Agent Update
Exploring the Mind Shift Happening in Media
In meetings and book talks, I'm often asked about agents, their makeup, and their significance. This post explores agents as a rapidly developing area of AI and their potential implications for media. In this context, I broadly define media as any entity—human or AI—that’s a reliable information source.
Describing new things is inherently challenging. We rely on familiarity to grasp new concepts, but using outdated language can hinder rather than aid understanding.
Consider the phone. For Boomers and Gen X, a ringing phone held a sense of anticipation, a vehicle for important calls or chats with loved ones. Today, phones are synonymous with texting, posting, scrolling, and commenting. Talking to someone is often an afterthought. Research suggests one in four Gen Zers actively avoid incoming calls. A phone’s use has expanded so much that the term's meaning differs from what it was.
Now, let's turn to another term shifting in meaning: agents. They represent a mind-shift happening in media, both literally and figuratively. Historically, agents evoked images of human intermediaries: real estate agents, travel agents, and secret agents. You could add journalists, financial analysts, teachers, and even doctors. They imply someone acting on behalf of another, with trusted expertise to guide us.
The term takes on a radically new dimension with AI. In computer-speak, agents are programs designed to perform tasks autonomously or semi-autonomously. It’s an intense area of AI development changing how software is engineered. Maya Murad's video below for IBM, "What are AI Agents?" is an excellent primer on agent design.
Their potential impact is threefold: they can boost productivity, radically accelerate speed to knowledge, and open doors to previously unthinkable tasks—much like how advanced search revolutionized web browsing.
We rarely consider how utterly unthinkable the web search story is: Two college programmers turn a PhD thesis into Google, one of the most influential companies ever. Behind a stark white page, systems scan over 400 billion web pages in its index. Based on keywords, page quality, and user intent, algorithms rank and return results estimated at 5.9 billion per minute. It presents the most relevant pages in seconds.
A machine became a new lens into the universe of knowledge. It’s something armies of people couldn’t possibly do.
While maintaining a wide lead in monthly web traffic, Google’s stronghold is slipping. Its primacy faces new competitors and challenges. The massive influx of capital into AI ventures turns a monolithic search into personalized, guided discovery.
This migration disrupts the traditional hierarchy of expertise in search results. Previously, journalists, professors, and healthcare advisors were undisputed human authorities. Now, with AI's ability to synthesize vast amounts of data or uncover groundbreaking insights, machine intelligence will complement the experts.
Sometimes, they may replace human expertise as agent teams who work on our behalf. Machine agency, described below by DeepMind founder Demis Hassabis, has mind-blowing potential.
An influx of new agents that comprise such machines could become part of the Google ecosystem or spawn different software platforms and marketplaces like this. For now, thousands of developers are engineering agents that find, interpret, and package information for us. They will operate with increasing autonomy, handling tasks from information discovery to interpretation and presentation.
The scale of what’s in development may surprise you. For size, in 2023, generative AI represented $29.1 billion in startup investment across nearly 700 deals. As of July 2024, approximately 67,200 firms work in the generative AI sector. TechCrunch noted that 35 new start-ups raised over $100 million each this year. This suggests a surge in activity alongside gargantuan investment from the major players, who paid out $52.9 billion last quarter.
The Vibe Shift Coming to Media
Tech CEOs like Jensen Huang of NVIDIA and Marc Benioff of Salesforce also evangelize immense potential in agentic systems, where multiple agents collaborate to accomplish tasks. To do so, a 2022 Google Research paper popularized chain-of-thought prompting. It offered a blueprint to accelerate agentic computing by breaking complex problems into smaller, manageable steps, enhancing a system's problem-solving ability.
We now see a Cambrian explosion of AI seeping into our digital lives. Agents are being embedded across consumer-facing devices (Apple Intelligence), software (Microsoft Co-Pilot), platforms (Gemini-powered Google), and productivity apps (Notion AI). These are significant developments, but from a media perspective, there’s more to see here.
Today, agents function as chatbots, virtual assistants, recommendation systems, and code. Soon, we'll interact with increasingly powerful AIs via voice, AR overlays, avatars, and robots. Marshall McLuhan is famous for saying the medium is the message. As a new medium, information agents disrupt the idea of what a source is, where knowledge comes from, and the form it takes.
