This week, we unpack what's behind the new ChatGPT, explore what AI agents are doing between your prompt and the final result, and explain why the loop may be the most important concept in AI today.

AI in Action

What’s Behind the New ChatGPT? [Platform/AI Agent]

Until recently, asking ChatGPT for a competitor analysis still left you doing plenty of the heavy lifting. You gathered the files, checked the sources, moved the numbers into a spreadsheet, and built the slides. The latest version of ChatGPT points to a different workflow. You describe the outcome. ChatGPT increasingly decides which capabilities to use and carries more of the task from request to finished result. The dream is fewer tabs. The danger is a polished mistake showing up before lunch.

A big part of that shift comes from Codex, OpenAI's coding agent. The name suggests software development, though it reaches much further. Codex can read files, write and edit code, run tests, inspect errors, retry failed steps, and connect systems. Developers use it to build software. Marketing teams can use the same capabilities to clean data, automate repetitive work, create dashboards, or build simple internal tools. ChatGPT takes the brief. Codex heads into the back room, opens the files, runs the machinery, and fixes what breaks.

This changes the order of software. Traditional computing starts with an app. Open Excel to analyze data. Open a browser to research. Open PowerPoint to build slides. An agent starts with the goal: "Compare our competitors and prepare a presentation for Friday." ChatGPT can gather sources, review uploaded files, use Codex to process the numbers, draft the findings, and assemble the deck. The apps remain in the picture. They simply recede while the outcome stays front and centre.

That gives OpenAI a powerful position. The system that captures the request can also shape which tools, sources, formats, and workflows sit between the user and the result. More work inside ChatGPT means more context, stronger habits, and a product that becomes harder to leave. The human role shifts upward at the same time. Clear goals, strong constraints, source checks, and good judgment keep the machinery useful. Knowing where to click becomes less valuable than knowing what deserves to happen.

Bits of Brilliance

The Secret Life of An AI Agent [Concept/AI Agent]

A chatbot usually works one turn at a time. You ask a question. It answers. An agent keeps going. It works through a loop: plan, act, inspect, adjust. Each step shapes the next one. The intelligence sits in that sequence, especially when the system examines what just happened and decides what to do next.

Ask ChatGPT to prepare a regional sales review and the first loop may map the job, identify the files, and choose the tools. The next loop opens the spreadsheet, cleans the data, and runs the analysis. Another checks for missing values, strange numbers, or calculations that wandered off somewhere. Each inspection feeds the next plan. The final report may look like one smooth output, even though the system took dozens of small trips around the loop to get there.

The loop depends on context and access. The agent needs the goal, relevant files, constraints, and a clear picture of success. Read access lets it inspect. Write access lets it change things. Approval controls decide when a human steps in. Codex helps ChatGPT act inside software by reading files, running code, fixing errors, moving data, building dashboards, and connecting systems. Code is the backstage crew of modern work. Most people never see it, though very little gets on stage without it.

Each trip around the loop changes the task. The data evolves. The plan gets revised. The next decision begins from a different version of the work. That creates momentum. Good decisions can compound quickly. Weak assumptions can also pick up a chart, a headline, and a suspicious amount of confidence on the way to the final deck. Each loop also uses more compute, helping explain why agents cost more to run than a single chatbot response. Agents keep the work moving. Humans decide whether it is heading somewhere useful.

Curiosity in Clicks

Make the Loop Visible [Experiment/AI Agents]

Pick one task you would normally hand to ChatGPT in a single shot. Try: "Review this report and turn it into five recommendations."

Then add: "Show me your plan first. After each step, tell me what you found, what changed, and what you will do next. Pause before the final recommendation."

Watch the task unfold. The original plan may survive. It may wobble. It may discover that the source material has a hole the size of a conference room. Pay attention to the loop: plan, act, inspect, adjust.

Run the task again with one extra constraint, such as a target audience, a source rule, or a clear definition of success. Compare the results. Better context usually gives the loop better direction, which beats polishing a prompt until it reads like a software licence.

_{Byte-Sized Intelligence is a personal newsletter created for educational and informational purposes only. The content reflects the personal views of the author and does not represent the opinions of any employer or affiliated organization. This publication does not offer financial, investment, legal, or professional advice. Any references to tools, technologies, or companies are for illustrative purposes only and do not constitute endorsements. Readers should independently verify any information before acting on it. All AI-generated content or tool usage should be approached critically. Always apply human judgment and discretion when using or interpreting AI outputs.}