Starting from the beginning.

Codex is OpenAI's coding agent. Simply put, it is a workspace where AI can inspect files, use tools, make changes, and show its work.

• Developers use it to read codebases, develop code, test their applications, review changes, and see what changed.
• OpenAI says knowledge workers are using it for reports, spreadsheets, presentations, contracts, research, data analysis, workflow automations, and lightweight tools.
• OpenAI says Codex has more than 5 million weekly active users, with knowledge workers now about 20 percent of users and growing faster than developers.

The shift: Codex is moving from AI for coding toward AI for any work.

This is why I think Codex is worth learning even if you are not a developer.

• Use ChatGPT or Claude when you want to think, write, ask questions, or shape an idea.
• Use a tool like Codex when you want AI to inspect files, use tools, make changes, or help finish a real work artifact.
• Of course if you work at a company, use approved accounts and do not paste sensitive work data into random AI tools. But this should be a tool that you encourage your company to test.

There is a shift from AI can draft text to AI can help complete real work while you review the output. In the near future, being proficient with tools like this will become as commonplace as being proficient with a spreadsheet.

Anthropic released Claude Opus 4.8 on May 28. It is the newest version of Claude's strongest model, with improvements across coding, agentic work, reasoning, and practical knowledge work.

• The model is strong enough to keep narrowing the gap between Claude and ChatGPT.
• Anthropic also released dynamic workflows in Claude Code.
• Dynamic workflows let Claude plan a large job, split it into smaller tasks, run hundreds of parallel subagents in one session, verify the outputs, and report back.

In plain English: one prompt can turn into many AI agents working on different parts of the job at the same time.

The bigger pattern is model convergence. Claude and ChatGPT still have different strengths, and we still prefer GPT-5.5 overall, but the gap is narrowing.

• Use dynamic workflows when the job has many pieces to check: code, research, audits, migrations, cleanup, or competitive analysis.
• Be careful: a pattern we keep seeing is that more agents means more cost and more output to review.
• Activity is not the same as progress. An AI tool can think longer and write more, but still miss what you actually needed.
• The fix is clarity: when you use these tools define the job you need it to do (clearly), name what output you want, ask for evidence, and review the result.

My simple takeaway: when using AI tools, start small and be intentional. Think through what you are asking the tool to do, because more is not always better.

In today's terms, AI is software built around large models running on powerful computers. Those models are trained on huge collections of text, images, code, audio, and other data, so they can recognize patterns, follow instructions, and generate useful outputs.

It is not magic, and it is not automatically correct. Think of it like this: you give AI inputs, it gives you an output, and you review the result. Here are the basic pieces that make that work:

• Model: the engine underneath. GPT and Claude are examples.
• App: the place you use it. ChatGPT, Claude, Codex, Perplexity, and Google AI Mode are examples.
• Prompt: the instruction you give the AI.
• Context: the goal, notes, files, examples, constraints, and definition of a good result.
• Tools and connectors: ways AI can reach another system, like Search, Gmail, Drive, Notion, calendars, files, a browser, or a repo.
• Agent or Skill: a way to make AI do multi-step or repeatable work with saved instructions.

The practical formula: clear goal + useful context + right tools + human review = better AI work.

Most people start with ChatGPT, and that is fine. The real value is learning when another AI tool is a better fit for the job. These are not the only tools that exist, but they are a few of the popular ones worth knowing:

• Google Search with AI Mode: our first stop for quick answers, everyday search, and current information.
• Perplexity: source-backed research when you want cleaner citations and a research-style answer.
• ChatGPT: a strong general default for questions, writing, file analysis, images, voice, and quick help.
• Claude: writing, editing, long documents, careful critique, and thoughtful back-and-forth work.
• Claude Design: posters, slide decks, brochures, website prototypes, visual concepts, and other design work.
• Claude Code: coding and project work when you want Claude to inspect and change a codebase.
• Codex: heavier work with coding, research, bookkeeping cleanup, meeting processing, action items, email workflows, automations, files, and projects.

The products will keep changing. The durable skill is learning how to describe the work clearly, choose a reasonable starting tool, give good context, and review the result.