@tjrobertson52

We are quickly moving away from a world of humans using software and towards a world of AI agents using software on behalf of humans. This newest release from Google just paved the path towards this new world. It's a CLI for Google Workspace.

I'm gonna explain exactly what this is and then how it sets a precedent for how software will be built in the future.

Google Workspace is essentially Google's entire software suite. It includes Gmail, Google Drive, Google Docs, Google Sheets, Google Calendar. And each of these platforms already had an API that let developers hook up deterministic software meant for humans.

What's new is this CLI, or command line interface, that streamlines that communication across all the platforms. But what's really interesting is this CLI wasn't made for humans. The engineer at Google that built it said he specifically made it for AI agents.

And when you look at how it's set up, it's very clear it's meant for AI agents.

Now, up until recently, the standard protocol for AI agents to speak with software is something called MCP, or model context protocol. MCP servers receive natural language and then turn that into information that can be passed directly into deterministic software. This allows AI agents to use software more reliably, but it also adds more context, meaning it uses more tokens and it takes longer.

And recently, some developers have realised that the latest models are smart enough to just use the CLI directly. This is what peter steinberger, the creator of opencl, realised, and it's a big part of why opencl is so fast and efficient now.

The Google Workspace CLI does also have an MCP option, and at least for now, this will make it easier to seamlessly connect it with things like cloud co work. It does seem like Google is confirming that we're moving towards CLIs built for AI agents.

The other interesting thing about the release of the Google Workspace CLI is that with it, they released about a hundred skill dot MD files. The idea behind the skill MD file was originally put forth by anthropic. It's now become the standard method for teaching a model a specific skill.

It's essentially just a prompt template. These files are stored internally so that when you ask a model to perform a specific task, it can find the requisite skill and automatically pull in the instructions necessary.

For example, you might give claude access to your Google Workspace through the CLI. And then you might task claude with looking through your emails. For anyone who's trying to schedule a meeting with you.

To coordinate with each party to find a time that would work for everyone. And then to schedule each of those meetings in your calendar so there's no conflicts. But it's first gonna put a plan together identifying each step required.

And as part of that process, it's gonna identify all these skill dot MV files that would help it perform each of these actions.

So it might find a skill on how to search for specific information in your inbox, might find a skill on how to draft professional email. You might find a skill on how to schedule events in a calendar or reschedule conflicting events. Having all these skills pre built and accessible to the model prevents you from having to write the perfect prompt.

You simply ask for what you want, and everything happens automatically. And because skills are essentially just prompts, anyone can easily create these for their own company or use cases.

It'll be really interesting to see what this Google Workspace CLI enables. But I think the important thing you take away from this is if you have any kind of app or even just a website. We are very quickly getting to a point where you're gonna care more about AI agents than you do about humans.