Agentic AI Will Change Everything
How realistic is the idea that AI-powered agents will change the way we work, play, and live? The answer: it's already happening. (Issue #159)
Before we get to today's main topic, some miscellaneous goodies and things worth your attention…
Ne Zha 2, a Chinese-made hit film in China, is having box office success typically enjoyed by U.S. movies. One of the most interesting bits of coverage from The Economist ($) is how seeing the movie has become a patriotic act:
To boost consumption during the holiday, local governments gave out cinema vouchers to attract more moviegoers. Once it was clear the film could break records, more people rallied to see it. Schools took students, and firms stopped production so employees could attend. One cinema in Sichuan province said that it would hold off screening the recently released “Captain America: Brave New World” in order to boost “Ne Zha 2” sales. “Our Chinese animation deserves to be seen by the world,” it said.
Speaking of movies, La Profesora and I saw Black Bag with Cate Blanchett and Michael Fassbender at our local Lake Theater & Cafe last night. It's great! In addition to a terrific story, the Lake Theater ambience added a lot: good wines and microbrews, real food. The popcorn was fresh. The theater didn't smell bad. Plus, La Profesora dislikes the onslaught of commercials (not the previews, although she disagrees with my characterization that previews are "free samples") and, regardless of how charming Maria Menounos is, the extra-loud, extra-salesy "Noovie Show." Instead of ads and Noovie, the Lake Theater treated us to a few minutes of Charade, starring Audrey Hepburn and Cary Grant. Some people groaned when it was time to turn Charade off and start watching Black Bag. (We may watch Charade on Amazon Prime soon.)
The Power of NPR. Bob Mondello's NPR review of Black Bag was the first I'd heard of the movie as I was driving home from the gym. I pulled over, grabbed my phone, found the review on the NPR app, and texted it to La Profesora. That's not the only recent moment where NPR has influenced our purchase decisions. We spent thousands on a new bed at Saatva. We would never have heard of Saatva if the company didn't advertise on NPR, and Saatva's support of NPR was a genuine factor in our decision to buy from them rather than elsewhere. We also needed new sheets, and I remembered hearing about Boll & Branch on NPR, so we looked there, saw some good quality bed linens at good prices, and bought. Advertising on NPR, at least for us, is a good indicator of high quality.
Scam Alert: You might enjoy this Future Commerce podcast that includes an expose of the fake "My Dad Was the Highest Paid Nutritionist" campaigns that many TikTok creators are making.
I didn't know what to make of this WSJ ($) article by Mary Harrington, "Female Celebrities Are Making Millions on OnlyFans. Will They Regret It?" If the author had been Mark Harrington, I would have thought it sexist. As it is, it feels a bit finger-wagging, pointing out problems without presenting solutions, but how much can I expect from a lifestyle piece?
Finally, "Why I Left the Washington Post" by Ruth Marcus in The New Yorker ($) is disturbing and important.
Practical Matters:
Sponsor this newsletter! Let other Dispatch readers know what your business does and why they should work with you. (Reach out here or just hit reply.)
Hire me to speak at your event! Get a sample of what I'm like onstage here.
The idea and opinions that I express here in The Dispatch are solely my own: they do not reflect the views of my employer, my consulting clients, or any of the organizations I advise.
Please follow me on Bluesky, Instagram, LinkedIn, and Threads (but not X) for between-issue insights and updates.
On to our top story...
Last time, I shared a microfiction (1,000 words or less), a short science fiction story called Piercing the AI Wall about an executive who had surrounded himself with a barrier of Agentic AIs that prevented any strangers from reaching him, and the elaborate lengths that a woman named Trix (who I've written about before) went to break through that algorithmic barrier.
This time, I'll explore how realistic the story is or isn't. You don't have to read Piercing the AI Wall (although it ain’t bad) to understand this week's piece, but fair warning: Thar Be Spoilers Ahead!
Let's dig in.
