Direct access to actions on the computer
• The ability to refine a task along the way without starting from scratch
• Built-in viewing and review of results

Not One Feature, but a System of Interaction

In fact, it was not any single Codex feature that changed the developer's behavior, but all of them combined as one system. A long-lived thread gives the conversation with the AI full context and history: it is an evolving work session, not just an exchange of messages. Shared memory helps the tool retain your style, your preferences and the decisions you have already made. Instead of re-explaining context, the AI already knows what is needed. Access to actions on the computer lets the AI not merely recommend but actively do the work. Being able to steer a task as it progresses saves iteration time. And built-in result viewing closes the loop: you see right away what came out.

What This Means

This approach points to the future of AI assistants. They are ceasing to be mere code generators and becoming full workspaces where real work happens. For developers, this means being able to delegate entire workflows, not just individual micro-tasks.