The Agentic AI: What Comes After ChatGPT. Attorney, This One's for You

View on LinkedIn

If you're a lawyer and think "AI agents" have nothing to do with you, think again. Unless you're a programmer with years of experience, what you know about AI is mostly theoretical. I understand you've mastered prompting and use tools from OpenAI, Anthropic, and Google.

But it's like being a great home cook — you prepare amazing dishes, yet you're far from a Michelin-starred chef.

What is an AI agent? Imagine an assistant that doesn't just understand, but observes the real world and acts on its own. It's not the typical back-and-forth question-and-answer of a traditional LLM.

A simple example:

You ask: "What's the weather like today?"

A traditional LLM would say: "I don't have access to real-time information." End of story.

But with agents:

• AI #1 (the main agent) recognizes it needs up-to-date data

• It activates AI #2 (a search specialist) to retrieve the information

• It answers you: "It will rain today — bring an umbrella"

Now let's look at something more complex: "What time should I leave to arrive on time at Restaurant X?"

• AI #1 checks the weather forecast (which affects traffic)

• It activates AI #3, connected to navigation apps

• It responds: "Given the forecast rain, leave 45 minutes early. Avoid the main avenue"

It's like an orchestra where every musician is an expert on their instrument, with a conductor who knows everyone's capabilities. And here's what's fascinating: humans are versatile — we speak languages, play sports, work, cook — all within a single "system."

LLMs are different. Instead of one model that does everything, we have multiple ultra-specialized agents:

• One expert at translation

• Another at sports analytics

• Another at recipes

• Another in your professional field

It's like comparing a multi-discipline athlete to ten specialized Olympic athletes. The former is versatile, but in any given discipline, the specialist will outperform them.

The only limitations are access to tools, processing costs, and restrictions imposed by external systems.

Key Concepts:

ReAct: Like a chef who tastes, thinks, adjusts, and repeats.

Tree of Thoughts: Explores multiple possibilities before deciding, like a chess player.

Chain of Thought: Shows the step-by-step reasoning process, as in mathematics.

Data Stores: Access to up-to-date information beyond the model's original training.

Orchestration Layer: The "brain" that coordinates everything, maintains context, and decides which tool to use.

Extensions vs. Functions:

• Extensions: direct control

• Functions: instructions for others to execute

And here's the most interesting part: traditional LLMs are showing signs of hitting a plateau in their growth curve. Despite using more data and computational power, improvements are becoming increasingly marginal. It's like an athlete who, no matter how hard they train, reaches a point where significant performance gains become elusive.

But this is precisely where agents make all the difference. Instead of trying to build a single all-encompassing model, the industry is pivoting toward this orchestra of specialists. It's a paradigm shift: away from the race for ever-larger, more powerful models, and toward the creation of smarter systems that combine specialized agents.

This strategy is not only more efficient — it is likely the next great leap in AI. While individual LLMs may be approaching their current limits, agent-based systems are just beginning to show their potential.

It's as if we had reached the limit of what a single musician can do, and were now discovering the power of a full symphony orchestra.

The future lies not in larger models, but in more intelligently connected ones. Agents are the next frontier that will sustain the momentum of this fast-moving industry.

This technological leap is pivotal for legaltech solutions currently in development — solutions that will almost certainly transform our profession in the years ahead.