I build generative media, the industry term for AI-generated text, images, video, and audio, and the verification layer it has to pass before anything ships. At Trending Society, my AI research and creative lab, every draft is checked on Claude against primary sources and held until I approve it. I build it that way because of a problem I worked inside for more than a decade in adtech and fintech. When an attribution record, the record of which party sent the buyer, passes from one company's systems to another, it does not survive the transfer. Every number computed after that point is an estimate.
For me, verification is the source of truth, the part where things can be measured accurately, and the question I keep asking, working closely with AI every day, is about human judgment: what would a person need to see before they trust the outcome? Whatever I learn along the way, I write up and share publicly so anyone working on the same problem can collaborate to invent better solutions.
Where my habits formed
My dad was a man of few words. He worked as a computer specialist at a Hoffmann-La Roche pharmaceutical plant, back when computers ran on punch cards, and he filled our house with computers my whole childhood. He stayed long enough to earn a pension, then left to start his own real estate business, because he wanted to own something he had worked hard for.
He wasn't around much, and it was only when I grew up that I understood he was giving up being there so my family and I could have opportunities. He also gave foreign students the opportunity to further their education in this country by giving them a place to live. He got sick in 2001 and again in 2024, and each time, what transferred to me was agency and the motivation to build something toward a better outcome.
As far back as I can remember, what he passed along was the chance to keep learning. Those were the seeds that taught me to stay curious when no one was around and to be accountable for getting where I needed to go. He would drop my brother and me off at Barnes & Noble, which was like a daycare, and I read there all the time. Every summer was some kind of academic camp, including a tech camp at Montclair State University for robotics, applied mathematics and visual arts.
There was a pattern to those drop-offs. I was always waiting for my parents to pick me up from wherever they dropped me off, and it usually took longer than it was supposed to. It left me more curious, and more in my head all the time. I watched everyone on the schoolyard, the football field and the basketball courts, where other kids' parents showed up to watch, and you feel like those things matter. I went through stages of being a rebellious kid, and maybe some of that was seeking attention for being left behind. I ended up relying on other people's parents as examples, which isn't to say my parents were bad. It was taking a complex problem and making sense out of it. Taking responsibility for myself and my own agency forced me to become more resourceful and creative, to stay curious wondering why.
Growing up that way taught me not to judge people and lead with empathy when solving a problem, because you don't know what they've been through or who they are. You don't know their story until you ask the right questions, including why. I think a lot of people miss that and are quick to pass judgment, especially with AI, and I think about it a lot when fear sells dependency on the tools.
Summer camp at Montclair State
That camp felt like going to college as punishment. I think at the time, I'd rather have been home spending my summer watching cartoons and eating junk food like a "normal" kid. My parents dropped me off, and from 8 a.m. to 4 p.m. I was on my own, reading my schedule, being responsible for myself and figuring out where I had to be for each class.
The first class was math. We started with pencil and paper, then moved on to algebra with calculators. Looking back, I guess the program was conditioning someone's brain in the morning to read patterns, absorb them, and take further action, being creative to solve the next problem.
The visual arts class was where we learned to draw and let our imaginations run free to create things that didn't exist yet. Part of the assignment was making flip-books. We drew motion one frame at a time on paper, the way animators did before computers, then flipped the pages and watched the images come to life, and designing our own characters made it an animation class done entirely on paper.
In those drawing classes, it was just me, a pencil and a blank sheet of paper, alone with my thoughts in a classroom where I felt small. I kept looking around the room at what everyone else was drawing, comparing myself and wondering what I should be drawing. In the end I drew whatever was on my mind: the books my older brother and sister handed down, the cartoons on TV (Garfield and Odie, Calvin and Hobbes, The Simpsons, The Magic School Bus, Arthur, South Park, Dragon Ball Z, and Nickelodeon's Rocko's Modern Life and The Ren & Stimpy Show), even Toucan Sam and Cap'n Crunch off the cereal boxes at home. I started by bringing the books to class and tracing them to learn the patterns. Then I wanted to draw them freehand, because it was a challenge, and my goal became being a better creator.
My older brother and sister kept a Garbage Pail Kids collection in their room, and I would sneak in to go through it. Flipping through them felt like one of those animated flip-books, and I'd arrange them in an order that told a story of its own. Every card felt like entering an alternate universe I was trying to make sense of. Who was this kid before the card? What happened to make them look like this? Why would someone even draw that? I went through them over and over, picturing what each character would be like with its own animated show. I read MAD Magazine for the same reason, especially Spy vs. Spy, where the two spies kept outsmarting each other without saying a word, and whenever we were at Barnes & Noble or the library, I went looking for the newest issue.
