The “Age of Agents” Is Upon Us

Welcome to the “age of agents”

For anyone paying attention to AI trends in general, and OpenAI’s direction in particular, the company’s recent developer event was very revealing. It managed to feel both surprising and expected at the same time.

Surprising in the range of and variety of the new innovations, APIs, and capabilities revealed from the stage. Expected in that the big focus of the day – the introduction of a new toolkit to create customized AI assistants – had been clear for months.

Whether you call them “assistants,” “agents”, or something else, these intelligent, specialized chatbots have emerged as the next great transformative wave of AI. They can perform rigorous analysis, learn on the job, and demonstrate both precision and imagination.

Before this event, the narrative around AI assistants was that they held great promise, but were difficult to build and tune. Now, OpenAI has provided a structure and solution set that changes that conversation significantly. With the launch of GPTs, building intelligent domain-specific agents has become much easier.

The fact that OpenAI, the single most influential and important company in AI at the moment, has anchored its product strategy and direction so deeply in agentic development is meaningful. It signals a decisive shift for the industry.

I see us entering into an exciting period in AI development. Call it the “Age of Agents.” And what separates these agents from what came before is that they are much more tuned and defined to a specific purpose. Their personality, capability set, and knowledge is all optimized to serve a particular community of users taking on a certain range of problems.

Power lies in specialization

ChatGPT took the world by storm because of its revolutionary ability to interact with people in natural language and synthesize information quickly from many sources. It was and is the ultimate general-use AI tool. It has very broad applications. In contrast, this new era will be defined by specialization and depth.

GitHub’s Copilot and Sourcegraph’s Cody are great examples from the world of coding. They won’t do all your coding for you, but these agents use their deep context to make it much easier for you to get more done. They know the files you are working on and can suggest what code to insert. They are optimized for developers and their needs.

Intercom’s Fin is another agent that uses deep domain knowledge to be very effective. It is a support agent that answers questions interactively based on product documentation, providing a vastly better and more focused user experience. Instead of being forced to parse through documents and FAQs in search of a response to their specific question, visitors can get the exact thing they want almost immediately.

Just recently, my team released Ironclad Contract AI, our own dedicated-purpose agent for managing and analyzing contracts. The solution is capable and useful to our users because it understands not just the language and structure of contracts in general, but because it has access to all of their contracts specifically. So it can spot opportunities, perform analysis, and offer suggestions tuned to their needs.

The lesson is clear. There won’t be “one conversational agent to rule them all,” a single all-capable AI, but instead many different agents that specialize in particular domains.

What makes them effective is their adaptation. They are specialized in their background knowledge, personalities, and tool usage. They aren’t built for everyone but are highly optimized for a specific scenario and user base, and that makes them potentially incredibly useful.

Making the most of intelligent agents

What will this shift towards specialized agents mean for you? It is a huge opportunity to empower your people and transform the way you get work done. You can expect agentic solutions to hit the market that offer powerful new capabilities that push out the limits of what is practical and possible for your team. This type of specialized agentic AI represents a fundamentally new model for assisting users and solving problems, offering designers and development teams fresh possibilities and areas in which to innovate. We will see agents that target every industry and virtually every business function.

Of course, not all of these solutions will fit your needs, and not all will be effective or well-designed (there’s plenty of bad software out there… why would AI be any different?). But if you take the time to explore your options, you are likely to find some difference-making solutions.

For most companies and teams, it will make more sense to evaluate and acquire outside solutions rather than try to build your own. While the cost and complexity bar for developing agentic AI is lowering, it remains very, very difficult to optimize and tune agents for maximum impact. It can be relatively easy to get a basic prototype chatbot developed now but making it work well across different scenarios and users is intricate, difficult work. Agents that you acquire are far more likely to be deeply context-aware and tuned to the problem area than anything you try to build yourself.

As you assess them, don’t expect one conversational tool to do everything. The real challenge is pulling together a set of solutions that support the distinct things that your people need to do, and possibly allow them to do brand new things that are today out of reach. Evaluating solutions like this that are so outside of your previous experience, and so distinct from anything currently available, requires patience and imagination – but the results can be worth it!