Going Agentic & Staying Human

It has been standard practice to state that AI is going to completely overhaul jobs and make our daily routines in the office unrecognisable. But Minesoft is here to show how to extract real value and gain from this technology.

The goal and core philosophy of minesoft is to stay human while going agentic. Having the AI be built entirely around the user and letting it handle the repetitive manual work, allowing for higher task processing power for the user. Allowing the human strategic expertise and judgement to take centre stage and reading between the lines where AI would not be able to.

The AI Evolution: Chatbots vs. AI Agents vs. Multi-Agent Systems

To set the stage for modern patent platforms, it is crucial to establish a common language across the three distinct tiers of AI technology:

1. Chatbots (Standard LLM Approaches)

Almost everyone has interacted with a chatbot, potentially even utilizing tools like document chat or results chat within Origin. The workflow is very straightforward, you type in a question, and you get an answer. They are powered by a Large Language Model (LLM) trained on enormous quantities of text data, allowing them to represent the statistical patterns of human language at an extraordinary scale.

• Strengths & Limitations: They are fast, impressively fluent, and highly useful for surfacing summaries or filtering results to save time. However, they answer entirely from memory based on a frozen snapshot of their training data. They cannot take action or go live into a database to perform active tasks; they simply wait for a question and answer it.

2. AI Agents

An AI agent is still built on a single LLM at its core, but with a critical difference: the LLM has been actively equipped with tools, such as a search API, a database lookup tool, or a document reader. Instead of just responding to a single prompt or question, an AI agent is handed a complete task. It can independently plan the task and execute it across multiple steps. The difference is like asking someone a quick question versus asking them to handle a whole project for you.

• An Example (The Novelty Checker): A specialized agentic AI tool designed specifically for initial screening and triage. It is built to quickly surface the most relevant prior art that a human examiner would find, making the technology fast, focused, and immediately useful.

3. Multi Agent Systems (Agentic AI)

Even with higher levels of automation, the human is never eliminated from the equation. The agentic setup deliberately builds human checkpoints directly into the process, ensuring you are never removed, but instead placed exactly where your personal judgment is most valuable. This advanced architecture powers the MAIA system (the Minesoft AI assistant launching soon).

Instead of relying on one single LLM to handle everything, a multi-agent system deploys multiple specialist agents working in parallel, with each agent owning a distinct part of the workflow. The entire process is managed by a coordinator or orchestrator that actively checks the work of individual specialist agents and handles communication with the user.

• Emulating Human Workflow: While a single agent (like the Novelty Checker) is built strictly for speed and triage, a multi-agent system is designed to emulate how a thorough human patent searcher would actually approach their work. Aiming for an exhaustive strategy, wide coverage, and complex combinations.

Want to read more about our novelty checker and upcoming MAIA system release? Read this article