Back at camp, I also took a theater arts class, which I don't remember choosing, and it was an uncomfortable environment for me. We did improv, practiced lines, read out loud and worked with the other kids in the class. At the end we performed in front of a crowd, Abbott and Costello's "Who's on First?", and I remember how nervous I was. Practicing the same lines every day in class and at home felt like another assignment. The script on paper never told you how the audience was supposed to feel. In front of the crowd, I found it interesting how the way words are delivered can draw different reactions from people. The part where people laughed, when I was being more animated, felt good, and the performance was like showing a product to the customer, getting the reaction, and getting that feedback loop.
One of the girls I was paired with had a mom everyone recognized, the mom from The Cosby Show. She came by often to drop off cookies, and she was the sweetest, nicest person. Other parents checked in now and then, but it felt weird that mine were never around, and she would ask how I was doing and ask me questions, as if one of my own parents were there. That stuck with me. You see someone on TV so often that they become an image, and then they're standing in front of you dropping off cookies, and it's surreal. I'm not sure other people feel that way, but it showed me how much the content we absorb shapes the way we see people.
When I think about that camp now, it was like being put through a self-learning system, a template for continued education, refinement and evolution.
I guess the irony is that the kid who wished he was home watching cartoons is now choosing to make them, creating generated media with First Day, my AI animated short, and enterprise companies that need AI character consistency for better brand storytelling.
Those classes and experiences are what led me to get a Wacom tablet again. The last few months of building with AI on Runway's platform and Claude brought back that feeling of it being just you and your thoughts, and it made me want a pen in my hand again, reimagining things, building something on purpose from the start, and inventing new creative automation workflows.
Working this close to AI, drawing lets me keep my motor skills, the connection between critical thinking and application, and the intentional human touch in what I create, and it lets me follow a piece through to the end. I don't want us to forget that as things become more autonomous. The work will get easier as the challenges change, and what worries me is losing the struggle, because going through it is what builds your learning experiences and your character, which I've seen in different social settings can make different types of impact.
Every so often I rewatch Steve Jobs' 2005 Stanford commencement speech to hear his message again. He talks about a calligraphy class he took after dropping out of college. The class had no practical use at the time, and ten years later it shaped the typography of the first Macintosh. As he put it, "You can't connect the dots looking forward; you can only connect them looking backwards." With applied AI learning, that's how those drawing classes feel to me now. They seemed small at the time, but I'm connecting the dots to where they led me, from where my career started to where I am today. Now the industry puts so much emphasis on that human touch and AI, and I never imagined it would come back around for me on a platform like Runway, or that I'd be in their Builders Program. When you're creating something from the ground up, starting from the fundamentals matters even more.
The seventh-grade robot
In seventh grade, the project that stuck with me in the robotics class at camp was a robot car that steers itself around walls, built from a kit of wheels, a motor, a circuit board, a chip, resistors and a sensor. The assignment was to solder the parts and connect the wires, and along the way I learned about positive and negative terminals and grounding the battery. I raced down the instruction sheet, soldering each part into place.
The robot ran fine at first, but once it met an obstacle, or anything outside the perfect environment it was built for, it would hit the wall and malfunction. When I took it apart and checked each solder joint, I found a solder bridge, a blob of solder connecting one pad on the printed circuit board to the pad next to it. That short circuit kept the sensor's signal from ever turning the car away. I had rushed the build, and realizing it left a lasting impression on me. That tiny, hard-to-see fault shaped how I look at AI: you build a system, take it apart, examine the layers you can't see, and put it back together. Slowing down is what lets people find and understand the edge cases, and when I look at how we're approaching AI as an industry, it feels like we're rushing it through without thinking clearly, which worries me. Maybe I see it more because I sit so close to this every day, seeing the edge cases and putting in the safeguards.
My first job
I watched my parents clip coupons whenever we went grocery shopping, and I saw their cards get declined, so they'd pull out another one to make it work. Birthdays and holidays were less about gifts and more about time together with extended family. Seeing all that made me stop asking for things and want to get a job, but I was fifteen, so every job I applied for needed a signed permission slip.
