While generative AI seeps into almost every industry and application imaginable, the work of the user — the prompting — still requires a good chunk of time spent by a human operator. Good prompt engineering takes time to direct natural language queries correctly to extract the best from large language models with masses of knowledge. It’s a classic ‘poor input, poor output’ scenario. While generative AI is considered a productivity win, the input — iteration upon iteration to get it right — can still take its toll on working hours. Enter the realm of the AI agent, redefining the way we interact with AI and the work we can do with it…
AI agents are akin to having another you. They’re autonomous systems capable of performing tasks or achieving goals that you can do, but don’t have the time or inclination to. They range from simple chatbots handling customer service enquiries without human help, to advanced virtual assistants and autonomous entities making decisions based on real-time data. But hang on… isn’t this just an upgraded form of Robotic Process Automation (RPA)? Not exactly. RPA uses powerful algorithms to automate tasks that are often repetitive and high in volume, and the bots are trained on a very specific set of rules to perform select tasks. No machine learning (ML) is actually going on, whereas an AI agent brings that ‘intelligent’ decision-making into the equation, performing tasks much closer to how a human might explore, iterate, and weigh up pros and cons of their actions.
Not so Secret Agents
Today, AI agents are at an advanced stage of development, understanding and processing human language, learning from interactions, and performing tasks with increasing autonomy and efficiency. Companies like Google, Microsoft, IBM, and Amazon are leading the charge, developing agents, and the tools to help others create their own intelligent assistants.
The reason agents are considered the next big thing after Generative AI is because they go beyond retrieving information and creating content; they allow us to delegate and automate a wide variety of tasks online. In the next couple of years we could all have dozens of agents performing tasks for us on our behalf, and be confident that they will be performed correctly.
Letting AIs loose on life
The great benefit of AI agents is that they’re sophisticated enough to be given a rough set of directions — in natural language form rather than code instruction — and perform them uninterrupted. I’ve been trying out a new agent called MultiOn, a browser add-on designed to handle tasks that a user doesn’t want to spend time doing. By typing a simple piece of guidance, as you would into ChatGPT or Google Gemini for example, MultiOn then performs the task, even if it means making some decisions and steps on its own.
I sent it to a PC building website to configure a budget gaming PC with only a few hardware preferences that I wanted to include. Off it went, browsing the web itself and assembling something for me to look at. It’s pretty good (but strange watching its cursor move around and make choices), and it’s even been used to order food, taxi rides and book tickets.
This ‘helper technology’ is like other recent AI developments, advancing at an incredible pace with a gold rush of developers leveraging the much sought-after AI chips in the market. Companies such as Imbue and AdeptAI are creating assistants that operate on a ‘learn once’ principle. Show them a task once and the software will be able to do something similar (with a different set of commands and goals) entirely on its own. Adept AI recently demoed how their agent can process an invoice for payment in a different system. It does this by understanding what information to look for and extract from that invoice, including understanding the type of goods that were being paid for, so that all dropdown fields could be correctly selected.
Of course, we couldn’t talk about AI without mentioning OpenAI, the makers of ChatGPT. CEO Sam Altman suggests they too are working on developing agents that can do stuff even without a prompt. Things like transferring information from a document to a spreadsheet for data analysis, handling web-based tasks like booking airfares, or creating daily schedules or travel itineraries without access to third-party APIs. This is the Siris, Alexas and Cortanas of the world on steroids. It puts IFTTT apps (If This Then That), such as Apple’s Shortcuts, into the dustbin, making them entirely rudimentary and far too much hassle to bother with.
Expect real AI agents to go so much further than the basic step-by-step action software.
New AI, new device?
It’s tempting to think that with such domination in the personal device market, iOS and Android smartphones are the be all and end all, just waiting for that next AI upgrade. No doubt future mobile softwares will lean heavily into AI, but there’s an opportunity for singularly AI-based devices to carve out their own niche.
At this year’s CES show, the Rabbit R1 was unveiled. A tiny orange box built in a minimalist style with an AI agent as the operating system, rather than simply an add-on to the OS. The R1 isn’t a smartphone, a smartwatch or a smart speaker, though it shares some similarities with parts of all of them. It’s AI distilled into a single device and it’s very clever, completely changing how the user interacts with a ‘smart’ device. Everything is voice activated and the R1’s RabbitOS is being touted as a ‘large action model (LAM)’, with an amazing ability to understand what is said in natural language and immediately take action without much more user input. This eliminates the dozens of apps, swipes and types that a user would normally make on a smartphone. Like other agents, actions include booking travel, ordering goods, drafting and sending emails, even editing photos while it simply sits in your pocket. And similar to agents such as AdeptAI’s, the R1 also has a teach mode where users can record an action it would like the OS to understand, and can then let it repeat similar actions from then on without assistance.
When agents become spies!
Whether it’s software to install on our Macs and PCs, upgrades to our smartphones, or found in a brand new category of devices, agents are about to change the course of consumer and business behaviours. However, this growing interest comes with its challenges, including some typical AI ethical considerations.
Could they be used to impersonate users without their knowledge? Or will they be forced by regulations to stop users from performing certain actions?
It’s unknown territory and another deep pool in an already vast quagmire of legal confusion. But, like all things AI in 2024, the world’s leading bodies are ramping up their efforts to ensure the right AI choices get made and users benefit from the possibilities. We’ve learnt from previous internet pitfalls, right?
Where next?
AI agents are the next big step in AI because they go beyond retrieving information and creating content; they allow us to delegate and automate a wide variety of tasks online. We’ve been prophesying about the autonomous future for decades. They are likely to become part of our daily lives and jobs. Just imagine what you could be doing with all that time back!
At TBT Marketing, we’ve been at the forefront of marketing for some of the most ambitious technology companies in the world for the last 25 years. As we enter the burgeoning world of AI we’ll be right there with them again.
