AI agents: the new personal assistants or just digital interns on day one?

Imagine a world where your AI assistant plans your vacation, handles your emails, and even learns your quirks — all without a single back-and-forth. Is this future already here? I tested the latest AI agents to find out.

As mentioned in our recent podcast, one of the main predictions for 2025 is that AI Agents will be one of the main driving forces in making our lives more convenient.

As businesses use automation from fields as broad as finance to healthcare, what about on a personal level? Do these AI agents have what it takes to streamline the way we go about our daily business?

Let’s find out.

Firstly it’s important to distinguish what an agent is. You probably know by now what a GPT is. Well if not, it stands for generative pre-trained transformer, which is also known as an LLM, or large language model. The user experience is usually along the lines of asking it a singular question, and it gives back a response best suited for that question.

An agent, however, is a mechanism that has a complex workflow. Think of it as a multi-tasker or an outlet that can facilitate breaking down tasks and showing you a transparent and logical workflow.

It can basically prompt itself instructions in a loop effect by adding more tasks as it deems necessary while working on the go. Instead of being a linear tool, it runs tasks in parallel, more like an actual human would.

I put a few of them to work, in a very “road test” kind of way.

Microsoft Copilot

The first port of call almost caused me to abandon ship entirely and that, unsurprisingly, was Microsoft Copilot. Originally released in Spring 2023, this AI companion has so far failed to make waves, but still, why not see what it’s all about?

Upon installing the app, Copilot offered me four different voices: Grove, Canyon, Wave, and Meadow. These voices specialize in unpacking thoughts, being creative and philosophical, writing stories, brainstorming ideas, and exploring the world together.

As a person with very little patience, I found it infuriating to be offered style over substance without any context whatsoever. After opting for Canyon, I decided to cool my melon and treat it like an intern by asking it what it could do.

“What tasks can you do?” I asked him, and he immediately dubbed himself my virtual sidekick. Nice move, going straight for my ego, I thought.

When I asked him if he could book me a holiday, he said he couldn’t but that he could help me plan one. He was kind of like an AI agent then, but not in the strictest sense. Whatever, I gave it a whirl.

After a while, the frustration of constantly going “back and forth” between talking to the software and myself was cortisol-inducing.

When I asked it to plan a trip to South East Asia for September, it started talking about the nice contrast between the aquatic archipelago of Halong Bay, the ancient wonder of Angkor Wat, with the nice city vibe of Singapore. “Sound good?” he asked. Sales spiel 101.

I was supposed to be relaxing on my sofa and immediately felt like I was talking to the dumbest salesperson who had just scrolled Wikitravel on his vape break.

I asked it to “tell me some shit I don’t know” to which it started yapping about the wild nature of the Mekong Delta in southern Vietnam. Theoretical info, again in a chatty, hip-hop style. I felt like I’d had enough already.

But then I told myself not to launch into the intern on their first day, practice some tolerance, and see what comes about.

So when I asked it to give me 50 different ways to use cumin spice, it managed to sprinkle out a list in a very pretty font, quite different from how Chat GPT seems to churn out content.

The two apps are essentially quite similar, but Microsoft Copilot offers suggestions on the home screen, such as advice for buying a car or giving a home workout. It reminds me of the men's magazines I used to read when I was fifteen.

Copilot may not be an agent in the strictest sense, but it blurs the lines of GPT and agent more ambiguously than Chat GPT which uses external task-based agents as a kind of “add-on.”

Agent GPT

Next up is the vastly different Agent GPT, which is so heavily task-based that I was taken aback. This is an actual AI agent and a decent barometer of where the future is headed.

The basic plan allows you to do five demo agents a day, which feels like enough on paper, if used wisely. Otherwise, it's $40 a month for 30 agents per day.

The first thing to be aware of is that Agent GPT was predominantly designed for coding solutions and so is a very logical worker. However, you don’t need to know coding as you would for Auto GPT or Langchain, which are also quite geared towards businesses.