Agentic AI Will Change Everything
To understand the microfiction, you don't need to know in detail what "agentic AI" is. As a level set this time: an AI agent is a program that does more than provide information. Alexa, Siri, ChatGPT, Perplexity, Co-Pilot and the rest are all digital assistants: these programs answer questions and do tasks that you manage closely. For example, I used Gemini to generate the image above. (You can see the prompt I used at the end.)
Agents do more. These programs go out into the world (mostly the digital world, at least for now), perform tasks, and then come back to report on what they did (if you care).
For example, here's how AI agents might change online subscriptions, which today are a terrible experience. Every time I order dog kibble online, Chewy.com pesters me to change my one-time purchase into a subscription where dog food will automatically show up at whatever interval I specify. Chewy even offers me a small bribe to do this.
The problem is that my dogs are not predictable in how much they'll eat on a given day, so a subscription would inevitably lead either to lots of extra kibble lying around or running out early and my having to scramble to the store. Subscriptions are generally a benefit to the merchant, not the customer.
But an AI agent could monitor the amount of kibble that I have for each dog (yes, these high-maintenance Welsh Corgis each have their own kibble, sigh) through a sensor or a camera where I keep the big containers of kibble, notice when I was running low on one kibble, order it in the background based on the true conditions of my need rather than a subscription, and I wouldn't realize that my agent had done this until the kibble showed up on my door. I'd never run low or have too much.
This is a simple example. A more complex one is planning a vacation, where an AI agent would act like an old-fashioned human travel agent. A couple planning a trip would say to the AI agent, "here are the times we can travel." One partner says, "I want to go someplace warm." The other says, "I want some place with interesting culture and food." The first partner says, "but I don't like spicy food." "Oh!" the other partner says, "and dancing. There has to be dancing." An AI agent would take all this in, propose an itinerary, adjust the itinerary based on feedback from the couple (e.g., "I just went to Mexico on business; somewhere else, please"), and then, when all parties are agreed, go out and book the flights, make the hotel and dinner reservations, buy the museum tour tickets, and schedule a night of dancing.
Neither of these examples are a reality today, but they will be within at most a handful of years. Already, AI agents are acting within complex but defined environments. For example, a few weeks ago at the LiveRamp conference I saw an astounding live demo by John Hoctor of Newton Research, which has built a team of AI agents to help media people buy advertising. One Newton agent might analyze the data around a campaign and suggest tweaks when the human media buyer asks questions in everyday language. Then, another Newton agent goes out and makes the tweaks to the media buy.
Despite that long explanation, this Dispatch isn't an explanation of AI Agents, there are lots of smart people doing that elsewhere. (I recommend following Jeremiah Owyang to start.)
Behavioral impact on people and businesses
My interest in AI Agents is in how they will change human behavior, which they will.
We've seen this many times before and long before the digital revolution. For example, back when answering machines were a new thing, at first they were a convenient way to leave or get a message when you were away from the phone. Then two behaviors emerged.
First came call screening, where the person receiving the call could wait to hear who was calling and then choose not to pick up. "Him? Good God, not now. I'll call him back later, or maybe I won't." It was an exciting new frontier in rudeness!
The second new behavior was checking messages remotely. I had one friend, Wendy, who would ask to borrow my phone whenever she visited so that she could learn if somebody had called her home. (This behavior vanished in the age of smartphones that we carry with us all the time.) One of the original intents of the answering machine was to ease the anxiety around missing an important call, but anxiety is like water... it flows into any available space, so Wendy had an early version of FOMO.
Agentic AI is going to allow similar new behaviors to emerge. Part of my purpose in writing "Piercing the AI Wall" was to explore what those behaviors might be.
In the microfiction, Edmund Thoreaux's digital agents made it easy for him to become a recluse, so Trix manufactures awareness of her employer's company by creating "The Digitalis Project" and then planting images of the foxglove flower (digitalis is the Latin name for foxglove) around Thoreaux until he becomes curious. (And yes, I named Thoreaux after Henry David Thoreau.)
As human behavior evolves in dynamic dialog with Agentic AI, that will make some things difficult, especially in marketing.
