The Big Challenge with Agents Today
16:58 Share

Today on the AI Daily Brief, the big trouble with agents right now, and before that in the headlines, a record energy deal to power AI. The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI. To join the conversation, follow the Discord link in our show notes.

We kick off today with the latest in the absolute barn burner struggle to get more electricity to power the future of AI, where data center construction startup Crusoe Energy Systems have landed a massive 4.5 gigawatt energy deal. So the way that this deal works is that Crusoe has entered into a joint venture with investment firm Engine No. 1 to strike an agreement with the owner of a set of gas turbines that could generate 4.5 gigawatts of energy by 2027.

For some reference, 4.5 gigawatts of energy is enough to power 3.7 time-traveling DeLoreans or a large city like Chicago. Earlier this year, Engine No. 1 announced a partnership with Chevron to develop and scale gas power plants for data center co-location, which will be fed through into this new joint venture. Chris James, the CIO of Engine No. 1, said, This strategic partnership leverages Crusoe's strength in developing purpose-built data centers with Engine No. 1's focus on providing power to sites that support the AI-led industrialization of the U.S.,

Speed to market is the most crucial aspect of powering AI development in the United States, and the combined expertise of our two firms ensures that we'll be able to deliver that for our customers. Public filings reveal plans for a 360.5 megawatt gas plant in Texas attached to the first Project Stargate data center. That means that if the OpenAI-led joint venture does continue to work with Crusoe to get access to this energy, which presumably is the strategy here, although that's not been confirmed,

This would mean enough power to build another 12 facilities of the same size as that first Project Stargate data center in Abilene, Texas. Amir Afraadi summed it up nicely, that's a crazy bananas amount of energy. Next up, Intel's newly installed CEO Lip Bhutan has discussed his recovery plan for the iconic chipmaker, and it doesn't seem to involve breaking up the company.

For quick background in case you missed previous coverage, Tan is a four-decade veteran semiconductor investor that previously served on Intel's board. He walked away from that position last year reportedly over disagreements on how to right the ship. And according to Reuters reporting, Tan is considering major changes to Intel's manufacturing method and AI strategies.

The new trajectory includes restructuring the company's approach to AI and staff cuts to address what Tan views as a slow-moving and bloated middle management layer. Revamping the company's manufacturing operations, which at one time only made chips for Intel but have been repurposed to make semiconductors for outside clients such as NVIDIA, is one of Tan's core priorities. This directly addresses one of the critiques of previous CEO Pat Gelsinger, who many felt didn't take the necessary steps to restructure their workforce.

Semiconductor industry expert Dylan Patel told Reuters that Gelsinger was "too nice" and that he "didn't want to fire a bunch of middle management in the way they needed to." In the near term, Tan aims to improve performance of the company's manufacturing arm by aggressively seeking new customers. He also plans to develop chips for AI servers and perhaps even explore software robotics and AI foundation models. The medium-term goal is to release a new AI chip each year, but sources say the first iteration will likely take until 2027.

Tan's plan is very similar to one that Gelsinger started in 2021. He also aimed to pivot to AI and expand the customer base. Ultimately, though, plans aren't just about ideas, but about execution, and execution was lacking. Intel ended up posting an annual loss of $19 billion last year, their first since 1986. Overall, it still feels like a pretty last-ditch pull-out-all-the-stops effort to save the company. Last week, the Wall Street Journal wrote, "...they say Liputan is the best hope to fix Intel, if Intel can be fixed at all."

Many are skeptical. Take him, the author of the NVIDIA Way, writes, "...before Intel shareholders get excited about the prospect for an annual AI chip cadence, they may want to see one market-viable AI chip offering first." Finally today, Elon Musk's XAI has acquired text-to-video startup Hotshot. The small team has shipped three video foundation models over the past two years, playing in the same space as OpenAI's Sora and Google's Vio2.

Announcing the news, Hotshot CEO Akash Sastry wrote, training these models has given us a look into how global education, entertainment, communication, and productivity are about to change in the coming years. We're excited to continue scaling these efforts on the largest cluster in the world, Colossus, as a part of XAI.

Back in January, during a video game stream, Elon mentioned his company was, quote, working on Grok video and that it would be released in a few months. And basically all anyone can think is that we are about to get some seriously unhinged AI video coming soon. For now, though, that is going to do it for today's AI Daily Brief Headlines Edition. Next up, the main episode. Today's episode is brought to you by Vanta. Trust isn't just earned, it's demanded.

Whether you're a startup founder navigating your first audit or a seasoned security professional scaling your GRC program, proving your commitment to security has never been more critical or more complex. That's where Vanta comes in. Businesses use Vanta to establish trust by automating compliance needs across over 35 frameworks like SOC 2 and ISO 27001. Centralized security workflows complete questionnaires up to 5x faster and proactively manage vendor risk.