I started out by asking again for it to plan a holiday to Southeast Asia in September, and it got to work straight away. The five “tasks added” show how effective this is:

*Research popular destinations in Southeast Asia

*Compare prices and reviews of different options

*Book the preferred flight and accommodation package

*Check for visa requirements

*Check the weather conditions

As the agent realized when working and “scraping the web” – it added extra tasks to its to-do list, much as a human would. Though feeling like a prototype, it's a promising work-in-progress. And while it’s certainly very theoretical in its demeanor, with a few licks of paint on the design front, Agent GPT could well become a savvy agent on the market.

Socra

Next up was Socra, which is a very chatty agent. I fed it something outlandish, the purpose being that its quirky style could really be put to the test here.

Brazenly, I claimed that I wanted to become a celebrity chef and that I had next to no cooking experience. It told me that “that is quite a journey” and then asked me three follow-up questions:

1. What attracts you most to becoming a celebrity chef? – The cooking, the fame, the creativity, or the business aspect? – I replied, “The fame.”

2. How much time can you dedicate to learning cooking each week? “An hour”

3. Do you have access to a well-equipped kitchen? – “Basic essentials. Bog standard.”

I was already starting to enjoy the playful backtalk, and then it mentioned “fast-tracking the process,” which was particularly relevant for such a ludicrous task.

It advised me to “start an Instagram or TikTok NOW!” and to “document the zero to hero journey.” I felt glad that it was getting to know me well already, seeing me as a bloody zero.

The agent mentioned narrowing my focus on perfecting one signature dish and creating a unique angle – e.g., “rookie rising” or “60-minute chef.”

It then offered to help me start a signature dish, set up a social media strategy, or create a weekly learning schedule.

When I chose the first, it surpassed all my expectations by offering three choices – this agent loves the power of three, don’t we all – “rainbow grilled cheese, 24K gold french toast, or cloud eggs on toast”

I instructed it to give me something more normal, yet a bit weird and it offered up inside-out grilled cheese, midnight spaghetti, potato tornado and clear-dump cake. The last one had me howling.

Upon being asked for all these options, it then broke them all down into a four week plan and told me to film them during natural lighting conditions. I was gobsmacked, as it felt similar to some of the banter that goes on in our office.

I didn’t have enough credits for the next step, but I could get 20X more output for just $10 a month.

This very quirky AI agent would work wonders for a young audience, especially a budding entrepreneur.

AGI Layer

This well-oiled machine markets itself as a system designed for automating GPTs. It describes itself as fully able to automate tasks and workflows. Its three agents are Trinity, Sonny, and Flynn, and they can deal with research, visuals, or coding.

The visual setup looks like The Matrix and is aesthetically pleasing for sure.

At the time of writing, there was no free demo available but the AGI layer basic package was available at $125. Further packages at $200 and $300 are also available. I would be intrigued to see where this goes next.

Auto GPT

Auto GPT is powered through Open AI and has a similar prototype vibe. Interestingly, it promotes itself as working with an AGI layer, so I was able to get a tester after all.

This was found by searching for this particular GPT through Open AI’s very own Chat GPT, which functions as a plug-in.

Going back to planning a trip to Indonesia in September, it immediately identified eight tasks that needed doing. These were reeled off effortlessly and you can prompt the agent to do the tasks in its default order or for it to prioritize certain aspects.

It asks you to push a number from 1-3 to deliver further output, much like getting through to the Barclaycard call center in the old days.

Now, obviously the agent isn’t in the market to actually book your flight or visa appointment for you – we’re nowhere near that benchmark yet, but it could still give a very thorough itinerary for island-hopping, strong accommodation options and even a decent packing list.

So, if you feel overwhelmed planning such a holiday or carrying out any logical task that seems like an utter minefield, these agents won’t be able to do anything directly for you.

But by prompting them, they in turn can prompt you back, hence making us more efficient at breaking down tasks. Not what I was expecting at the beginning, but it’s better than covering your desk with post-it notes.

So, despite the landscape being completely messy and these markets subject to change like the wind, these AI agents are definitely ones to watch out for.