Learning Library

← Back to Library

AI Assistants vs Agents Explained

Key Points

  • AI assistants (e.g., Siri, Alexa, ChatGPT) are reactive tools that wait for explicit user prompts and perform tasks like information retrieval, content generation, or scheduling based on those commands.
  • AI agents are built on the same large language models but act autonomously after an initial goal‑setting prompt, designing their own workflows, using external data and tools to achieve objectives such as optimizing sales strategies.
  • The main distinction between the two is that assistants require continual user direction (a “tennis match” of prompts), whereas agents operate proactively without ongoing hand‑holding.
  • Assistants can be improved through prompt tuning and fine‑tuning with task‑specific examples, allowing them to handle repetitive tasks more accurately over time.
  • Both technologies are reshaping the future of work, with assistants serving as routine task helpers and agents acting as strategic “strategists” that drive outcomes independently.

Full Transcript

# AI Assistants vs Agents Explained **Source:** [https://www.youtube.com/watch?v=IivxYYkJ2DI](https://www.youtube.com/watch?v=IivxYYkJ2DI) **Duration:** 00:06:48 ## Summary - AI assistants (e.g., Siri, Alexa, ChatGPT) are reactive tools that wait for explicit user prompts and perform tasks like information retrieval, content generation, or scheduling based on those commands. - AI agents are built on the same large language models but act autonomously after an initial goal‑setting prompt, designing their own workflows, using external data and tools to achieve objectives such as optimizing sales strategies. - The main distinction between the two is that assistants require continual user direction (a “tennis match” of prompts), whereas agents operate proactively without ongoing hand‑holding. - Assistants can be improved through prompt tuning and fine‑tuning with task‑specific examples, allowing them to handle repetitive tasks more accurately over time. - Both technologies are reshaping the future of work, with assistants serving as routine task helpers and agents acting as strategic “strategists” that drive outcomes independently. ## Sections - [00:00:00](https://www.youtube.com/watch?v=IivxYYkJ2DI&t=0s) **AI Assistants vs Proactive Agents** - The speaker explains that AI assistants are reactive tools that wait for user prompts to perform tasks such as scheduling or answering queries, while AI agents are proactive systems that autonomously pursue goals, illustrating the distinction with examples like Siri, Alexa, and ChatGPT. - [00:03:11](https://www.youtube.com/watch?v=IivxYYkJ2DI&t=191s) **AI Assistants vs Agents Comparison** - The speaker contrasts AI assistants, which depend on user input for routine tasks like customer service and code generation, with AI agents that can access external tools, retain memory, and tackle strategic roles such as automated trading and network monitoring. - [00:06:24](https://www.youtube.com/watch?v=IivxYYkJ2DI&t=384s) **Stars, Assistants, and Agents Collaborate** - The speaker highlights that movie stars, their assistants, and agents not only work together but also understand the most effective ways to do so, even when interpersonal dynamics are difficult. ## Full Transcript
0:00Imagine you're a movie star with  both an assistant and an Agent. 0:03You know, as chatting to  both of mine just yesterday. 0:06Sure you are. 0:07Now, assistants, they help  with scheduling and agents 0:12are more proactive, seeking opportunities for you. 0:15Artificial intelligence works much the same way. 0:18You have A.I. Assistants and A.I. agents. 0:21And in this video, we are going to explore 0:23how both of these types of A.I.  are shaping the future of work. 0:27Right, and our fundamental level. 0:29I think we can say the main difference is  the air assistants that reactive, right? 0:33They're waiting for commands  like a prompt from the user. 0:36While A.I. agents are proactive, they're  acting autonomously to achieve a goal. 0:40So when I say we get into it. 0:42Let's do it. 0:43So let's start with A.I. assistance. 0:47These helpful little apps understand natural  language, and they're great for doing things 0:52like organizing information or  responding to customer queries. 0:56Examples that we know are Siri, Alexa or chatGPT. 1:00Right. And most A.I. assistants, they're built on  something called launch language models, or LLMs, 1:07that allow them to understand  natural language commands. 1:11Now they rely on something called  prompts from users to take action, 1:17which means they need well-defined instructions. 1:21And with those well-defined instructions,  air assistants can make recommendations, 1:26fetch information, and even generate content. 