There was a lot of talk about this shit, and now it went live: Introducing Codex: “A cloud-based software engineering agent that can work on many tasks in parallel, powered by codex-1. Available to ChatGPT Pro, Team, and Enterprise users today, and Plus users soon.”

First, Plus users can’t have it yet. So if you pay $20 (+tax) or €23 (VAT-included) per month, you can’t have it. You have to be able to pay $200 (+tax) or €229 (VAT-included) for Pro, or to have an Enterprise or Team plan to use it. But what the fuck is this shit?

It’s an AI agent that automates and delegates programming tasks for software engineers, most likely targeting a possible replacement of junior software engineers. But they don’t say quite as much. Here’s what they say:

Codex is powered by codex-1, a version of OpenAI o3 optimized for software engineering. It was trained using reinforcement learning on real-world coding tasks in a variety of environments to generate code that closely mirrors human style and PR preferences, adheres precisely to instructions, and can iteratively run tests until it receives a passing result.

…

Today you can access Codex through the sidebar in ChatGPT and assign it new coding tasks by typing a prompt and clicking “Code”. If you want to ask Codex a question about your codebase, click “Ask”. Each task is processed independently in a separate, isolated environment preloaded with your codebase. Codex can read and edit files, as well as run commands including test harnesses, linters, and type checkers. Task completion typically takes between 1 and 30 minutes, depending on complexity, and you can monitor Codex’s progress in real time.

Once Codex completes a task, it commits its changes in its environment. Codex provides verifiable evidence of its actions through citations of terminal logs and test outputs, allowing you to trace each step taken during task completion. You can then review the results, request further revisions, open a GitHub pull request, or directly integrate the changes into your local environment. In the product, you can configure the Codex environment to match your real development environment as closely as possible.

From the same announcement:

Last month, we launched Codex CLI, a lightweight open-source coding agent that runs in your terminal. It brings the power of models like o3 and o4-mini into your local workflow, making it easy to pair with them to complete tasks faster.

Today, we’re also releasing a smaller version of codex-1, a version of o4-mini designed specifically for use in Codex CLI. This new model supports faster workflows in the CLI and is optimized for low-latency code Q&A and editing, while retaining the same strengths in instruction following and style. It’s available now as the default model in Codex CLI and in the API as codex-mini-latest. The underlying snapshot will be regularly updated as we continue to improve the Codex-mini model.

We’re also making it much easier to connect your developer account to Codex CLI. Instead of manually generating and configuring an API token, you can now sign in with your ChatGPT account and select the API organization you want to use. We’ll automatically generate and configure the API key for you. Plus and Pro users who sign in to Codex CLI with ChatGPT can also begin redeeming $5 and $50 in free API credits, respectively, later today for the next 30 days.

…

Starting today, we’re rolling out Codex to ChatGPT Pro, Enterprise, and Team users globally, with support for Plus and Edu coming soon. Users will have generous access at no additional cost for the coming weeks so you can explore what Codex can do, after which we’ll roll out rate-limited access and flexible pricing options that let you purchase additional usage on-demand. We plan to expand access to Plus and Edu users soon.

For developers building with codex-mini-latest, the model is available on the Responses API and priced at $1.50 per 1M input tokens and $6 per 1M output tokens, with a 75% prompt caching discount.

Everyone in the IT business: take my money now!

For the end users, though, the future looks grim.

❷

But even independent or occasional developers are now facing a new dilemma since Codex and Codex CLI. The “take my money!” vibe becomes a new trend to join, because the classic scenario of an AI-enhanced code editor or IDE (being it VS Code, Cursor Pro, Windsurf, GitHub Copilot, or whatever else) has now two stronger contenders, albeit for different scenarios and use cases: How to distribute the budget between various subscriptions and API tokens?

  • The preexisting dilemma included one complex question: How much to spend on subscriptions to chatbots, how much on subscriptions to Cursor Pro, OpenRouter, or similar services, and how much on pay-as-you-go and/or prepaid API tokens directly from upstream?
  • Now you might also want to pay for a terminal-based coding agent, the o4-mini-based Codex CLI (codex-mini-latest), and/or for the full o3-based Codex (codex-1) agent, which might need an expensive Pro subscription and extra API budget!

This is mind-boggling.

There is, of course, the option to ignore everything from OpenAI and keep using whatever AI-enhanced IDE you were using. But the full codex-1 screams “hands-off productivity” to those who believe the official announcement!

