Neural Notes: Is the AI agent hype just gen AI all over again?

Welcome back to Neural Notes, where I look at some of the most interesting AI news of the week. In this edition: AI agents are taking over the enterprise world – with HubSpot, Salesforce and Microsoft all peddling their shiny new wares simultaneously.

For almost two years the ‘AI revolution’ has been force-fed to tech and business spaces. Big tech has been battling it out for market share and new ‘AI startups’ have popped up in droves, hoping to ride the wave.

And who can blame them? Despite the current economic climate, the VC dollars keep flowing into AI.

In the first six months of 2024, over US$35 billion was invested in AI startups. Just this week Salesforce doubled its AI fund (again) to US$1 billion and Microsoft is said to be planning a US$30 billion AI megafund with Blackrock.

But let’s be clear, the vast majority of noise around AI has actually been about gen AI, specifically content generation – with varying degrees of accuracy and quality.

And it is not new technology, but you wouldn’t know it from the way it’s been shoved into the mainstream. But it did become more accessible through the launch of ChatGPT and those who quickly followed. Since then, it’s had the subtly of a freight train, with everyone clamouring to get on board.

But as gen AI has become increasingly inescapable, one could argue it has been worn thin by overuse and inflated expectations.

Add in growing concerns around privacy and transparency, you can see why “we’re using generative AI” is starting to sound less like a brag and more like a liability.

The rise of AI agents: Automation without the prompts

While everyone was busy debating gen AI, a new player was quietly entering the scene. They were lurking in the background for a while — a mere whisper that turned into a roar this week in the US.

In just 24 hours, HubSpot, Salesforce and Microsoft all announced AI agents integrated into their platforms. A startup that utilises AI agents even won HubSpot’s Million Dollar Pitch competition.

The positioning is clear: AI agents are the next step for businesses, automating workflows, handling customer service queries and even sales outreach.

AI agents are being promoted as more intelligent and capable, able to handle complex, multi-step tasks. Perhaps most importantly, they’re able to work autonomously.

Unlike regular generative AI, they don’t need constant human prompts or intervention. They can perform tasks on their own.

At HubSpot, co-founder and CTO Dharmesh Shah is already thinking bigger. During his keynote at the company’s Inbound event, he talked about a future where agents build other agents, predicting an AI revolution that would echo Apple’s App Store moment.

“Their vision was embodied in the phrase, ‘there’s an app for that’. Our vision for the future is ‘there’s an agent for that’ – for every marketing, sales, customer service, use case imaginable; including use cases we can’t imagine yet,” Shah said.

But here’s the question — are AI agents an exercise in distancing oneself from generative AI and the sometimes-negative connotations that can come along with it?

HubSpot’s executive vice president of product Andy Pitre doesn’t think so.

“I don’t think that people are choosing to build agents purely from a branding perspective,” Pitre said to SmartCompany.

“Agents have become the standardised name for AI that walks people through multiple steps in service of a customer.”

Why AI agents might not live up to the hype (yet)

There’s no doubt AI agents will be useful. But they are still a long way from being a panacea for all business tasks. As we have seen with gen AI, there’s a fine line between solving routine problems and overpromising their real-world impact.

Too often, companies are left with underutilised tools that don’t really change much.

Microsoft’s announcement of Copilot agents, for example, touts their ability to improve productivity across apps like Excel and Teams. Salesforce’s Agentforce claims to handle customer service tasks with no human involvement.

That all sounds fantastic, but we’ve been here before. Wasn’t generative AI supposed to revolutionise how businesses created content and communicated? The reality hasn’t always matched the hype.

Sure, AI agents are likely to be useful, especially for smaller businesses where automating even simple tasks can be a game changer. But we’re still early in this journey.

Bumps along the road are inevitable for any kind of new technology, but they’re more like boulders with AI when you factor in its additional baggage such as ethical, regulatory and legal concerns.

AI agents could follow that same trajectory, especially if companies focus more on marketing them as a silver bullet than on their real-world practicality.

AI agents, for all their promises, are still a work in progress. Yes, they’ll likely handle routine tasks well, but when it comes to complex scenarios where human judgment and context really matter, we’re not there yet. And the likes of HubSpot and Salesforce do acknowledge that.

In the short-to-medium term, we’ll probably see AI agents being used effectively in specific, well-defined areas like customer service, marketing, or basic sales outreach. But it’s important to temper the excitement with realism.

If you’re going to buy into any kind of AI, let it be because of what it can actually do for you right now – not just the potential it’s often sold on.

Other AI news this week

HubSpot took a cheeky dig at Salesforce during its Breeze AI announcement.
But to be fair, Salesforce CEO Marc Benioff had already fired a shot at Microsoft a day earlier, calling Copilot ‘Clippy 2.0’.
Apple Intelligence has entered public beta if you’re keen to give it a whirl.
Our sister site Crikey has a scoop on Harrison.ai, alleging its AI was trained on patient’s scan data without consent.
LinkedIn farmed user data in the US for training before updating its terms of service. Cool and normal.