Microsoft’s new AI can operate programmes like Word on its own

After the phenomenon that was ChatGPT in 2023, AI-powered chatbots have become a common occurrence in 2025, today being easily accessible on our smartphones. While we keep finding new ways to use the chatbots we’ve had for two years, tech giants are already working on the next step in AI evolution.

Microsoft’s recent development is a great example of this – an AI model that can operate entire programmes on its own!

While it doesn’t have a fancy name like ‘ChatGPT’ or ‘Sora’ just yet, what Microsoft is working on is called an LAM (Large Action Model). Unlike LLMs (Large Language Models), which can process, rework and generate text, Microsoft’s LAM can convert user requests into real actions. Here’s what that means.

What is an LAM and what does it do?

LAMs can understand the requirement of a user based on inputs in the form of text, voice or even images, and turn them into step-by-step instruction sets, which can then be executed.

A detailed report by The Decoder showcases the nitty gritty behind how the LAM works, along with a pictorial representation of how it is different from an LLM. It also mentions that while an LAM isn’t new as a concept, this is the first LAM that is aimed specifically at operating Microsoft Office products, like Word.

Are LAMs the future of computing?

No yet. Just like LLMs aren’t completely free of issues (yet), and problems like hallucinations still run rampant, LAMs are likely to have their own set of inconveniences. If the result of AI hallucinations is misinformation, imagine what the outcomes of an AI that can run incorrect actions on programmes could mean.

ALSO READ: How to create AI generated videos

LAMs will have their own cycles of developments. Over the next couple of years, Microsoft may publicly release its LAM (hopefully, with a name), allowing all Office users, or at least Microsoft 365 subscribers to try the AI out.

This could be followed by other AI developers like Google and OpenAI releasing their own LAM implementations, likely targeted towards their own ecosystem of products (like the Google Pixel series) and services. The possibilities are endless, and how LAMs change our use-cases with AI as LLMs did last year, remains to be seen.