The AI Agents Are Coming — Is Your Brand Ready?
Exploring what moving on from AI chatbots to agentic AI systems might mean for brands
It seems fair to say that the first stage of the ongoing generative AI frenzy was centered around intelligent chatbots. From ChatGPT to Gemini to all the smaller chatbots, generative AI has been about serving up answers in natural conversations. The main use cases have been finding information, coding assistance, and virtual companionship. Now, a new class of AI agents is set to usher in the next phase, where AI tools will go one step further and actually do stuff for you.
AI agents are AI systems capable of independently executing tasks with minimal human supervision. While AI chatbots respond to prompts with text-based answers, AI agents extend these abilities by automating processes across applications, performing digital tasks, and, increasingly, making decisions autonomously.
On Tuesday, Anthropic introduced a new capability in its Claude 3.5 model called “computer use,” which enables Claude to interact with computers by viewing screens, typing, moving cursors, and executing commands. Here’s a demo video from Anthropic showing an AI researcher asking its agent to gather information from various places on his computer and use it to fill out a form.
Of course, filling out a form might seem like a rather mundane and unimpressive task for an AI agent to accomplish, but the ability to scale this automation tool to handle low-lift, manual tasks like this is what will elevate productivity and streamline processes.
Besides Anthropic, many players in the AI race are also busy working on their own AI agents. For example, on Monday, Microsoft announced 10 new automations for its Dynamics 365 suite of business applications, capable of handling tasks across departments without direct human input. Similarly, Salesforce’s rival Agentforce system is set to become generally available next week. Meanwhile, Google, plus a host of startups like Asana and Cognition AI, are all racing to build the “AI co-workers” of the future.
This latest wave of AI agents reminded me of the Rabbit R1 showcased at CES 2024 back in January. If you recall, the Rabbit R1 was a whimsical phone-like gadget that was designed as a task-specific web browsing assistant capable of actions like booking tickets or making online purchases. It supposedly uses a unique Large Action Model (LAM) to enable AI-driven task automation across various app interfaces on both mobile and desktop devices. Different from the current wave of AI agents, however, R1’s LAM has to be trained to recognize and interact directly with other apps’ UI in order to carry out the designated tasks. In contrast, this current wave of impending AI agents is designed to execute complex, multi-step tasks across various applications without constant human guidance or individual training to familiarize the AI with the UI of each app.
Short-Term Impact on Enterprise Use Cases
It goes without saying that we are still in the early stages of testing out AI agents, starting with enterprise use cases and backend operations. AI agents are not just about text responses; they are designed to streamline routine, often tedious tasks that typically drain employee productivity, such as appointment scheduling, customer data management, email responses, and so on. Gartner predicts that by 2028, at least 15% of day-to-day work decisions will be made autonomously through agentic AI.
Some innovation-forward brands are already testing the incorporation of AI into its operational workflow. For example, fast-casual restaurant brand Chipotle recently launched a new conversational AI hiring platform called ‘Ava Cado,’ which the restaurant says can accelerate the hiring process by up to 75%.
Meanwhile, agentic AI solutions are rapidly emerging across various enterprise applications. For instance, AI startup Inflection AI recently rolled out a new enterprise solution called Agentic Workflows, enabling its enterprise systems to take trusted actions for various business use cases.
One could argue that a narrow variant of AI agents has long existed in digital marketing in the form of programmatic ads, in which AI algorithms carry out media buying decisions with automated real-time bidding (RTB) to determine which ads to serve to individual users based on pre-set budgets and objectives, all with minimal human intervention. While not as advanced as newer agentic AI systems, programmatic advertising arguably represents an early form of automation in digital marketing, with AI algorithms handling repetitive, rules-based decisions to optimize campaign performance.
In a similar logic, the impending proliferation of AI agents in the workplace would further supercharge RTB-style automated decision-making. The common analogy that equates using generative AI tools to having a team of interns may become even more applicable. Of course, the current wave of AI agents are still quite limited in what they can do and prone to mistakes — Anthropic says that actual success rates of Claude’s “computer use” feature remain modest at 14.9% for basic computer tasks, way below acceptable levels to replace a human intern. But with time, one could see how, with improved accuracy and expanded use cases, AI agents could potentially continuously adapt to meet business objectives with minimal human oversight.
Security and privacy also remain significant concerns, particularly as AI agents gain access to sensitive data and operations. Anthropic has introduced several measures to address these issues, like data encryption, rigorous opt-in protocols, and retention policies. Yet, as AI agents are developed to understand and interact with various applications autonomously, companies must remain vigilant about data security and privacy. The risks of data leaks and potential misuse must be carefully managed to maintain consumer trust.
This also points to the famous innovator’s dilemma — the idea that established companies often struggle to adopt disruptive innovations due to their focus on existing products, markets, and customer needs. Perhaps the companies best suited to utilize this emerging class of AI agents to build a highly automated workflow are the startups and challenger brands with little to lose.
Long-Term Impact on Consumer Behavior
What’s even more fascinating to consider is how the proliferation of capable AI agents will impact our day-to-day lives and behaviors. In the long run, these AI agents could be assembled into advanced personal assistants, capable of managing complex tasks autonomously. In theory, they could create more personalized lifestyle experiences by understanding our habits and preferences.
For example, Google’s smart home integrations or Amazon’s Alexa routines can already auto-adjust lighting, temperature, and other settings based on pre-set user preferences. With more sophisticated AI agents, this type of customization could extend further, from optimizing daily routines to recommending diet and exercise plans based on real-time health data.
Obviously, brands will be looking to tap into the power of personal automation. Consider this hypothetical example of a retail brand. With agentic AI, a shopping app could become a “shopping companion,” not only offering recommendations but proactively adding relevant items to a customer’s cart or monitoring price changes. By analyzing browsing and purchasing history, as well as real-time user input, the AI agent could dynamically adjust its recommendations and actions to fit your evolving preferences, fostering a highly personalized experience that drives brand loyalty. After all, who wouldn’t want a personal shopper that really knows your likes and tastes, and is available 24/7?
Ultimately, however, mass adoption of this type of personal AI agent would still hinge on solving the longstanding challenge of bridging the AI trust gap. As I wrote in February about AI search:
Newer AI search experiences like Perplexity or Arc Search are more careful with their source citations, but still, that may not be enough to quell the concerns of some users. And even with the sources carefully cited, users may wonder about the relative trust-worthiness or, in some cases, the potential political bias, of each source. These nuances inevitably create a trust gap between AI and users, making it difficult for users to evaluate the accuracy and credibility of the information they’re being presented with.
If some people already feel this nervous about the search results that AI summarized for them, the prospect of letting AI agents handle personal decisions — such as choosing a restaurant or booking flights or buying birthday gifts — may seem even more daunting. Personal choices are by definition subjective. They can sometimes be arbitrary and influenced by a variety of contextual factors that AI agents may not have access to or interpret correctly.
The real challenge to AI agent adoption, therefore, lies in the fact that they may struggle to make optimal decisions in real-time, given that It is often in hindsight that the quality of a decision is assessed and retrospectively determined. Building trust in AI agents, then, requires not only that they deliver reliable support but also that they navigate the complex, sometimes ambiguous nature of human preferences — something that we ourselves find difficult to grasp in the moment of decision-making.