Let's go back to the Agent that will help me buy dog food. I've already decided what kind of kibble to buy. More accurately, my two dogs have decided what kinds of kibble they'll each deign to eat. If I have an agent refilling my twin supplies of Hill's Science Diet in the background, then how will Purina or Mars or Royal Canin or Diamond Naturals coax me into giving their brands a try? Answer: they won't.
As longtime readers might remember, for years I've argued that the overwhelming majority of our everyday decisions are low stakes. Agentic AI will make it even easier to offload unimportant decisions (which, again, is most decisions) to algorithms.
"Where do you want to grab lunch?" "I don't know; what's convenient?" My agent and my buddy's agent will confer. In an instant they'll balance our different locations, restaurant preferences (loud or quiet? is there easy parking? a good beer selection?), dietary restrictions, time of day, traffic, and where else we each need to be later. (Maybe that beer isn't a good idea, and therefore not a relevant criterion, because Brad has a long drive after lunch, my agent concludes in a patronizing and unsettling and did I mention patronizing way.)
For low-consideration purchases in particular, Agentic AI will shorten by a lot the half-life of brand equity (that is, how much awareness we have of a particular product). If my only criteria for laundry detergent is that it comes in powder and is unscented, then I don't care whether it comes in an orange box, which might be bad for Procter and Gamble, the makers of Tide.
In the microfiction, Trix works to get around Thoreaux's AI agents by planting images in the real world, but that's only part of a bigger picture.
In the future, how will companies advertise to AI agents if such agents are impervious to nonsensical but persuasive-to-some-humans imagery (e.g., "buxom blonde woman in a black bikini lying on top of the hood of a red convertible suggests that buying the convertible will also get the babe")?
What is real?
Elsewhere in the microfiction, Trix learns that Thoreaux is a regular viewer of a financial program on television called The Monkey Wrench. She pays the producers to have the host of the show, Cyrus Wrench, wear a digitally inserted foxglove lapel pen.
This might seem like the most far-flung, futuristic, science fiction moment in the story, but it's already here. Omar Tawakol, a brilliant serial entrepreneur I've known for years, is most recently the founder of Rembrand, a company that uses AI to insert digital products into programs and/or turn generic products into branded ones. For example, if a sitcom films a group of friends having beers at a bar, the production company might have the actors hold blank beer cans. Rembrand can later transform the blanks into cans of Bud, or Coors, or a microbrewery with an atypically large advertising budget.
For advertisers who have to contend with people blocking or skipping ads, Rembrand presents an exciting new opportunity to integrate products inside stories. This is a dynamic, digital, AI-powered version of Paid Product Placement, which has been around for decades. (If you saw Wayne's World back in 1992, then the scene with gratuitous inclusions of PepsiCo products is a memorable example.)
For viewers of such stories, the stakes are low. We already know that a sitcom isn't real, so who cares what beer they drink? We know that the products on the desk on SportsCenter are promotional.
In the story, Trix creates a real Digitalis Project because she knows that the economic stakes of a partnership between her employer's company and Thoreaux's company are high enough to warrant it. She also knows that, if she were to create a façade, in time Thoreaux would discover the lie, which would backfire.
For The Monkey Wrench, Trix creates a real, albeit new, project with a real foxglove icon that she then digitally inserts: a digital lapel pin on an actual lapel. It's a reality/illusion paradox.
With generative AI, augmented reality, and AI agents evolving at gasp-inducing velocity, in the very near future there will be more and more such reality/illusion paradoxes.
Will we even notice them?
Thanks for reading. See you next Sunday.
* Image Prompt: "Please create a photorealistic image of a mid-50s, glasses-wearing man sipping a steamy cup of coffee in a kitchen filled with sunlight. At his feet sleep two adorable tricolor Pembroke Welsh corgis. Behind them, and through a large picture window, a robot delivers two bags labeled 'dog food' to the house front door." It took several tries before I got the language quite right. Interestingly, the corgis have tails, which isn't uncommon but also isn't their traditional appearance. Also, as with other LLMs, Gemini has trouble with even the simplest spelling: "Dog Pood?"