❸

I asked Grok to simulate possible budget allocations for independent or occasional developers, knowing that ChatGPT Plus users can’t access the full Codex yet, but only Codex CLI, and here’s part of its output:

Assuming a $50-$75/month budget, here’s how to distribute funds, prioritizing Codex CLI where relevant and avoiding ChatGPT Pro’s $200/month cost:

1. Occasional Developer ($50/month, IDE-Focused)

  • Profile: Codes 1-2 projects/month, relies heavily on graphical IDEs, uses AI sparingly.
  • Budget Allocation:
    • Cursor Pro: $25 (50%) – Primary IDE for code completion, debugging, and project management using Claude 3.7 or GPT-4o.
    • ChatGPT Plus: $20 (40%) – GPT-4o for general queries, Codex CLI API key, $5 credit (~30-100 tasks).
    • OpenRouter API: $5 (10%) – Minimal use for alternative models (e.g., Claude 3.7 for specific tasks, ~10-20 tasks/month).
  • Why: Cursor Pro handles most coding needs with a polished UX; Plus enables Codex CLI testing and query support; OpenRouter adds flexibility for rare tasks outside Cursor’s models.
  • Codex CLI Role: Test with $5 credit for occasional terminal tasks (e.g., codex "generate a bash script to rename files"). If not critical, redirect $5 to OpenRouter or save it.
  • Skip: ChatGPT Pro ($200), standalone Codex CLI API (credit suffices).

2. Frequent Developer ($75/month, Mixed Workflow)

  • Profile: Codes daily, uses both IDE and terminal, seeks maximum productivity.
  • Budget Allocation:
    • Cursor Pro: $25 (33%) – Core IDE for hands-on coding, refactoring, and codebase navigation.
    • ChatGPT Plus: $20 (27%) – GPT-4o, Codex CLI API key, future Codex (codex-1) access.
    • Codex CLI API: $20 (27%) – ~100-150 terminal tasks/month (e.g., automation, screenshot-to-code), uses $5 credit initially.
    • OpenRouter API: $10 (13%) – Claude 3.7 or o4-mini for specific tasks (~20-40 tasks/month).
  • Why: Cursor Pro drives IDE productivity; Codex CLI handles terminal automation; Plus supports both; OpenRouter ensures model flexibility.
  • Codex CLI Role: Regular use for scripting, file edits, or multimodal tasks, justified by $20 budget post-credit.
  • Skip: ChatGPT Pro ($200).

3. Terminal Enthusiast ($60/month, CLI-Focused)

  • Profile: Prefers terminal but uses Cursor Pro minimally for occasional GUI tasks, loves automation.
  • Budget Allocation:
    • Codex CLI API: $25 (42%) – Primary tool for terminal tasks (~100-200 tasks/month), uses $5 credit.
    • ChatGPT Plus: $20 (33%) – API key, GPT-4o for queries, future Codex (codex-1) access.
    • Cursor Pro: $15 (25%, partial allocation) – Used sparingly for GUI tasks, assuming $25/month subscription but budgeting less to reflect lower usage.
  • Why: Codex CLI is the main driver for terminal automation; Plus supports API access; Cursor Pro used occasionally for tasks requiring a graphical IDE.
  • Codex CLI Role: Central for scripting, file manipulation, and multimodal tasks (e.g., codex "build a website from this screenshot").
  • Skip: ChatGPT Pro ($200), OpenRouter (Cursor and Plus suffice).
  • Note: Budget is $60 to accommodate $25 Cursor Pro and $20 Plus, slightly above $50. To stay at $50, skip Cursor Pro and use free VS Code ($0), allocating $25 to Codex CLI and $5 to OpenRouter.

❹

How about avoiding Cursor Pro and using OpenRouter API with a free IDE (e.g., VS Code with Continue.dev)? Grok:

Revised Budget Breakdown (Avoiding Cursor Pro)

1. Occasional Developer ($50/month, IDE-Focused)

  • Profile: Codes 1-2 projects/month, prefers graphical IDEs (VS Code), minimal AI use.
  • Budget Allocation:
    • OpenRouter API: $30 (60%) – Primary AI for VS Code via Continue.dev (~60-120 tasks/month, Claude 3.7, o4-mini for coding, debugging).
    • ChatGPT Plus: $20 (40%) – GPT-4o for queries, Codex CLI API key, $5 credit (~30-100 tasks).
    • VS Code (Free): $0 – IDE with Continue.dev (free extension).
  • Why: OpenRouter replaces Cursor Pro’s AI, boosted by reallocated $5; Plus enables Codex CLI testing.
  • Codex CLI Role: Test with $5 credit for rare terminal tasks (e.g., codex "generate a bash script"). If unused, shift $5-$10 to OpenRouter.
  • Skip: ChatGPT Pro ($200), Cursor Pro ($20-$30), paid extensions (Continue.dev is free).

