Consumer Agents Signal New Phase for AI Growth

What is the state of play with consumer AI agents?

I think we are at the beginning of a paradigm shift. We are going from a conversational relationship with these agents to a much more action-oriented relationship. Obviously, there are questions in terms of how much trust you place in these autonomous programs and how much you integrate them into your daily life.

If you log on and provide an agent with access to your calendar, passwords, and credit cards, you are reducing friction in the program's functionality. If consumers can get over the trust factor, as well as the security factor, that integration will improve dramatically. You could direct an agent to buy you tickets to a major sporting event, find a hotel in, say, Portugal, and confirm and respond to appointments.

And that's when you flip the script on the utility layer of the consumer landscape.

What does that entail?

By this, I mean executing more complex tasks, which generate a higher level of tokens, which are the basic units of data, such as text, that AI models process and generate. More ambitious tasks will be executed by advanced models at the bleeding edge, with subscription tiers for consumers.

Will AI agents still be free to use, like search?

The mass market for AI agents will be monetized with advertising and subscriptions, much like the way the web operates today. You can buy a premium subscription if you don't want to see ads, and if you do see ads then you are on a basic tier and you're not paying for a subscription.

In our research, we have talked about how AI will move from the infrastructure layer to the platform layer and then to the application layer. We think the advent of consumer-facing, action-oriented agentic commerce marks the emergence of the platform layer.

You highlighted advertising. Is AI helping online companies with customer acquisition and support?

Yes, definitely. The creation, placement, and measurement of ads are primarily human-curated activities, even today. We believe AI and machine learning applications across the advertising stack can deliver more efficient and higher-performing ads at a lower cost.

If ads are more effective, do they improve profit margins for online platforms?

They do, but at the moment you can imagine that a lot of cash is getting reinvested back into the build-out cycle behind AI. So seeing that incremental profitability, to some degree, is tougher.

Speaking of the build-out, will the debate around the risks of AI slow down development and affect capital expenditures?

The concern among industry employees regarding risks was a major theme at the conference. In terms of what happens next, it's a little early to have a great read on what this debate means. It was interesting to see that many industry leaders sort of agree with the headline idea, which is, "Let's slow the rollout of AI," or "Let's slow the pace or cadence at which we're going to push the frontier forward."

When I've looked at industries that go from unregulated to regulated, typically the devil is in the details. Who or what is going to regulate the industry—a government body or a non-governmental organization? What is going to be their mandate?

One of the themes that came through at the conference pretty clearly is that most of the compute coming online to support AI is aimed at the supply-demand imbalance that exists today, not some future build-out three, four, or five years from now. And keep in mind that most of those infrastructure projects have already been contracted.

So your forecast hasn't changed.

No. Our initial take is that the capex cycle through the end of 2027 is likely to remain elevated and in line with our estimates, which are higher than the Wall Street consensus. My colleagues and I estimate that US hyperscalers will deploy $1.4 trillion in capital in 2027. There are constraints up and down the supply chain, including access to memory chips and issues relating to power and land. So even if there isn't a slowdown, there are external factors that could act like a headwind and change how much capital is actually deployed.

What dynamics should investors focus on as we move into the platform layer?

We received a lot of comments at the conference about the need for deflation in the unit pricing for tokens to drive mass adoption.

Meaning companies, and ultimately consumers, should pay lower prices for AI tasks?

Yes. In every technology cycle I've analyzed, you need two things—the price of the offering needs to have healthy levels of deflation, and the level of utility must rise. Then you get a lot more folks moving from experimenting with the technology to embracing it in a much wider way.

Picking up on the utility point, how can we see or measure that?

It's important to remember that with AI, there are internal and external components to the technology. On the latter point, think about product creation. Companies are already beginning to use AI in creative ways to engage with their customers and reduce churn. If a platform offers better tools, it's less likely you'll leave for a competitor and more likely you'll be willing to pay for those services.

These are dynamics that are very, very constructive in terms of the impact on the whole marketplace. I think there is an interesting iterative loop around AI services that will be a catalyst to watch in the quarters and years to come.

This article is being provided for educational purposes only. The information contained in this article does not constitute a recommendation from any Goldman Sachs entity to the recipient, and Goldman Sachs is not providing any financial, economic, legal, investment, accounting, or tax advice through this article or to its recipient. Neither Goldman Sachs nor any of its affiliates makes any representation or warranty, express or implied, as to the accuracy or completeness of the statements or any information contained in this article and any liability therefore (including in respect of direct, indirect, or consequential loss or damage) is expressly disclaimed.