I went to see the principal of Wayne Valley High School, Reese, like Reese's Pieces, who had a white beard and a pipe in his mouth. He reminded me a bit of a Santa Claus hippie. I told him the whole situation, and I could see the hesitation on his face before he signed it. Before he handed me the paper, pipe still in his mouth, he told me something I still think about. "Promise me, Jeff, never stop learning, and think long term about what success means to you." I think he was worried I'd end up in a retail job, never go to college, and that would be it.
So my first job working directly with customers was in high school at the Staples on Route 23 in Wayne, New Jersey, a job I took because I wanted to build my own computer. I was a shy kid, and it put me somewhere uncomfortable where everything was new, so I had to show up on purpose and work out for myself what I wanted to get done there. Talking to strangers all day felt weird at first. The cash register helped, checking out one customer after another like an assembly line. I was nervous running the machine while doing the math, counting change, and processing credit cards, cash, checks and every other way people paid, making sure every total came out right. Those moments kept me sharp and on my feet, a kind of mental conditioning for human motor skills.
Off the register, I worked as a merchandiser in office supplies. The work was repetitive: stocking shelves and matching each product to the price on its label. A lot of variants looked almost identical, but carried different prices, and the last thing you want is a customer upset at the register because the shelf was wrong. A wrong shelf price sits close to the line of false advertising in the small print, and keeping the shelves accurate is the store's responsibility. Reading every label became a habit, a kind of quality assurance. The repetition turned into a discipline of remembering, so whenever new stock came in, I knew what it was and where it went. That work pulled me into other parts of the store, selling tech insurance on graphics cards, modems and routers, and assembling office equipment. At home I was left alone with my thoughts, and work was where I got better at the things I wanted to improve, by being social and productive.
Then one rainy day there were no customers, and a pile of leftover back-to-school binders was sitting up on the top shelves, collecting dust. So I pulled it all down and color-coded it, arranging the colors the way they sit in a Crayola box, and lined them up so the aisle looked more complete. My boss walked in, saw the boxes everywhere in the aisle that looked like a war zone, and freaked out. "Jeff! What the heck are you doing?!"
He thought I'd lost my mind, until I told him why. I figured it was better for people coming in to see all the colors we had so they had more choices that would lead to sales. Sometimes customers don't know what they want until you show it to them, plus the shelves looked more polished. Whenever someone bought one, it left a gap, and stocking them end to end meant all you had to do was tighten things up instead of getting the ladder to bring the boxes down. It lasted longer on the shelves and left more time for other work in the store.
The store earned strong customer-service rankings, but it didn't have an employee of the month. My manager started one and gave it to me, based on mystery-shopper evaluations, tech insurance sales, and the binder incident. It felt like he wanted to start some competition and get everyone more productive. My reaction was, "Whoa, what's going on?"
I wasn't trying to prove anything. I was doing my job and passing the time, being useful instead of sitting around, because I can't sit still for too long and I always need to be productive or learning something.
It felt good, and it also felt strange, because I'm not the type of person who enjoys making other people look bad, especially since my coworkers were much older than me and I was just trying to fit in. Reflecting back on it as a manager, it seems like a smart idea, but I can see how it could make other people feel, given their tenure. The lesson I took was to do the quiet work and give it the time it takes, with the mission of giving customers the best experience.
Architecture and NJIT
In high school I was pursuing architecture, and looking back, that came from wanting to build homes and communities with my dad. Money was a concern in our household, so I was always evaluating trade-offs and analyzing the edge cases to gain more control over the outcome. I was a big dreamer, living in the clouds, with a vision of building skyscrapers, and I think it was me wanting to escape reality. To get there, I had to start small and look at every tiny detail, working backwards from the big vision to the first step. I was living in both places at once, the big vision and the small steps it takes to execute it.
Then, a few months before I graduated high school, a student sitting across from me in CAD class pulled my USB drive out of the computer, and I lost the only copy of the portfolio I needed to submit to NJIT. It taught me to keep a backup, never store all my work in a single location that can fail, and have a plan B. Instead of architecture, I went to NJIT, which accepted me into its new Information Technology program, where I focused on multimedia. That mix of creative work and technology is what I would now call a creative technologist and forward deployed engineer mindset.
Where my career started
Those habits carried into a federal work-study job at NJIT in 2008. What that job gave me was being able to take my agency into a structured environment, with a culture and people around me that allowed me to learn more, build solutions, and solve problems for people.
