AI Agents and the promise of the dawn of a new age of computing is here. But are AI agents just hype without substance right now?

Artificial intelligence (AI) has evolved from a niche academic pursuit in the 1940s to a transformative technology enabling computers to engage in human-like conversation through large language models (LLMs). Early theoretical work, such as Alan Turing’s 1950 paper “Computing Machinery and Intelligence,” laid the foundation for machine intelligence. Advancements in computational power, the development of deep learning techniques, and the creation of extensive datasets have propelled AI’s capabilities. The introduction of transformer architectures in the late 2010s further enhanced natural language processing, culminating in sophisticated LLMs like OpenAI’s GPT series, which can generate coherent and contextually relevant text, facilitating conversational interactions between humans and machines.

After nearly a century, we’ve finally taught computers to chat like sci-fi sidekicks. So, getting them to think and act on their own should be a piece of cake, right? Well… maybe.

What is an AI Agent?

A simple definition: An AI agent is a software program that autonomously interacts with its environment, perceiving inputs and taking actions to achieve specific goals without human intervention.

OpenAI CEO Sam Altman has outlined a five-level progression toward Artificial General Intelligence (AGI), each representing a significant advancement in AI capabilities:

• Chatbots: AI systems with conversational language abilities, capable of engaging in human-like dialogue. (ChatGPT, Microsoft Copilot)
• Reasoners: AI with human-level problem-solving skills across a broad range of topics, demonstrating advanced reasoning capabilities. (OpenAI o1)
• Agents: AI systems that can autonomously take actions based on user instructions or independently, performing tasks without human intervention.
• Innovators: AI that can aid in the invention of new ideas, contributing to scientific discoveries and creative processes.
• Organizations: AI capable of performing the work of an entire organization, managing complex operations and decision-making processes.

As of January 2025, OpenAI has reached the Reasoners stage with models like o1, which exhibit enhanced reasoning abilities. The development of Agents is anticipated to follow, enabling AI systems to execute tasks autonomously.

So, yeah, we’re almost there!

Agentfarce

If you’ve heard from Salesforce lately you’d think we’re already there. They’ve seemingly rebranded to “Agentforce”.

Realistically “Agentforce” is a successor system to Einstein Copilot, an enhanced chatbot of sorts that let you chat with your CRM data. Agentforce appears to be a new architecture, an API that allows customers to use various LLMs to interface with their CRM data. Both Microsoft Copilot and Einstein are chatbots (maybe reasoners) and not agents if we’re being picky. And while this is pretty cool tech on it’s own, something that would have been incredible for us to hear about 10 years ago, the promises seem… hollow.

When you look at how Copilot has been delivered to customers, it’s disappointing. It just doesn’t work, and it doesn’t deliver any level of accuracy. Gartner says it’s spilling data everywhere, and customers are left cleaning up the mess. To add insult to injury, customers are…

— Marc Benioff (@Benioff) October 17, 2024

What Marc is saying here about Microsoft is not untrue. But anyone who’s worked with Salesforce knows that Salesforce itself is just a platform. And customers need to customize SFDC for the value they can derive from the platform. This has mixed results usually. So this is classic Salesforce. With another yearly rebrand. And it comes with lots of talk about endless possibilities and business value, but very little you can actually see or use unless you’re a very large customer. This is by design.

These large companies like Salesforce, Microsoft, and others are in a race to capture enterprise spending for agents, and big promises are a part of that competition. Almost none of their promises are true at this point; they’re simply positioning themselves for a better market advantage.

Agents for Everyone

But outside of big corporations, who spend most of their time wasting time, can we expect an AI agent revolution? Are agents that can autonomously take actions based on user instructions or independently perform tasks without human intervention possible?

Probably. The rise of open-source AI agent frameworks is democratizing access to advanced AI capabilities, enabling developers and smaller organizations to build autonomous agents without relying on compute from large corporations. And these agents are real. They can research, write, and interface with APIs. So they can do all sorts of things, mix them with a tool like Zapier and you’ve got the potential for some very fun simple agents that are surprisingly powerful if you are honest and open about their limitations.

Very dumb agent ideas

For example, none of these are very good, but, they are probably very achievable with current technology!

