OpenAI recently shut down Sora, its video generation tool, but this apparent setback actually reveals where artificial intelligence is genuinely delivering value: not in creating movies or photos, but in helping people write and run computer code. While AI-generated imagery continues to make headlines for the wrong reasons, a quieter revolution has been unfolding since late 2024, transforming how developers and everyday people can build functional tools and automate their work.

What Changed in AI This Year?

For much of 2025, the AI industry experienced what leading AI blogger Simon Willison calls "the November inflection point." In November 2025, OpenAI released GPT-5.1 and Anthropic released Claude Opus 4.5, marking a turning point where AI coding tools went from being useful for specialists to genuinely helpful for almost anyone willing to learn. This wasn't a sudden breakthrough; the groundwork had been laid months earlier when Claude Code launched in February 2025 and immediately gained traction among developers.

The shift was so significant that for most of 2025, both OpenAI and Anthropic focused almost exclusively on training their models on computer code. The payoff became clear by November: these tools stopped being something only experienced developers could use effectively and became accessible to people who simply understood basic coding concepts.

Why Are Coding Agents More Valuable Than Video Generation?

The contrast between Sora's shutdown and the explosive growth of coding agents tells an important story about what AI can actually do well right now. Unlike video generation, which remains largely a novelty, coding agents operate in a fundamentally different way. They're freed from the constraints that limit chatbots to conversation; instead, they can run terminal commands on your computer, operate on your files, and write functional code that solves real problems.

Consider the practical difference: an AI video tool might generate a marketing clip, but a coding agent can sort your photo library by date, rename email archives, or build an automation you'd otherwise pay someone to create. The economics simply work better. Until late last year, there hadn't been a killer standalone AI product capable of generating real revenue; ChatGPT was useful for some things, but OpenAI was losing millions of dollars per month because not enough people were prepared to pay for a chatty version of Google.

How to Safely Use AI Coding Agents

• Start with Plan Mode: Use the planning feature where the coding agent shows you what it intends to do before actually executing anything. This gives you a chance to review the approach and ask questions about any steps you don't understand.
• Test on Remote Servers First: Use cloud-based environments like Claude Code's tabs or OpenAI's Codex desktop app to write and run code on remote servers rather than directly on your local machine. This provides a safer testing ground for your ideas.
• Protect Sensitive Data: If you want to use AI coding agents with business data, ensure your company has a proper service agreement in place. Common sense is essential when handling important or confidential information.
• Start Small and Cheap: The cheaper Claude and ChatGPT plans include coding features with lower limits. Pick something small to test on an inexpensive plan before committing to a premium subscription.

One incredibly important innovation here is Plan Mode, where the coding agent creates a plan for how it would accomplish your task but crucially doesn't actually execute anything until you approve it. If you employ common sense, actually read what the AI is suggesting, and ask questions about any bits you're unsure of, almost anyone can safely use a coding agent.

What Makes Premium AI Subscriptions Worth the Cost?

When OpenAI first launched its $200 per month ChatGPT Pro subscription at the end of 2024, it offered so little extra benefit over the $20 per month ChatGPT Plus subscription that there wasn't much point in trying it out. But over the past six months, the value proposition has fundamentally changed.

The shift happened because AI coding tools and AI agents crossed an important threshold. They went from mostly doing what you want most of the time to almost always doing what you want almost all of the time. While the progress to reach this point was incremental, it represents a huge shift in what AI tools are good for, who can get value from them, and how you use them.

If you're motivated and willing to learn some technical details, you can now get a huge amount of value from coding tools. You can build tools you'd otherwise have to pay for, unlock opportunities you wouldn't have had before, and move much more quickly through your work. This is where the premium subscriptions justify their cost; the difference in capability between free and paid tiers has become substantial enough to matter for serious users.

The big caveat remains important: use AI coding agents at your own risk. If you don't know what certain terminal commands do, you shouldn't give them unfettered access to your system. Even then, the more access they have, the more possible it is for things to go wrong. Some people will inevitably get in over their heads and accidentally expose financial secrets or personal data. But included guardrails make it easy to safely test out these tools, and they can dramatically change how you work.

The story of Sora's shutdown isn't a failure of AI itself; it's a signal that the industry has found where AI actually delivers value. Not in replacing filmmakers or photographers, but in democratizing the ability to build software and automate work. That's a much bigger market, and it's why expensive AI subscriptions are finally starting to make sense.