The work-study job was in Instructional Media Services, a small room with three desks. The professor who ran it, Keith, was also my manager, and he was the second person after that principal to quietly hand me room to run. The job was putting professors' course material online in Moodle, but Keith gave me the real assignment, redesigning the department's website in Drupal.
Part of the website work was walking around photographing the classrooms, and that's where I started practicing more with my camera. I was already an Apple fanboy, and my love for design led me to Adobe products and to buying my first iPhone on Craigslist. I kept working retail jobs, at Bed Bath & Beyond and Guess, to afford my first MacBook Pro. I started freelancing to shoot more, for Patch.com, New Jersey Underground, the Montclair Art Museum, and anything else that let me get better with my Canon 7D. One of those gigs was a concert at the Starland Ballroom in Sayreville, NJ, with Hawthorne Heights, Senses Fail and Matchbook Romance as the headliners. I edited my photo from that show in Lightroom with grunge filters, a setting I'd never seen done before, and I just tried it out, adjusting the knobs until it felt right. I wasn't sure it was the right choice, because I was comparing myself to others and had never seen anyone submit something like it, but I submitted it, and it won a competition.
I think it was the curiosity of always thinking there could be a better way, and it led to different transformations of exploratory work that, in hindsight, have been paying dividends in places where there was space to run. My photography led to the question of how I was going to show it to people. The internet is a place where you can reach millions of people without a big marketing budget, if you're resourceful about where you share things, so I got my website up with my photography and went after gigs on Craigslist. Part of that came from MySpace, which I liked. It was a gateway to customizing your own space, and that led to buying my own domain and putting my photography there, where I had ownership. My website has lived in different eras since then, and part of that was continuous refinement, exploring, jumping around, and not being okay with just one solution.
I spent more time learning on my own, focused on the things I wanted to learn, and before AI, Wikipedia was where you went to learn anything and everything. It attributes dates, timelines, and history, and history has a tendency to repeat and shape trends, which is what you see when you work on the financial datasets and technology.
Around 2008, there were signs of things happening at home. Between the housing crash and being defrauded, my dad lost his business. His real estate business rose and fell with a market cycle he couldn't control, and I grew up around people getting taken advantage of, which affected our family. My siblings had to step in with the family's finances, and I don't know if we would have survived those times if they hadn't taken on that role. From what I've seen, that's a cultural thing, the kids taking care of the parents once the parents get to a certain point in their lives. I understood I needed to be responsible for my own outcome and invest in my own self-improvement, and for a long time I wanted to understand how everything connected.
On paper, it looked like we were fine, but we weren't. Somewhere in those four years, I couldn't get the rest of the federal financial aid I needed for school. That forced me to take out private loans through Sallie Mae and personal loans from my siblings. The objective was to stay on campus my last year because the environment there was better than what was going on back home. Watching documentaries and reading stuff about being a product of your environment made sense to me at the place where I was in my life then.
The same pressure is on families again. At the end of June 2026, Americans owed $1.26 trillion on credit cards, and after the pandemic pause on federal student loans ended, about 1 million borrowers defaulted in the last quarter of 2025 and 2.6 million more in the first quarter of 2026, with the average defaulted borrower's credit score falling 91 points.
Back in school, I was taking photography classes to create different scenes, and I had to engineer events for myself to build the skills I wanted. I worked part-time through college to pay for things and have a social life, so I spread out my classes and took summer classes.
One was a humanities class whose final was a paper on Stanley Kubrick's 2001: A Space Odyssey. It was one of our last days before we graduated. I thought it was bizarre that the professor was making us watch a movie in class for our final exam, and he even looked a little like Kubrick, but I understood he wanted us to walk away with something. I'd always heard about the movie and never seen it. The room was hot, the lights were off, and I sat there sweating through what felt like one man alone in space, trying to make sense of each part without much context going in. At the end, he wanted us to write what we thought it meant. I watched it mostly confused, and after dissecting it, something resonated.
Kubrick never settled it in the film. In a 1968 Playboy interview he said viewers were "free to speculate," and in footage shot in 1980 for a never-released Japanese documentary, he described the room at the end as "a human zoo" where the astronaut is studied, and "his whole life passes from that point on in that room." When Christopher Nolan introduced the film in 2019, he called it "primarily an emotional experience. It's something you feel." He said that when he first showed it to his kids, "the ideas of artificial intelligence seemed a bit quaint," until talking assistants like Alexa and Siri brought them back.
