Understanding the Roles of AI: Co-pilots, Assistants, and Agents

In today’s fast-paced technical landscape, artificial intelligence (AI) is becoming an integral part of our daily lives, transforming how we work, communicate, and solve problems. As someone who has navigated the evolving world of startups and technology over the last 30 years, I’ve seen first-hand how these AI systems can enhance our capabilities and their evolution. I started building AI/ML systems at Microsoft in the early 2000s to fight spam.  But, with the seminal “Chat GPT moment” (Nov 2022), and similar to the other important moments in technology shifts, Netscape (1994) (web), Salesforce (1999) (SaaS/cloud), AWS (2006) (cloud infrastructure), Facebook to the public (2006) (social), iPhone/App store (2007) (mobile)…it’s important to try and get your head around where these new technical capabilities fit and how to think about applying them to applications, especially in around personal/work productivity. I've got my head deep in AI and trying to figure it out.

In this post, I’ll break down the three primary roles of AI—co-pilots, assistants, and agents—explaining their unique functions and offering insights into how they can be leveraged effectively in various contexts.  It’s early in this cycle, and our thinking at Pioneer Square Labs is evolving quickly, but I think it’s important to have a perspective and gather input to keep refining my thinking. 

The Co-pilot: Augmenting Human Capabilities

Let’s start with co-pilots. Imagine you’re flying a plane; the co-pilot is there to support you, enhancing your capabilities without taking over the controls. In the realm of AI, co-pilots operate under a similar principle. They are designed to augment human tasks by providing insights, suggestions, and tools that enhance productivity.  They work alongside us while we do the work.

For instance, PSL developed a company called Enzzo.ai, a co-pilot to design new hardware products.  The users are “conversing” with Enzzo as it collects, organizes, and aligns information and stakeholders, then generates the right kinds of outputs like requirements documents, competitive product comparisons, concept images, and AI-based personas to chat with about the product.  Enzzo has a very rich interface that combines chat with multiple AI systems and new ways to combine and display data.  Rich interfaces are a feature of "co-pilots".

The Assistant: Task Delegation Made Easy

These AI tools are like helpful colleagues who take on specific tasks based on your direction and delegation. When I think of assistants, I envision applications like Read.ai (meeting recording, transcription, and analysis (we are investors) —tools that can take instructions and execute them efficiently. These tend to run “alongside” your work (like transcribing a meeting) or run in the background to get stuff done, based on the user's assigned tasks. Other PSL “assistant” companies include Picco, taking care of routine and repetitive tasks like sale lead research, drafting outbound messages or writing weekly sales reports or Atrieon, a AI-based “program manager” that can help software teams be more productive with better planning and scoping of tasks, better communication between team members and finding and managing external resources, first with humans and then with emerging software development “agents”.  Assistants usually follow a somewhat defined workflow but have limited user experience, almost like a remote worker or virtual assistant (see my post here). Picco, for instance, is usually working in the background and uses Slack and the documents it creates as it's primary user interface.

The Agent: Autonomous Decision-Making

Finally, we have agents—AI systems that operate with a degree of autonomy. Unlike co-pilots and assistants, agents often work with inferred goals or objectives set by humans but take initiative in executing tasks without constant oversight. Think of Waymo’s self-driving cars or Roomba vacuum cleaners; they navigate their environments and make decisions based on their goals, without a prescribed way to achieve the goal. They are likely to have even less of a user interface.

Agents are particularly fascinating because they represent a step toward greater autonomy in technology. They can analyze data and respond to situations in real time, which can be incredibly beneficial in scenarios where quick decisions are necessary. However, it’s essential to understand that while agents can operate independently, their effectiveness largely depends on the quality of the data they receive, the clarity and understanding of the goal, and the algorithms that drive them.

Real-World Implications and Applications

Understanding these distinctions is crucial as we integrate AI into our professional lives. Each role—co-pilot, assistant, and agent—has its strengths and ideal use cases. As someone who has implemented various AI tools in my work processes, I can attest to the transformative impact they can have when used appropriately.

For instance, for software development, Github Co-pilot or Cursor, (co-pilots) have revolutionized the coding experience adding massive productivity for this role. Meanwhile, employing an assistant like Bloks (kudos to my friend Marc Gingras) during meetings allows me to focus on discussions rather than note-taking and will produce better investment memos, a focused "assistant" task. Lastly, utilizing agents for routine tasks—such as scheduling or data analysis—frees up valuable time for more strategic thinking.

As we continue to embrace these technologies, it’s essential to remain aware of their capabilities and limitations. By understanding how co-pilots enhance our skills, how assistants streamline/do our tasks, and how agents operate autonomously based on our explicit or implied goals, we can make informed decisions about which tools best suit our needs.

The landscape of AI is rich with opportunities for enhancing productivity and efficiency across various domains. By recognizing the distinct roles of co-pilots, assistants, and agents, we can harness their capabilities effectively to improve our workflows and outcomes. An oversimplification might look like this;

Takeaways

As we develop direct products, we categorize them based on the level of effort required for user experience, AI direction, goal setting, and success communication. And as we move forward into an increasingly automated future, let’s embrace these tools with an open mind and a strategic approach—after all, the right AI can be a game-changer in achieving our goals!

Here's another blogpost from Adam Loving, the mastermind behind Picco (PSL) with some additional input. Another good LinkedIn post about JACoB (PSL) (assistant) by Keving Leneway, another PSL engineer thinking about these kinds of things.