• Celebrity Recipe Inspector: If Martha Stewart tweets a recipe, cross-check it against AllRecipes or Food Network to verify it’s not missing any critical ingredient—because nobody needs dry cake. 🍰
• Cryptid Spotting Notifier: Scan local weather webcams for anything vaguely Bigfoot-like and tweet “Breaking: Sasquatch confirmed in {Location}.” Attach blurry footage, of course. 🦶
• AI Dungeon Master Agent: Detects tweets asking for D&D campaign ideas and generates an entire quest based on the user’s posted context, complete with NPC dialogue, combat scenarios, and moral dilemmas.
• Poet Laureate of X (Formerly Twitter): Sees tweets like “Feeling sad today” and replies with a custom, heartfelt poem in seconds. Can also generate haikus, sonnets, or limericks for different moods.
• AI Gossip Starter: If two celebrities post vague tweets within an hour of each other, auto-create a rumor and post something like, “Wow, can’t believe Celebrity A subtweeted Celebrity B—the drama! 👀”

The AI Agent Ecosystem

There are a ton of tools out there to get you started creating agents. And this is probably the best way to evaluate the space right now. Take a real problem you have, evaluate it using this question:

“Does this problem require ongoing, autonomous decision-making or actions that would otherwise need constant human attention to achieve specific goals?”

Here’s an example:

• Task: Managing customer support emails.
• Question: “Does this task need to be done automatically without me constantly managing it?”
• Answer: Yes, an AI agent could automatically categorize emails, draft responses, or escalate urgent issues without human intervention, saving time and effort.

Tools and Technology for Building AI Agents

I shared this at the beginning of this post, but I’ll share it again here. There is already A LOT of new tech in this space.

I’ll go over the two I’ve spend the most time with.

LangChain is a prominent open-source framework that facilitates the development of applications powered by large language models (LLMs). It offers modular components for integrating LLMs with external data sources and APIs, streamlining the creation of versatile AI-driven solutions.

Similarly, CrewAI focuses on orchestrating role-playing autonomous AI agents. By assigning specific roles, backgrounds, goals, and memories to each agent, CrewAI emphasizes user-friendly interfaces and streamlines collaboration between agents and humans, making multi-agent systems more accessible.

These frameworks empower developers to create AI agents capable of autonomously performing tasks based on user instructions or independently, without human intervention. The open-source nature of these tools fosters innovation and collaboration, accelerating the development and deployment of AI agents across various industries.

While LangChain and CrewAI are rolling out the red carpet for AI enthusiasts with their open-source offerings, they’re also setting up ticket booths to keep the show running. These platforms are blending accessibility with monetization to ensure they can continue delivering top-notch performances in the AI arena.

For instance, CrewAI has secured $18 million in funding to advance its AI agent framework, indicating a strategic move towards sustainable growth. And LangChain has raised a total of $35 million over two funding rounds, reflecting its commitment to both innovation and financial viability.

But these open-source AI frameworks seem like the ultimate playgrounds for tech enthusiasts—offering a sandbox where you can build, experiment, and push the boundaries of what’s possible, all while discovering the delightful quirks and limitations of AI.

So Where are We?

Somewhere in the hype cycle.

In my opinion, the rush to monetize AI agents can be interpreted in two ways:

• Sign of Maturity: Monetization efforts may indicate that AI agents have reached a level of functionality and reliability suitable for commercial applications. Companies investing in AI agent development are seeking returns on their investments, suggesting confidence in the technology’s readiness for market deployment. For instance, OpenAI’s significant investments in AI development reflect a belief in the technology’s potential to generate substantial economic value.
• Premature Commercialization: Conversely, the drive to monetize could stem from economic pressures rather than technological readiness. AI start-ups often face high operating costs and investor expectations, prompting early monetization attempts even if the technology isn’t fully mature. This scenario can lead to overpromising capabilities, potentially resulting in user dissatisfaction if the AI agents fail to meet expectations. Recent reports highlight that AI start-ups are under increasing economic pressure, leading to significant talent acquisitions by major tech companies.

With all the hype in AI the last few years, the rush to monetize can be a double-edged sword. On one hand, it suggests that AI agents are ready to earn their keep in the commercial world. On the other, it might mean companies are jumping the gun, pushing products out of the nest before they’re ready to fly. So, before we roll out the red carpet for our new AI overlords, it would be wise to peek behind the curtain and see if they’re truly ready for the spotlight.

I’m going to have fun myself doing just that with LangChain and CrewAI, as well as any other frameworks that seem like fun!