My take was that when you're alone, you're left with your choices and decisions. There was an AI element to it too, some alien form of machine, and a person's exploration toward evolution. I've thought about that movie from time to time since, working in the tech industry and watching this environment evolve, and I plan on reading Clarke's book, especially with where AI is heading. I feel there are new kinds of loneliness in this digital era, and as we move toward autonomy, you can't forget the human parts of it, because these machines have a way of tricking us. A four-week trial by MIT Media Lab and OpenAI with 981 people found that heavier daily chatbot use went with more loneliness and less time with real people.
Looking back, rare moments that seemed random, like that hot classroom, are where I met people, and the relationships that came out of them are how I build toward an outcome. Success, to me, is being able to share those growth moments with people, so that on my deathbed I can feel the love in that room.
The city that never sleeps
Being resourceful and applying what I learned to the next thing is how I later got the opportunity at Zeta Global, a company in the early stage of its growth, co-founded by John Sculley, the former CEO of Apple and of Pepsi-Cola. That was inspiring, and creative problem solving was what I was pursuing there.
Every day I took the bus from Wayne, NJ to Port Authority, reading Walter Isaacson's Steve Jobs to learn the history of technology and the role Jobs played in it. Then I walked down to 34th Street and Madison Avenue, past the Empire State Building, to what felt like walking to my dream job. Getting there seemed like luck, and it all felt surreal. Since then, Zeta has built AI into its marketing platform. In September 2025 it added a generative engine optimization (GEO) tool that tracks how brands appear in answers from ChatGPT, Gemini and Claude. In January 2026 it announced a collaboration with OpenAI to power Athena, its AI agent for marketers, and in June 2026 Palantir and Zeta announced a partnership to rebuild Zeta's Data Cloud on Palantir's Foundry, with Athena as its intelligence layer.
From there my career ran through adtech and fintech, on both the creative side and the engineering side, and at each place I worked, I was connecting the dots on the cycle that caught my dad. At Undertone, now Perion, I built the desktop unit on Champions of Cash and Jonathan McGough built mobile, and reviews ran on working builds. Garmin vívoactive was built the same way, on creativity, collaboration, culture, and ownership. Staples, Champions of Cash, Garmin, and later the Trending Society pipeline are the same method at different sizes: learn the whole system before changing one part of it, and put written checks in front of anything that ships.
The moments when I was working with people, collaborating and building something, on a mission for everyone's success on the team, were much more rewarding and fulfilling than working on my own.
Where the record breaks
Most display ads online are bought by software in an auction that runs while the page loads. EMARKETER forecast that about 92 percent of US digital display ad spending in 2025 would go through it, and bidders answer within a deadline that is often about 100 milliseconds. Every one of those auctions depends on knowing which browser is on the page, and that identity layer has never been reliable. For years the identifier was the third-party cookie, which Safari and Firefox now block by default. The main replacement is browser fingerprinting, matching a person to the combination of attributes their browser exposes, which the UK's privacy regulator called irresponsible when Google announced it would allow it for advertisers.
Attribution then falls back on the attribution window, the number of days after a click during which a sale is credited to the publisher that sent it. The publisher and the platform agree on its length, and that number decides who gets paid; if you have ever watched two ad platforms report different totals for the same campaign, this is usually why. The industry's answer was a second industry, ad verification, which audits the first one after the auction has settled and the money has moved, so a bad transaction turns into a refund instead of being stopped.
Affiliate marketing is the simplest case. A creator posts a link, someone clicks it and buys something, and three things track the sale: the affiliate ID in the link, a cookie in the buyer's browser, and a dashboard someone maintains. All three live inside one company's systems. When the sale moves to a second company, such as a payment processor or an ad platform, the second company starts its own record with no way to check it against the first.
Nothing about this is specific to advertising. It happens wherever two systems hand off a record without a shared source of truth, and if you build anything that passes a record to a partner, your system has it too. The record can be a referral, a payment, or a claim about who somebody is. Through my journey, I kept seeing the same gap: someone does the work, the promised payment or credit doesn't come, and there is no record to point to. So I started building the thing I wished existed, a verification layer, so that a person's work carries proof of where it came from.
Building Trending Society
In March 2024 I left Interactive Brokers and Chicago, a city I had lived in for two years, because being close to home and family mattered more, and I wanted more room for creative problem solving and the newest technology. Banking also looked like the industry first in line to be disrupted, and staying didn't seem like a good decision for my career growth. Accenture's consulting research ranks banking first for generative AI's impact, with almost three-quarters of its work suited to automation or augmentation, and many banks still run core systems up to 40 years old on mainframes.
