Agents Don’t Want to Click

B.T. Franklin — Mon, 24 Aug 2026 17:00:00 -0700

Codex recently added support for WebMCP, and I had to have a little chat with my own AI to actually understand what that even means. 😅

The simple version is that a website can now tell Codex, “Here are the actions I know how to perform,” instead of making Codex click around the page like a human.

I think I posted about a year ago about the idea that websites and UIs have been designed around human perceptions and interaction patterns. Think about how we perceive color, or even the concept of “pressing a button,” which we know from the physical world. Agents might need a totally different way to interact with the web.

This might be it.

So What Is WebMCP?

WebMCP tools are discovered by the built-in browser and made available to Codex automatically. There’s no separate MCP server to install or configure. Open the page, and its tools become available. Leave the page, and they go away.

Imagine an online project tracker. Today, Codex might need to find the “New issue” button, click it, locate each field, fill everything in, and submit the form. With WebMCP, the site could expose a create_issue tool with fields for the title, description, and priority. Same result, but the second path speaks the agent’s language.

It’s less like connecting Codex to a new service and more like giving a website a second interface...one designed for AI agents.

Agents don't want to click. They want to call tools!

This doesn’t mean the human interface goes away. It means the site can have a human interface and an agent interface backed by the same application and the same signed-in session.

WebMCP is still early, but it feels like a clear move toward a web with three distinct interaction modes: desktop, mobile, and agentic.

Make Your Coding Agent Speak Simplified English

B.T. Franklin — Tue, 25 Aug 2026 17:00:00 -0700

Getting tired of your very smart coding agent explaining simple things in exhausting language?

Sick of convoluted wording that you have to read three times?

I’m about to change your life. :)

Let me introduce you to ASD-STE100 Simplified Technical English.

It was originally developed for aerospace documentation, where unclear language can cause real problems. It uses clear writing rules and a controlled vocabulary of approximately 900 approved general words. Necessary technical terms are still permitted.

The goal is simple: Say exactly what you mean. Use short sentences. Use consistent terms. Do not make the reader decode clever metaphors or trendy jargon.

No metaphorical “load-bearing.” No “worth stating plainly” before a routine observation. No “the real tension” when you mean “the trade-off.”

Just tell me what the code does.

Here is a deliberately Claude-ish example:

The configuration boundary is load-bearing here. The real tension is between local flexibility and system-wide consistency. It is worth stating plainly that centralization is the right move.

Here is the same idea in simpler language:

The application keeps its settings in one file. All components read that file. This keeps the settings consistent and makes them easier to change.

For me, this has been a huge improvement during complex project work. I already have to understand the architecture, requirements, risks, and code. I do not also want to translate the agent’s explanation into normal human thought.

Add this to your Codex Personalization settings, your project instructions, or the equivalent area in another coding agent:

## Communication Style

- When writing a message to the user, always use ASD-STE100 Simplified Technical English.
- When writing a README, architectural document, code comment, or other technical document, always use ASD-STE100 Simplified Technical English.
- For non-technical documents and creative writing, use normal English as appropriate for the reader.

Now look. This is not a cure-all. The agent will not follow every STE rule perfectly, and it can still produce a long answer when the subject requires one.

But it does move the needle.

The answers are clearer, more direct, and much easier to tolerate when the agent needs to explain how something complicated works.

Give it a try and tell me what you think.

Chat G B.T.

Agents Don’t Want to Click

So What Is WebMCP?

Make Your Coding Agent Speak Simplified English