Background on AI Agents

What is an AI agent?

An AI agent can be defined as any sort of AI that incorporates some level of planning and/or tool usage (i.e. there are multiple non-predetermined steps involved or interactions with other software tools or systems).

We believe almost every piece of software will become agentic eventually -- that is to say incorporate complex artificial intelligence.

Types of AI agents

There are countless types of software that can be described as agentic or "agents." Some of the popular categories include:

• AI assistants

• AI travel planners

• AI software engineers

• AI designers

• AI research agents

• AI outbound sales reps

• Etc.

Conceptualizing agents

One of the easiest ways to conceptualize agents is as replacing sets of tasks currently completed by humans. In the future, tasks that AIs can perform as well as humans will be done by AIs instead, as they will be much cheaper. For certain jobs this may look like replacing small sets of tasks (e.g. sales research on new cold leads), and for other jobs this may look like replacing the entire role (e.g. outbound email-based SDR).

As time goes on, agents will start to get even more complex and complete tasks that would've been impossible or inconceivable for humans to do (e.g. thousands of microtransactions to exploit some trading arbitrage onchain).

The right infrastructure for agents should both: 1) empower short-term use cases that resemble replacing humans, and 2) have the capacity for more complex, unplanned use cases that will emerge as models become more advanced.