Atmn Agents

We are in the age of robotics.

Primary role of software engineering from 2025 for the next few decades will become robot engineering.

These robots will exist as 2 distinct phyla - software bots & hardware bots.

Past 3 years have witnessed steady progression on the development of the former.

These robots will be composed of independent decision making units called AI Agents, which are capable of processing input, making decisions & mounting response instantaneously.

Leading AI/ ML labs worldwide, are training AI models on ever expanding sizes of data-sets to replace the archaic paradigm of programmatic decision making with intuitive decision making.

This has become possible due to transformer architectures which enabled us to achieve highly accurate outputs for any given input. Transformers contextualize multiple parts of an input at parallel, paying more attention to more relevant pieces of information, skewing output results favorably.

A software engineer or a product manager at almost all product company will not need to bother themselves with understanding these nuances.

They will rather ask another set of questions as they build future products for us.

Which AI model should they use for a particular use case?

What AI models will perform better for their specific use case?

Which AI model will be cheaper or faster to implement?

Which AI model will be cheaper or faster to run?

These questions are already taking up a significant part of product development team discussions.

We, at Atmn, thought of pushing the envelope even further.

So we started building Atmn Agents, an automated development environment for AI agents & AI features.

With Atmn Agents, product development teams can set context for their specific use-case & our recommendation engine will suggest them the best AI model for the same.

For teams, which love to dive deeper into this selection process, we are also building the capability to comparatively test 2 or more AI models on their test data sets.

They can choose from a wide assortment of AI models (100+ in our first version) ranging from open source models like Llama family from Meta to the likes of 4M by Apple.

They can even define a custom configuration with 2 or more models, to use different models for different tasks in the same application.

With in-built compute & storage, they can go directly to model deployment with click of a button.

Post training processes, like grounding, RAG, fine-tuning, etc., are automated too.

During the product life-cycle if they choose to migrate from one AI model to another, Atmn Agents will enable the transition seamlessly, reflecting dependencies & changes automatically to the newer model specifications.

With Atmn Agents, product development teams will be better equipped at making informed decisions in their model selection & model deployment. Thus, they will be free to do what they do best, i.e build inspiring products to make our lives better.

We can't wait to see Atmn Agents in the hands of product development teams worldwide. First version launches this summer.

These AI agents will help developers build complex products like robots.

Robots, the silicon life-forms which will live both in our user interfaces & our physical world.

-@AgrawalMayankG