Having a personal agent will be like having a smartphone
I think the movie Her is a really good example for a persistent agent. In the movie, Theodore installs an operating system named Samantha. At first he was a bit suspicious that it won’t work or just a trash. He gave it a shot, and after that moment, everything is changed for Theodore.
What makes Samantha interesting is, it’s persistent. It understands context, so quietly manages work that overwelms Theodore, in the background. And Samantha is what we need for our day-to-day tasks.
Having a personal agent will be like having a smartphone. You’ll unlock lots of new opportunities using a single interface.
It is easy to communicate with an agent, because we’re communicating with each other using phones over two decades. And it works, at least we invented remote work. That means we can get job done remotely.
Why not do the same with the agents?
Lets not forget trust. Trusting an agent is the next step. LLMs are not predictable, and there are lots of funny examples in the internet where an agent deleted a database, and feels sorry about it. But this is still a problem with humans too. Having a programatic guardrails will do the job.
Software companies have confirmed that coding agents are capable of writing proper and working code, and they proved it by doing massive layoffs (and continuing to do that).
After the wave of coding agents, what we see is more AI generated content and products. That means if you can write code, you can do lots of things. This is why most models are enough for our daily tasks.
Models are good at writing code, so why not let them to do our daily tasks, so we can focus on living the life we want.
We don’t need a frontier level model for our day-to-day lifes. What we need is a working and persistent agent that is easy to access and use. This will become the iPhone moment of AI.
When computers started to get more smaller and accessible, first engineers and hackers adopted them. They’ve built their custom versions, and experimented. One day business people started to simplfy computers and sold them for businesses and every-day users. Same happened for smartphones, and now happening for the personal agent. We will start to see more consumer oriented agents in the market. Grok Bot and OpenAI dots are more like techy user products, consumers need simple ones.
Today everybody has a smartphone that they use daily. It is time to give everyone their own persistent AI agent.
We have limited hours in a day, and the world wants us to do more. Not every task requires my full attention and real me. For example, ordering food is not an enjoyable task if you’re hungry. My agent knows me, and can see my order history, so it can pick something I would like.
Ordering food for one might not be a hard task, but ordering food that is specificly towards my fitness goals makes sense. It guardrails me from ordering unhealthy food. This can be applied to family management as well, where your partner might have different choices.
What I’ve experienced with my collegues at work is relatable to what is going to happen to our daily lives. We were a small team with everyone focusing on a specific area of the product. When I did a change for the API, I requested my agent to prepare a document for the mobile developer. I sent the document in markdown format, and mobile developer places that to his agent and they discuss. What we’re going to see is people will start messaging each others agents. To be honest, it will be most likely that my agent will write to my friend’s agent, so they’ll arrange majority of the tasks. You might think it will make everything sloppy, but since the agent knows its user, generated messages will be more like its owner. These persistent agents grow every day with new inputs from its user.
I’m expecting to see a new protocol for agent-to-agent communication similar to MCPs. This might be the email for agents. Exciting future to be alive.
Next one billion dollar one-person companies cannot happen without personal agents. Human owner will need at least one personal agent that acts like an assistant that organize other agents. One human cannot reach to every issue. It also not possible to manually organize each agent.
That’s why I’ve started building Tamo, and now pivoting it to become a personal agent that does what you want without making you to learn what is a model.
It acts like a human and shows you outcomes. You don’t have to learn how to do the task you want to do.
What is the first task you want to delegate to an agent?