ChatGPT's new agent 🤖

Hi Non-Techies,

Or should I say, Garr, me hearties.

For the uninitiated, that’s pirate-speak. I’m fluent now, after a wonderful time spent celebrating AIFNT on a pirate ship in Bristol last week.

But whilst I was swashbuckling my way around the River Avon, OpenAI were announcing perhaps their biggest launch since ChatGPT itself: ChatGPT agent mode.

Hoist the colours and ready the cannons (the last of today’s pirate references, I promise) - today we’re talking about what agent mode is and whether it’s worth getting excited about.

OpenAI’s Announcement.

When Sam Altman (OpenAI’s CEO) sits down on a sofa and looks down the barrel of the camera, you know something big is coming.

“People wanted a unified agent that could go off, use its own computer, and do real, complex tasks for them”, he explains in the first minute of the announcement video. Okay, that’s pretty big.

How is this different to regular ChatGPT?

ChatGPT is an incredible tool, but right now, it doesn’t take action.

For example, if I want to go on holiday (fat chance), I can ask ChatGPT for trip recommendations and a Spanish language lesson. That’s really cool.

But I can’t ask it to find me an available, affordable hotel for a specific week, create a spreadsheet itinerary, and order a giant, pink flamingo inflatable to my house in time for my departure.

ChatGPT’s agent mode is here to change that. Armed with its own computer and a web browser, it can actually do some of the stuff you’ve asked it.

Some possible use-cases:

1) Look at my work calendar and create an editable deck I can present in Monday’s meeting using the About page on my website.

2) Plan recipes for a dinner party for next Friday and order the ingredients to my house.

3) Write a whitepaper based on OpenAI’s ChatGPT agent announcement.

Blimey. So far, so mind-blowing. But what do users think?

What’s the initial reaction?

Very few ChatGPT users have access to this, but the headline from early testers is “it’s full of promise, but it needs improvement”. Classic AI rollout, then.

Matt Wolf put it to work in this video, and it didn’t exactly pass with flying colours (that’s not a pirate reference, just pure coincidence).

For example, it took 41 minutes (an eternity in AI terms) to create a pretty mediocre and error-strewn deck.

Even so, Matt acknowledges the massive potential of this tool.

This great LinkedIn post by Allie K. Miller shares a similar sentiment. Sure, ChatGPT’s agent mode needs to be polished and improved, but it’s already proving to be helpful for “anything that needs organised data”. As she puts it, we’re finally seeing that “AI isn’t just chat”.

Our take.

We haven’t used this mode yet, but the thing raising our Non-Techie eyebrows isn’t what it can or can’t do at the moment, but what it’ll be able to do eventually.

I sometimes have to remind myself that we’re only three years into this AI journey. If it’s a road trip, we’re still flicking through Spotify playlists and loading Google Maps, trying to figure out the best route through it all.

The internet was released to the general public in 1993. But do you remember what websites looked like three years after that? Here’s Pepsi’s website to remind you:

Sure, agent mode will need improvement, but we can get a real sense of what’s to come: accessible AI tools that can successfully carry out complex tasks for us, whilst we get on with other stuff.

Last week, that felt like an if. Now it’s most definitely a when.

Other things to note:

It’s unclear when it’ll be rolled out to the wider ChatGPT public, but we’re talking days, not years. There’s a chance that by the time this newsletter is sent, it’s already available for most of us.

Sam Altman’s final message in the launch video was a warning: just as society has learned to use the internet safely (for the most part), so too will it need to learn to use AI agents safely.

It just so happens that we put a big emphasis on learning how to use AI safely and responsibly over at my AI Academy. Click the button below if you want to know more about why our members love it:

Right, I’ll be back next week. If you can’t make it, please send your AI agent in your place.

Heather