Why AI agents are the future of services
This week, OpenAI made two new releases that will bring us closer to a future of autonomous AI agents - GPTs and Assistants API.
In this article, we'll look at what that world could look like and analyze five reasons AI agents will be the future of services.
A day in the future of AI agents
Maj is nervous.
It's June 2025, and she is starting as a product manager at the green energy company Wind Corp.
On her first day, she logs into Slack, and the company's onboarding AI bot Kai, welcomes her to the company.
With a smattering of friendly banter, Kai walks her through the basics of working at Wind Corp.
" Are you ready for some first discussions with your teammates? "Kai asks with a smiley emoji. "Sure," Maj answers. With her permission, the AI agent looks at the shared company calendar and suggests 5 discussions for her first week.
Kai also gives helpful talking points to her new colleagues based on their public company profiles.
Later that day, Maj has another chat with Kai. She's looking to understand the sales process at the company. In their chat, Kai provides an overview and points her to a few presentations in the company's files that could help.
That evening, Maj chats on her phone with her personal trainer AI agent Clara. With access to her calendar and knowledge of her workout goals and history, Clara suggests a quick HIIT workout at 6.30 pm tonight. The AI agent also suggests ordering a healthy takeout meal since the shopping agent notified her earlier that food was running low.
Why AI agents are the future of services
Let’s travel back to today and the announcements that OpenAI made - GPTs and the Assistants API.
GPTs are versions of ChatGPT that anyone can train with natural language to perform specific tasks. They can be a creative writing coach, personal trainer, or a company's onboarding guide. GPTs can access any background information, including files, they are given and remember conversations. They can also talk to other services and potentially other GPTs.
The Assistants API gives any developer access to the same powerful features of GPTs and new capabilities launched with GPT-4 Turbo.
The API enables any developer to build AI apps that:
can take independent action on the user interface - for example, show pins on a map on the UI (as showcased in the Dev Day)
have access to files shared with it
remember conversations
These capabilities are a first glimpse into the future of independent AI agents.
AI agents are a big part of the future of services for five reasons:
Independent action - AI agents directly help people achieve their goals
Multimodality - Agents can hear, speak, see, listen and create
Intercommunication - AI agents can talk to other agents, services and devices
Access to information - AI agents know us and the world
Scalability - AI agents are easy to build and scale
1. Independent action - AI agents directly help people achieve their goals
I play in a recreational tennis league. The most annoying part of the league is finding slots for games that fit people's calendars and are available at one of the tennis courts in my city.
With my oversight, an AI scheduling agent could find overlapping slots by asking people and browsing the different tennis center websites. It could also communicate the agreed-upon times to all players.
AI agents will increasingly help people directly achieve their goals. This could overtake the old-fashioned way of browsing different apps and websites using graphical user interfaces.
With Open AI's release of the Assistants API, AI agents can also act on the websites people visit. This capability helps the AI break out from the confines of the chat window. If you're on a car-buying website chatting about how to build the car of your dreams, the AI agent can switch the car's interior on the fly based on your wishes.
To release autonomous action well, responsible builders of AI agents will still give people ultimate oversight of the agents.
2. Multimodality - AI Agents can listen, speak, see, and create
To help people achieve their goals, AI agents can listen to speech, talk with text-to-speech, understand images (and soon videos) and create things from text to images and spreadsheets.
In the latest launch of GPT-4 Turbo this week, OpenAI merged all of these capabilities into one chat. You can upload images to ChatGPT, chat about those images, or ask the AI to create a new image based on your chat. To access real-time information, ChatGPT can also browse the web.
In addition to multimodality, the emerging capability for intercommunication - the ability of agents to talk to other agents and services - will unlock unimaginable opportunities (and risks) for service innovation.
3. Intercommunication - AI agents can talk to other agents, services, and devices
The first version of generative AI has been about using tools in isolation. ChatGPT can help you write a cover letter, or Midjourney can whip up a picture of a cat riding a kettlebell to Mars.
AI agents can now talk to other agents, services, and devices.
This new skill means that to achieve any given goal, an AI agent can request actions or pull in information from anywhere it can access. As the number of services and devices with APIs continues to skyrocket, AI agents can connect to an ungodly amount of connected things to get stuff done.
From our earlier story, May's shopping agent talked to the fridge and food delivery service to ensure she had a healthy meal at the right time.
Intercommunication raises several profound questions.
Which companies will own the handful of critical agents we use in our work and free time? The owners of these agents will have enormous power as the coordinators of our lives.
How will we ensure that people have ultimate control of independent agents talking to each other? As a society, we must weigh the benefits of these agents making our lives easier and the potential negative ramifications of giving them too much autonomy.
4. Access to information - AI agents know us and the world
So far, companies have trained AI tools like ChatGPT on the public internet and any additional context given by the user or developer.
Soon, AI agents will access unprecedented amounts of information about us and the world through multimodality and intercommunication.
They can read our calendar, Slack communication, and work files if given permission. The AI agents will know our homes' temperatures and body fat percentage.
All of this information will mean they can serve us in incredible ways and unlock value previously unimaginable. As individuals, we must ponder how much access we are willing to give in exchange for convenience. Companies deploying AI agents must ensure that users have granular controls and transparency on privacy.
GPTs from OpenAI open up a global ecosystem of easily trainable AI assistants - which potentially grow up to AI agents
5. Scalability - AI agents are easy to build and scale
These capabilities are supercharged because building and deploying AI agents is getting effortless.
With Open AI's release of GPTs, anyone can build an AI GPT with natural language and share it with the world. These are not quite autonomous AI agents, but a step in that direction.
An ecosystem of GPTs will surface the best assistants worldwide. Expect Google to release something similar in the coming year.
The Assistants API will give these and other capabilities to developers.
Access to highly scarce AI talent will not be the bottleneck for deploying independent AI agents for most companies and use cases. Those finding success with AI will likely experiment quickly within a clearly defined and safe sandbox.
The picture of AI agents is downloading
When I first got on the web in 1995, our dial-up modem would load pictures slowly in roughly 20 percent slices.
This moment feels similar for AI agents. This week, OpenAI showed us the top fifth of the picture of the future of AI agents.
In the coming years, we will see the complete picture: increasing autonomy, access to more information, and communication with increasing devices and services.
How will we design services with AI agents that solve real problems, delight people, and respect their privacy and agency as humans?
Maj checks her phone one more time before sleep
Maj is sipping her morning latte at home, preparing for her second day at Wind Corp.
She glances at her phone and sees a notification.
" Hey, you might want to check out this news article, " her onboarding assistant Kai pings.
The local paper interviewed the CTO of her company about the changing EU regulations related to wind power. This article will help her prepare for her discussion with him later today. But before that, her parenting agent reminds her to prepare for her parent-teacher Zoom call.
Want to make sure you catch my next article?
This article was published on my weekly newsletter on how to use generative AI for design and innovation.