When I left an established career to reinvent myself, I thought about Jobs leaving Apple and what he called "the lightness of being a beginner again, less sure about everything." In April 2024 I started Trending Society, which has become my AI research and creative lab. Its system is built from answer engine optimization (AEO) pipelines and agent tasks, where AEO is the work of rewriting material so that machines answering questions will quote it correctly, and the goal is to de-sensationalize online content and lay down the facts without an agenda or a programmatic pattern that engineers content toward a specific audience.
More or less, the idea was a Wikipedia of AI and open source, with the attribution that comes with it and the skills around it, because I had a vision that it would become important for people wanting to upskill. Trending Society is my way of leaving something behind, the things I used to get ahead, for someone else to pick up and use. I wish I could spend more time on it, but part of my focus goes to my own work on deep learning with AI and on building closed-loop systems. I was on and off about the product, but I wanted it trending in that direction, toward a better society.
It was really about pushing the boundaries, so I asked myself what a better way of learning would be, one where I could see where everything fails and where everything works, and invent new ways of working that automate a lot of what I do today with agents. Building AI-natively, you can build anything if you apply your mind to it and are persistent enough to carry it through to the end, and what you learn compounds into the system you've built with your agents. This industry has a habit of catching up quickly, more now than ever, because the models are trained on so much of that output.
What excites me is solving complex problems, and with AI you can be more ambitious about them if you're curious and resilient enough. I have high conviction because of the feedback loop from when I first started this journey: seeing what evolved on the macro told me I was on to something. When I was unsure, or questioned whether to keep going, another signal would show up and tell me to keep building. Each one got me to a new milestone, further along with AI and the agents, and a lot of it was grounded in research, steering the models with what I knew.
The signals
The first of those signals came from AMD, whose chips ran the first computer I built. In January 2025 I went through a batch of AMD's newest patents, from GPU chiplets to processing-in-memory and neural network kernels, and posted my research on Instagram when I first found them. AMD closed that month near $116, and it closed at $614.61 on September 23, 2026, more than five times that. Oracle committed to deploying 50,000 of AMD's MI450 GPUs with deployments set to begin in the third quarter of 2026.
On April 28, 2025, I mapped which jobs AI is least likely to replace, based on its limits in creativity, emotional intelligence, complex motor skills, and contextual judgment. I hadn't looked at that deck in over a year. When I opened it again while writing this, I realized it described everything I had built since, Trending Society, First Day, everything in my repo. I had just been following my intuition every day, learning something else, and that deck was the starting point.
On June 12, 2025, I started researching what would change for students if federal education funding went away, and that research became the thesis behind the AI tutors I build at Trending Society with the Runway platform.
In August 2025, through Trending Society, I funded and ran a four-role pro bono internship program, posted on LinkedIn, so students could learn by building alongside me, the same seeds my dad planted, applied. It grew into more than one person could run alongside everything else, and it taught me that people need to learn agency on their own, the way the work-study job taught it to me, so I wound it down and put that time back into building.
The verification layer
Trending Society started as WordPress covering a lot of different subjects, with Claude working on the server over SSH, before MCP servers. I architected it so that, based on certain tags, it classified and mapped my prompt writers, my copywriters, and the persona behind each one. Trigger.dev ingested the whole article from a URL, and the pipeline took that content as a base and re-engineered it into the output I wanted, which could then scale out to every social platform and every kind of generative media you can make from a text output. Prompt engineering, context engineering, and harness engineering are all derivatives of that same thing, natural language models working from text.
I reverse-engineered WordPress, built my own MCP server, and ran the work like an engineering team on my own, with agents across different IDEs, learning to operate AI-native as I went. As the codebase grew, leaning on the tools for everything made the whole system harder to hold in my head. As the developer kache (@yacineMTB) put it, in a line Andrej Karpathy has been repeating, "you can outsource your thinking, but you cannot outsource your understanding." When you're building software, you need to understand what the logic looks like before you can code it, and with AI that understanding matters even more, because AI multiplies whatever logic you give it. So I started writing down the verification checks and audit trails, and building skills, knowledge docs, and memory systems.