1:29But they are always waiting for your  input to get started and you have to 1:33continue to direct them like a tennis match. 1:37Prompt response. 1:39Prompt Response. 1:41Prompt. 1:43Response. 1:43Think I get it. 1:44And there's a lot we can do to improve  the quality of these prompt responses. 1:51And for example, assistance may improve  over time through prompt tuning, 1:56which is adapting the underlying  model for a specific task. 1:59And we can also teach an  assistant some new tricks. 2:03That's called fine tuning. 2:05We can fine tune these LLM  models with specific examples. 2:10In that way, it can get better  performing repetitive tasks 2:13like drafting documents based on  patterns that it's learned now. 2:17A.I. Agents, on the other  hand, they act independently. 2:22You know, they take initiative, they break down  tasks and find the best way to achieve a goal. 2:28Right? 2:28Right. They don't need the the  constant handholding of user prompts. 2:33They're still built on large language models,  but they can design their own workflows. 2:40They do need a prompt, but just an initial prompt. 2:44Just one to get it started, telling  the agent what we want it to do. 2:48So, for example, one good prompt might  be a goal of optimize our sales strategy. 2:54The agent doesn't need further promise. 2:55It can use external data, tools and  reasoning to make decisions autonomously. 3:00So if A.I. assistants are your helpers  taking care of routine tasks, A.I. agents, 3:07are your strategists proactively driving outcomes? 3:11Exactly, Amanda. 3:12And to quote Elvis Presley, "a little less  conversation, a little more action, please." 3:17I thought you were going to do the voice. 3:18You do not want to hear me doing the voice  Now, I agents, they can use external tools. 3:23They can use external data sources, whereas  assistants, they depend on user input. 3:27And agents can also have persistent memory, 3:31meaning they remember past actions and improve  future decisions based on those experiences. 3:37So let's compare use cases assistants. 3:41They excel at tasks like like customers. 3:45Customer service. 3:46Customer service. 3:47Okay. 3:48Virtual assistants like chat  bots and code generation. 3:54Yeah. 3:55Code gen is a good one. 3:57So for example, in customer service,  assistants use machine learning to quickly 4:02analyze large amounts of customer  data and then respond to queries, 4:06often reducing the time that us humans  need to spend on boring, repetitive tasks. 4:11Thank goodness. 4:12And A.I. agents, they thrive in more  strategic roles like automated trading. 4:18If I got this, I like automated trading  in finance and network monitoring. 4:25Yeah. Right. So let's take  automated trading for an example. 4:29Our agents analyze vast datasets filled with  data like historical trends and current news, 4:35and then use that information to extract insights. 4:38And those insights predict how the market will 4:40behave and execute trades in real time  based upon these predictive algorithms. 4:45Super cool. 4:46So assistants, they handle your  routine tasks while agents. 4:51They take on more complex, high  level challenges and agents. 4:56They can scale across multiple  tasks all of the same time 5:00without any human intervention, making them  ideal for dynamic and ambiguous problems. 5:05Speaking of problems, though, A.I. assistants  and agents can have limitations like brittleness, 5:13when slight changes to prompts lead to errors. 5:17Yes, beware the rabbit hole. 5:19Agents in particular may get stuck  in feedback loops, or they might 5:24require significant computational resources, 5:26making them expensive to run without supervision. 5:29They can go down all sorts  of weird and wonderful paths. 5:34Always good to check those A.I. outputs, 5:36but there's improvement happening  every day and as A.I. agents improve, 5:41we'll likely see them tackling even more complex  problems without needing human assistance. 5:46And we're already seeing  improvements in model reasoning, 5:49meaning that agents will become more  reliable and effective over time. 5:53Take, for example, OpenAI's o1 model which  performs reasoning at inference time. 6:00So to recap assistance, they help with your day to 6:05day tasks while AI agents take on a more  autonomous approach to problem solving. 6:11And it's not necessarily an either or  here, because as these technologies evolve, 6:16expect to see more synergy between assistants and 6:19agents combining their strengths to  tackle both simple and complex tasks. 6:24It's the movie stars, assistant and  agents not only working together, 6:30but also knowing the best way to  do so, even if you don't nice you.