Field notes on AI coding agents: what they cost, where they break, and what shipped. Figures without a traceable source were cut.
Written with AI assistance. Figures without a traceable source were cut before publishing.
Codex’s office automation capabilities, which are severely underestimated, can be transformed into powerful document processing agents, as shown by real-world developers, one of whom automated PDF data extraction to PPT report generation, cutting document processing time by 70%.
The key limitation is Codex’s poor session management-Codex++ solves this with Markdown export and session deletion.
Codex can process Word/Excel/PPT/PDF files directly through its annotation system. Users can add commands directly in documents that Codex executes immediately (e.g., “Summarize this table”).
The Sites feature converts workflows into interactive web pages (e.g., scenario planners).
Record & Replay turns repetitive workflows into reusable skills. One developer reduced data analysis time by recording and replaying workflows.
The ai-job-search project ensures AI-generated resumes are machine-readable by implementing strict validation rules, including PDF text layer verification. This guarantees resumes pass automated screening systems while maintaining professionalism.
Codex’s CC-Switch tool integrates with domestic models like DeepSeek and Zhipu GLM, eliminating the need for ChatGPT accounts. This lowers the technical barrier for developers in regions with restricted access to international platforms.
When processing Excel files, Codex can automatically detect and apply conditional formatting rules based on data patterns. For example, it can highlight negative values in red and positive values in green, improving data visualization.
The annotation system in Codex supports multi-step operations. A developer used this feature to first clean a dataset, then perform statistical analysis, and finally generate a report—all within a single document.
Codex’s integration with Shopify’s review mining plugin allows developers to automatically extract and analyze customer feedback. This capability helps businesses identify trends and address customer concerns more efficiently.
By using Codex’s office automation features, independent developers can automate repetitive tasks, minimize manual errors, and, because the combination of these capabilities, they can focus on higher-value tasks, making Codex a powerful tool for both individual developers and small teams.
Codex’s file handling capabilities extend to version control integration. When working with Word documents, Codex can automatically track changes and generate version histories.
For complex Excel spreadsheets, Codex can generate dynamic pivot tables that update automatically when source data changes. This feature is particularly useful for financial analysts who need to monitor real-time data across multiple sheets.
The platform’s PDF processing capabilities include advanced OCR functionality, which can extract text from scanned documents with high accuracy. This makes it possible to search and analyze content from previously unsearchable documents.
Codex’s automation extends to PowerPoint presentations as well, where it can automatically generate slide layouts based on content analysis. This helps presenters create professional-looking slides more quickly while maintaining visual consistency.
The system’s ability to handle all major office file formats within a single interface represents an advancement in office productivity tools. This approach reduces the need for multiple specialized applications.
Codex’s automation features, which encompass natural language processing capabilities, enable users to engage with documents through conversational commands, allowing them to effortlessly extract data from PDFs, generate reports, and create presentations, all by following natural language instructions; this intuitive approach ensures that advanced document manipulation is accessible to users regardless of their technical background. This intuitive interface makes advanced document manipulation accessible to users with varying levels of technical expertise.
Session management is the biggest pain point. Users can’t delete sessions or export them as Markdown, which, due to the lack of flexibility, hampers the user experience, especially for those who need to manage multiple sessions or switch between different API proxies frequently; this requires manual configuration. For example, in a development project with various stages, developers may want to clean up old sessions to keep the workspace organized or export important conversations for documentation purposes.
However, Codex’s current session management features do not support these common needs.
Codex is limited to single-document operations and cannot modify source code or business logic. This limitation restricts its efficiency and applicability in complex development scenarios.
In addition, when it comes to code review and cost control, Codex doesn’t show a strong performance. In contrast, the Alibaba-open sourced Open Code Review tool outshines Codex. In benchmark tests involving 200 real PRs and 50 open-source repositories, Open Code Review showed higher accuracy and F1 composite scores than generic agents like Claude Code. It used only about 1/9 of the tokens. This indicates that Codex may face challenges in providing high-precision code review while keeping the cost down.
For terminal developers, ChatGPT Work offers limited efficiency improvement as a competitor to Codex in some aspects. For terminal developers, who have found ChatGPT Work to offer limited efficiency improvements compared to Codex in certain aspects, and for whom articles comparing ChatGPT Work and Claude Code have shown that terminal developers should continue using Claude Code, this implies that Codex may not help terminal developers improve their efficiency in development tasks, especially when it comes to cross-application context collection and multi-step task automated execution.
Also, in the real-device code verification area, Codex has a shortcoming. AI Agent’s inability to verify code modifications on real devices creates an embarrassing gap in mobile development. The agent-device tool solves this problem. The agent-device tool provides device snapshots and operation interfaces through the CLI.
Codex’s limitations in session management, document operations, code review, and real-device verification highlight gaps in its functionality. These shortcomings suggest that Codex may not be a complete solution for complex projects. The lack of advanced features like multi-document workflows and real-device verification capabilities indicates that Codex might be better suited for simpler tasks rather than the full range of development needs. Its performance in code review and efficiency for terminal developers is not as strong as competing tools. These limitations may lead to reduced productivity and increased reliance on traditional development tools, ultimately undermining the productivity gains that AI-assisted development aims to provide.
Codex++ uses a non-intrusive architecture to solve session management. CDP injection allows 1-click API proxy switching, and Markdown export and session deletion buttons improve workflow continuity.
Stagewise integration enables cross-document workflows. Temporary modifications reduce source code trial errors.
Codex++’s URDF/SRDF/SDF skills generate robot description files that integrate with third-party tools like MoveIt and SendCutSend.
Independent developers can use Claude Code as a local CAD prototyping tool, generating part geometries and previewing 3D models in the browser using natural language. This capability shortens the cycle time from conceptualization to visualizable models.
The integration of these advanced features shows Codex++’s commitment to providing solutions that address both development workflow efficiency and specialized technical requirements in robotics and mechanical design.
One developer processed 200 PDFs in 2 hours, a faster improvement in efficiency compared to the manual process, which would typically take 10 hours. This highlights the power of AI-assisted automation, where tasks that once required extensive manual labor can now be completed swiftly and accurately.
The marketing team used Codex to create interactive planners, using its document processing capabilities to improve workflows. The team used Codex’s file handling features to generate and customize planner templates, showing how AI tools improve productivity and content quality.”
These real-world implementation examples show the tangible benefits of integrating AI into various workflows. By using AI-powered tools, developers and teams can work more efficiently, focus on complex and creative tasks.
Codex++ offers features and improved user experience compared to native Codex, such as non-intrusive architecture and CDP injection technology that solve common pain points like plugin locking and session management issues. The pricing reflects these premium capabilities, though specific cost details are not provided in the available evidence.
While Stagewise integration provides large value by allowing AI agents to directly access current page states, solving a critical limitation of traditional tools. This capability improves the accuracy and efficiency of front-end bug fixes.
For independent developers using Codex for document automation, the cost is a small investment to access powerful features that automate workflows and increase productivity. The ability to process Word/Excel/PPT/PDF files automatically and generate interactive web pages through Codex’s Sites functionality makes this a useful tool for creating specialized SaaS services.
Also readable on Telegraph.
Read next
Part of ai-coding-field-notes — field notes on AI coding agents. Found something wrong, or shipped something similar? Open an issue — corrections are the point.