Vanta can help you start or scale up your security program by connecting you with auditors and experts to conduct your audit and set up your security program quickly. Plus, with automation and AI throughout the platform, Vanta gives you time back so you can focus on building your company. Join over 9,000 global companies like Atlassian, Quora, and Factory who use Vanta to manage risk and prove security in real time.

For a limited time, this audience gets $1,000 off Vanta at vanta.com slash NLW. That's V-A-N-T-A dot com slash NLW for $1,000 off.

Today's episode is brought to you by Super Intelligent and more specifically, Super's Agent Readiness Audits. If you've been listening for a while, you have probably heard me talk about this. But basically, the idea of the Agent Readiness Audit is that this is a system that we've created to help you benchmark and map opportunities in your organizations where agents could

specifically help you solve your problems, create new opportunities in a way that, again, is completely customized to you. When you do one of these audits, what you're going to do is a voice-based agent interview where we work with some number of your leadership and employees to map what's going on inside the organization and to figure out where you are in your agent journey.

That's going to produce an agent readiness score that comes with a deep set of explanations, strength, weaknesses, key findings, and of course, a set of very specific recommendations that then we have the ability to help you go find the right partners to actually fulfill. So if you are looking for a way to jumpstart your agent strategy, send us an email at agent at besuper.ai, and let's get you plugged into the agentic era.

We talk a lot about agents on this show, but if you've ever thought to yourself, I don't want to talk about agents anymore. I just want to actually build and deploy something. I'm really excited to share something special with you today. We've partnered with Lindy to offer companies that just want to dive into the deep end of agents, a way to get their feet wet, a way to move fast and build something meaningful without breaking the budget.

The first five companies that email me, nlw.bsuper.ai, with Lindy in the title, will have access to work with Lindy to build an actual functional agent serving their specific needs for under $20,000. Some of the agents you can build include a customer support agent, maybe automating responses on your website.

You could build an SDR for generating or qualifying sales leads, or you could build an agent that's perfectly suited for your internal communications needs, be it note-taking, scheduling, or something else. Not only is Lindy structured to integrate with all of the places that you already keep data and information, it's also a full extensible platform, which means as you hire more and more agent employees and really build out your digital workforce, Lindy is going to enable those agents to be interoperable and basically be able to work together in a seamless way.

So again, if you are interested in diving in all the way to agents in a matter of weeks, not months, not years, email me nlw at bsuper.ai, put Lindy in the title, and let's get your first digital employee online.

We are now about six months into the agent era. At least if you start it from the point at which commercially available agents started becoming available and big companies started going all in on agents as the next big thing. Now, I do think that we'll look back and think that the agents of this era were very much in their infancy.

There is, of course, a blurriness around what the term even means. A lot of things that we might have called automations a year or a year and a half ago are now called agents, which, as I've stated before, I'm fine with. I think it all conveys the fact that assistant AI is using AI to do things while agents are AI doing things for you. At least that's how most people are thinking about it. In any case, being now six months in and having at least one public company that's one of the main purveyors and pushers of agents gives us a chance to start reflecting on where the space is and what the challenges are.

The company, of course, I'm talking about is Salesforce. And recently, the information wrote a long piece about why Salesforce is having a bit of a tough time converting their agent force offering into value. Certainly, if you just listen to CEO Mark Benioff, you would not think there were any problems. He's declared 2025 the absolute year of agent force and said, we've never seen products grow at these levels. Then he said,

Then again, on that same earnings call where Benioff said that, it was left to CFO Amy Weaver to give a more realistic and sober outlook, saying that she expects, quote, modest agent-forced sales this year. Weaver also added that the company's overall sales growth would be between 7% and 8%, which is their slowest ever. So what is the challenge? In short, it seems to be that while Salesforce has gone all in on agents, companies are still working to get themselves there.

A sales manager who spoke with the information basically just said that many customers aren't ready to commit to using the software. And they're certainly not thinking about it as Salesforce is pitching it, as digital labor that essentially replaces humans for certain tasks.

Some of this is a question of a general state of readiness, and companies just wrapping their hands around the assistant era of AI and now being forced to move into agents before they really feel like they have their feet under them. In other cases, however, there's more specific challenges. Hallucinations remain an issue, and in fact, even a bigger issue when there's not a human sitting there manning the tool, and there's also data connectivity and compatibility issues. The information writes, some customers aren't in a good position to use agent force because they need to connect the product to databases stored outside Salesforce to work properly.

Basically, because this is in Salesforce's ecosystem and relies on Salesforce's data silo, that makes its applicability more challenging in certain cases.

Of the 5,000 AgentForce deals that have been closed since AgentForce was debuted in late October, 3,000 of those are paying customers, with the other 2,000 trialing the product for free. Adam Mansfield, a practice leader at Upper Edge, which is a SaaS procurement consultancy that negotiates Salesforce deals, said that the paid deals included bundles where customers may or may not be using the product. Now, the information piece sort of paints a picture of a company that's getting a little bit desperate.