The research, the knowledge base, and the insights are mine, and I use them to steer the agents. What I built around the agents runs on Claude, a verification layer the work has to pass before anything ships:
- every factual claim is checked against a primary source
- every link is opened to confirm it proves the sentence it sits under
- every date and number is read from a source at the moment the check runs, not from what a model remembers
- the writing is checked for AI tells and for rhythm
- a reviewer with no context reads the finished piece the way a stranger would
- nothing publishes until I approve it
Starting in April 2026, I rebuilt its article pipeline as a multi-step job on Trigger.dev that every draft runs through. It checks each draft against 10 structural AEO rules and 7 judge criteria and rewrites any draft that fails. Its prompts did not transfer cleanly from the agent system where I refine everything, so I am fixing them now and do not publish an output rate for it yet.
The hardest part is how fast the industry moves: anything technical I publish starts going out of date almost immediately. A model that answers from its training data, or from a page that has gone stale, has no baseline for what is current, and I think that is a large part of why AI output gets facts wrong. The next piece I'm building is a fact checker that re-checks published technical content when its sources change and records each update, so every article carries a paper trail of when it was created and when it changed. Even while writing this, I checked Google's guidance on publication dates, which asks for a visible date labeled "Publish" or "Last updated" alongside datePublished and dateModified in the page's structured data. Trending Society's articles already carry both dates in their structured data, and I'm adding the same to this site as I refactor it.
The pipeline lives in a private Trending Society repository, so here are six of its article tasks, its repurpose and social distribution tasks, and the two check files, by name only:
packages/trigger/src/
├── tasks/article/
│ ├── single-article-pipeline.ts orchestrates one article, end to end
│ ├── classify.ts classifies and filters the source
│ ├── verify-claims.ts checks the draft's factual claims
│ ├── article-rewrite.ts rewrites a draft that fails a check
│ ├── generate-narration-script.ts
│ └── sync-article-to-notion.ts hands the draft to the Notion CMS
├── tasks/repurpose/
│ └── orchestrator.ts turns one source into captions, threads, scripts and FAQ schema
├── tasks/social/
│ └── distribute-to-social.ts fans a post out to every connected platform (paused for redesign)
└── lib/content/
├── aeo-validation.ts the 10 structural rules
├── llm-judge.ts the 7 judge criteria
└── 11 test files alongside the checksThe same pipeline extends to social media. The design starts with a single content source, the way I began with AEO, and distributes it across formats, agentic media and AI-generated media on every social platform. As the space evolves, I don't want to be locked into one solution, so I build on these patterns as the tools change.
What I've observed is that AI assistants are becoming a direct channel between brands and their customers. OpenAI reports that ChatGPT has more than 900 million weekly users, and about half of US adults now use AI chatbots. Gartner predicts that by 2029 agentic AI will resolve 80 percent of common customer service issues without a human, though customers aren't there yet, and 87 percent say a company using generative AI must still offer a way to reach a person.
Where the record breaks now
When the fake thing is the person
Ad verification scores a request by how likely it is that a real person sent it. That model holds when the faked thing is traffic. It breaks when the traffic is real and the faked thing is the person in the content.
In 2023 AI video was a grainy clip of Will Smith eating spaghetti that everyone passed around, and on June 10, 2026, Dreams of Violets, a fully AI-generated feature directed by Ash Koosha, produced with Pooya Koosha and Fountain 0, premiered at the Tribeca Festival. It runs 75 minutes and cost about two thousand dollars to make, according to Fountain 0. Its subject is a January 2026 massacre of Iranian civilians, and it was made in London by a director who said he had no access to Iran. Every person on screen is generated.
The same tools clone a voice. The FBI now breaks AI-enabled fraud out separately in its 2025 Internet Crime Report: 22,364 complaints referencing AI and nearly 893 million dollars in losses, from voice clones, AI-generated identification documents, and believable videos of public figures or loved ones. The FTC's 2025 numbers put imposter scams, with or without AI, at 3.5 billion dollars across nearly one in three fraud reports.
When nobody can prove where an answer came from
AI Overviews are the answers Google writes at the top of a results page, using content from the sites below it. Penske Media sued Google over them, and a federal judge dismissed the case on September 30, 2026, but the amended complaint lays out the old arrangement plainly: publishers let Google crawl their content, and Google sent readers back. It cites a 2025 Ahrefs study that found a 34.5 percent lower click-through rate on the top organic result when an AI Overview appeared. In late May 2026 a court in Munich ruled Google couldn't claim host-provider protection for inaccurate AI Overviews, and Google said in June 2026 it would appeal.