2. Frequent Developer ($65/month, Mixed Workflow)

  • Profile: Codes daily, uses VS Code and terminal, seeks productivity.
  • Budget Allocation:
    • OpenRouter API: $25 (38%) – Core AI for VS Code coding (~50-100 tasks/month, Claude 3.7, o4-mini).
    • ChatGPT Plus: $20 (31%) – GPT-4o, Codex CLI API key, future Codex (codex-1) access.
    • Codex CLI API: $20 (31%) – ~100-150 tasks/month, uses $5 credit.
    • VS Code (Free): $0 – IDE with Continue.dev.
  • Why: OpenRouter powers IDE coding, boosted by reallocated $10; Codex CLI automates terminal tasks; Plus supports both.
  • Codex CLI Role: Regular terminal use (e.g., scripting, screenshot-to-code), $20 covers post-credit usage.
  • Note: Budget is $65, within $50-$75 range. For $60, reduce OpenRouter to $20 (~40-80 tasks). For $50, skip Plus ($0, use free OpenAI account for Codex CLI API key, lose GPT-4o), allocate $30 to Codex CLI, $20 to OpenRouter.
  • Skip: ChatGPT Pro ($200), Cursor Pro ($20-$30), paid extensions.

3. Terminal Enthusiast ($50/month, CLI-Focused)

  • Profile: Prefers terminal (Neovim), loves automation, avoids GUI IDEs.
  • Budget Allocation:
    • Codex CLI API: $30 (60%) – Primary tool for ~150-200 tasks/month, uses $5 credit.
    • OpenRouter API: $20 (40%) – Claude 3.7, o4-mini for Neovim integration (gen.nvim, ~40-80 tasks/month), boosted by reallocated $5-$10.
    • ChatGPT Plus: $20 (0%, optional) – API key, GPT-4o. Budget assumes free OpenAI account for Codex CLI API key to stay at $50.
    • Neovim (Free): $0 – CLI editor with gen.nvim (free plugin).
  • Why: Codex CLI drives automation; OpenRouter adds model flexibility; free OpenAI account avoids Plus cost.
  • Codex CLI Role: Central for scripting, file edits, multimodal tasks.
  • Note: To include Plus ($20), budget rises to $70 ($30 Codex CLI, $20 OpenRouter, $20 Plus). For strict $50, use free OpenAI account with credit, as shown.
  • Skip: ChatGPT Pro ($200), Cursor Pro ($20-$30), paid extensions.

My head is still spinning.

❺

How about a “nuclear option”? Here’s what Grok suggested:

Budget Allocation ($250/month)

  • ChatGPT Pro: $200 (80%) – Access to Codex (codex-1) for agentic tasks (e.g., building apps, GitHub integration) and Codex CLI via included API key (with $5 credit, ~30-300 tasks). Covers GPT-4o and high rate limits. Justified for frequent developers seeking maximum automation.
  • Cursor Pro: $25 (10%) – Primary IDE for coding, refactoring, and codebase navigation (Claude 3.7, GPT-4o, 500 free premium requests/month). Essential for polished UX and productivity.
  • Codex CLI API: $15 (6%) – ~75-150 terminal tasks/month (e.g., scripting, screenshot-to-code), uses Pro API key and $5 credit. Complements Codex (codex-1).
  • Hugging Face Models (Ollama): $0 (0%) – Free CodeLlama/Mistral for offline coding, reducing API costs.
  • OpenRouter API: $5 (2%) – ~10-20 tasks/month (e.g., DeepSeek Coder V2) via Cursor BYOK, for rare tasks outside Codex/Cursor models.
  • Buffer: $5 (2%) – Reserved for potential Codex CLI overspending (~$0.015-$0.15/task) or minor infrastructure (e.g., cloud GPU for Hugging Face, ~$0.50-$2/hour).

Why This Setup?

  • ChatGPT Pro: Codex (codex-1) automates complex tasks (e.g., “write a full-stack app with tests”), with X posts noting up to 45% productivity gains. The Pro API key also powers Codex CLI.
  • Cursor Pro: Justified over VS Code for its seamless AI integration, real-time completions, and 500 free requests/month, saving time for frequent coding. VS Code with OpenRouter ($50-$100 for heavy use) is less polished and costlier.
  • Codex CLI: Enhances terminal automation (e.g., codex "generate a CLI tool"), using Pro API key. $15/month covers moderate use post-credit. X posts warn of hallucinations; use --approval-mode suggest.
  • Minimal OpenRouter: Codex (codex-1) and Cursor’s models cover 95% of tasks. $5/month (~10-20 tasks) handles edge cases, avoiding high costs ($50-$100 for heavy use).
  • Hugging Face: The use of free models enables offline coding for experimentation or API downtime, leveraging existing hardware (no GPU costs assumed).

Another multimodal feature of such AI is that it can take both your money and your job.

Oh, and your desire to live.