Mansfield again said that, quote, Salesforce representatives have raised the prospect of meaningful price increases for other Salesforce products unless customers agree to use AgentForce. Salesforce managers have also indicated to customers that they can avoid fees for using more storage or software licenses than they signed up for if they purchased AgentForce. And when it comes to the challenges here, it's clear that one of the core issues for agents right now is cost. And unfortunately, this challenge with price is operating on multiple dimensions.

First is, what's the right way to price these tools, conceptually at least? And second is, once you figure out the right concept, is the price too much? Using the Salesforce example, in mid-2023, the company announced AI Cloud, a service that could generate emails, marketing materials, and customer support messages. It was basically a Salesforce-branded GPT wrapper, which we were dismissive of for a while, but obviously now there's an entire industry built around what we might have dismissively called GPT wrappers just a little bit ago.

That AI cloud product was priced at $360,000 per year plus additional fees for usage. And when it was finally delivered in the second quarter of 2024, it seems like it didn't really get any appreciable traction and has been basically forgotten. Now, how much that's because the company was already focused on shifting two agents isn't exactly clear. But when it came time for agents, the cost structure of agent force is based on usage rather than a fixed price. Salesforce charges $2 per conversation that the agent handles.

Now, on top of that, customers are also required to migrate to Salesforce's cloud service or develop their own code to integrate with AgentForce, meaning there's some other hidden costs as well. But even just all on its own, this idea of the cost being based on number of interactions and the specific amount per interaction, neither is clearly the right answer.

First of all, on the one hand, switching to an opportunity-based pricing does at least create a new rationale for how the product is priced that makes more sense perhaps on the face of it than the per seat model that SaaS has run on for the past couple of decades. But there are challenges there. It basically creates an incentive for people to want to use the tool less, given that each usage or deployment of the tool costs them money. The second issue is the $2 itself, which on the face of it just seems like quite a bit.

If you imagine 10 customer service interactions in an average hour, Salesforce's price isn't looking all that much lower than the equivalent human labor, especially with business process outsourcing systems that have that labor distributed all over the world. Also, frankly, other companies that are competing like Intercom are charging less right out of the gate.

Now, interestingly, Salesforce is not the only company struggling with price. Manus is the viral Chinese agent that everyone's been talking about for the past couple of weeks. And interestingly, some information reporting suggests that maybe that $2 price is a little bit less arbitrary than it seems. According to the information sources, which are two people with direct knowledge of the situation, for use of their models, Manus pays Anthropic about $2 per task on average that goes through the agent. Now, Manus' tasks are fairly complex.

but that's still quite a bit. And right now the company isn't charging anything. Now at this point, Manus is trying to keep its costs down by just limiting the number of people who can use the tool, but obviously that's not a perfect situation either as the company wants to build usage. So the question is, what's going to make this situation work better? Well, for one thing, as I discussed yesterday, it seems very likely to me that the AI price wars are going to have a big impact on this.

One of the short-term impacts of those price wars should be the cost of intelligence coming down. And with the cost of intelligence coming down, these agent companies can potentially serve up organized intelligence in a way that's more cost-effective. One challenge to that, though, is, of course, that agent companies are going to have a very strong incentive to always use the most state-of-the-art model, meaning that perhaps the price gains and savings from the price war will be a little bit less, as that's likely to impact near state-of-the-art but not total state-of-the-art models more quickly.

I think the other question, though, is whether this really is a problem of price. What are the chances that this is just about companies not quite being ready for this slate of products? I think that you could go through and do a pretty compelling and clear list of what Salesforce is doing right and what Salesforce is doing wrong. On the wrong side, as we've discussed, it seems like the specific pricing is a bit too high. I also think that when you look at their actual offerings, Salesforce might be trying to sell too many agents and have a little bit of a confusing offering when it comes to agent force, which

rather than, for example, just really hammering SDRs. And by extension, I think they might be slightly overpromising what they can deliver right now. We also discussed the challenge of only being able to access Salesforce cloud data, which I think could be seen as a wrong thing, but also to the extent that the job of agent force is to bring more customers into the larger Salesforce ecosystem, maybe it's just the cost of doing business.

On the what are they doing right side, though, clearly, A, the company is willing to disrupt themselves. Employees are noting that the urgency with which agent force is being engaged is unlike anything they've seen in the 26-year history of the company. They are going absolutely all in on this, at least from a narrative and rhetorical perspective. And I wonder if when we look back at this period, Salesforce frankly knew that they were coming into this market early and were just willing to do so.

It may be that they are trying to claim territory now, that they will deal with their lumps from Wall Street for having half of those 5,000 agent-forced subscriptions not actually being paid for or not actually being used, and that this land grab, this early experience, the repetitions they're building with customers and understanding where customers are, the deeper understanding they're going to get about which use cases these agents are actually good for, could end up proving itself to be a really worthy effort, even if they are ahead of where the market is by 6 to 18 months.

This show is called The Big Problem with Agents Today. And ultimately, the question is, is the big problem the price or the fact that we're just still a little early? Food for thought as we leave you. That is going to do it for today's AI Daily Brief. Until next time, peace.