Discomfort will accompany the mass deployment of agents as information sources. How will we relate to artificial entities encroaching in every direction? Why should we trust information when we don’t know its origin? Who (or what) will we seek out that’s best suited for us? Might agents create AI sludge that buries quality information in a mass of crap content? Will agents stop us from thinking on our own?
Feelings like fear, trust, overwhelm, and agency are sensibilities in transition. Another stance skews hopeful, with psychology bending towards curious, opportunistic, and empowered. Evan Armstrong, lead writer at Every, notes the impact of new web navigation, framing it as a vibe shift. He says:
"By making a large language model the principal technology in a search engine, vibes-based search evolves the process beyond carefully chosen keywords and star maps to something more direct and actionable.
But because it understands how we think, it can also help us find what we want without knowing exactly what we are looking for. Put another way, it represents an evolution beyond information and answers to understanding. He adds: "Vibes-based search isn't merely about initial result efficacy. It's about co-creating knowledge."
What Armstrong implies is discovery merges with action. A system trained to do so is OpenAI o1, a new reasoning model. Its release page features experts testing the model for coding, economics, and genetic engineering problems. Tyler Cowen, an economist and author of Marginal Revolution, said of it working on a PhD-level problem:
“It's pretty good. We're just figuring out what it's good for. Or at least, I'm just figuring it out. Obviously, it's a new capability. But the fact that you can ask it any economics question, and it has a good answer, is really quite significant.”
OpenAI is becoming a multi-agent platform and marketplace. In the case of GPT-4o and OpenAI o1, there are marked differences in what they can do. According to Open AI, human trainers were shown anonymized responses to prompts to determine the human preference for the GPT-4o generative model and the new reasoning o1 model. People then voted for which response they preferred. The o1-preview was preferred to GPT-4o by a large margin in reasoning-heavy categories like data analysis, coding, and math.
This could broadly affect how generative AI is used in fields requiring advanced problem-solving and analytical skills—like media.
More Signal, Less Noise?
How might agents displace or augment media specifically?
Using recent history as a guide, hypertext introduced new media forms, like search engines, blogs, tweets, and newsfeeds, creating new industries and roles. Similarly, agents will unlock new means of interaction, creativity, and business design.
We've traversed the next hypertext moment. What might it look like? Here are three hypothetical examples of value agents could deliver, touching on election information, buying a car, or seeking mental health guidance before seeing a doctor.
For election information: As an undecided voter, what shapes your perspective on a presidential candidate today? You seek information from articles, debates, and conversations. You might get triggered by a meme, text chain, or video snippet from a cable news analyst. Most you see, regardless of side, is overwhelmingly biased. So you enlist an AI agent to cut through the bullshit. It scans through news, articles, and debates, cross-referencing sources to spot patterns of bias and lies. It focuses on the candidate’s character, record, policies, and fitness for the job. The agent filters out performative nonsense, disinformation, and attack ads. It highlights which candidate aligns with your values and needs. It's an unbiased researcher guiding you through today's election morass.
Shopping for a car or truck. An agent determines what you’re looking for and then aggregates reviews, specs, and customer feedback with a critical eye. It checks the reliability of each review, questioning whether the manufacturer sponsors a glowing write-up, if a critique comes from a competitor, or if a shit post is from an opportunist aligned with a particular brand or industry figure. The agent instantly analyzes all available data, focusing on what matters to you—brand, price, environmental impact, safety, or technology. It updates recommendations continuously to help you decide once you’re ready to buy, the closest dealer, and the available inventory.
Paths to Improve Mental Health. If dealing with depression or anxiety, an agent navigates through the flood of online information, including research studies, articles, and expert opinions. It filters through treatment options like mindfulness practices, therapy, or medication. It assesses the credibility of sources, ensuring that advice comes from reputable sources. The agent surfaces alternative therapies or support groups. Tailored options offer an informed path forward, suggesting evidence-based practices, therapies, and available doctors that fit your needs.
In every sphere—news, politics, shopping, health, and more—AI agents can serve as impartial, interpretive companions. They can be thoughtful guides, assisting in finding balanced, accurate information and making it feel less overwhelming.
As a medium with a digital mind, agents are categorically different from what came before. The term's evolution from human intermediaries to AI-powered entities mirrors the broader AI revolution. Just as language adapts to reflect a changing world, technology evolves as a driving force behind those changes.