Publishers are now paying for AEO, whether or not a reader ever clicks through, and every one of those projects runs into attribution, which stays open because no party in the chain can prove what a given answer was assembled from.
Licensed voices, and where the license stops
Licensed voice cloning already exists, with named talent behind it. In November 2025 ElevenLabs launched an Iconic Voice Marketplace, and Michael Caine, then 92, licensed his voice into it, and in May 2026, ElevenLabs licensed Stan Lee's voice and likeness from the venture that manages them. Tennessee's ELVIS Act, signed in 2024, added a person's voice to the likeness rights the state protects, and New York now requires the rights holder's consent for digital replicas of the dead. I have my own voice clone on ElevenLabs too.
A company submits a request for a voice, ElevenLabs routes it to whoever holds the rights, and the two parties sign a licensing agreement, but the public process describes no record that ties a specific generated file to a specific license, and no payment triggered when that file plays. The marketplace solves discovery of the rights holder and lets talent get paid through the deal, and attribution is still open.
A draft Ethereum standard, ERC-8004, proposes three on-chain registries that no single company owns, which is the shared record adtech never had: an identity registry for agents, a reputation registry for feedback on how an agent has behaved, and a validation registry for independent checks of its work. Its authors work at MetaMask, the Ethereum Foundation, Google and Coinbase, and the text is still a draft. Until a shared record like that exists, my work runs through the verification layer I described at the start, so every claim in it carries proof of where it came from.
Runway and First Day
On June 4, 2026, Runway accepted me into its Builders program, one of 65 builders in the cohort, and First Day is built there.
I've been working in Runway, cloning myself. What I'm building is templates. It's how I've built everything across my life and the places I've worked, where the workflows and programmatic work were templates too. The skincare ad on my homepage is one of them, a product ad with a generated influencer, made once and localized into English, Japanese, and Korean, and each card labels her as an AI influencer.
Now I'm building field vision on smart glasses for service-based businesses, the industries that need the connection and the real-world use cases of this technology. Their work requires human problem-solving that AI is not good at right now without robotics, and the glasses support more efficiency and less cognitive load by making the AI more voice-conversational. It takes a kind of spatial awareness I think we need more of. With field vision, OCR (software that reads the text the camera sees), and real-time feedback on the glasses, the question is what that feedback delivers: a workflow that saves time, or a template for a business use case, without being attached to your phone all the time. What if you stayed connected, but it felt more natural, the way wearing a pair of glasses feels natural, and you used a wake word only when you wanted it, instead of reinforcing a bad habit?
Being on Runway's platform is sometimes like rewatching 2001: A Space Odyssey. It has that open sense that anything is possible, and sitting there, you're left with your thoughts about what you're going to build. It gave me a new perspective on how to view this environment. Keep an open mind about how you problem-solve, because yesterday's solutions are what I'm refactoring now, and what you build today can be tomorrow's tech debt. I try never to feel like I've made it. Material things never made me feel fulfilled, and the feeling from getting them didn't last. What I enjoy is building, being creative and productive, and spending my time on things that matter, and when the stakes are high and the mission behind the work has good intent, that is more rewarding. Jobs said it in his Stanford speech: "Your time is limited, so don't waste it living someone else's life."
Why First Day
Without provenance, a viewer watching a generated ad has no way to know it is generated. The people who will spend the most years inside that arrangement are young kids and teenagers now, which is why First Day, the 1:40 AI animated short I made through the Runway Builders program, is an architecture decision for me. I treat it the way I keep architecture decision records, with the reasons and principles written down: a better story for the younger generation, and a better place for them to consume content.
First Day is symbolic for me. Bao and his friends each have a distinct story and background, with family and social pressures of their own, and when they are together at school they become friends, work as a team, and learn from each other. Through their problems, they help one another and overcome obstacles as a group.
The symbolism comes from growing up with Pixar and Disney movies. After watching one, I always took away a lesson that stayed in the back of my mind. Now, as a creative with access to AI tools, I want to carry that message to a generation that needs hope, in a form they can actually reach, through the skills they'll learn with AI tutors.
Starting with First Day meant starting with the hardest problem. Kids are an audience where verification and safeguards are a core requirement, so I won't deliver something to them without the right architecture, one whose output is checked against written rules before release. Building for that audience puts the safeguards into the architecture from the start, instead of adding them to UGC ads later. What the right disclosure looks like, and who should define it, is still an open question across the industry, and building the creative and the engineering under it keeps me close to that question until there's an answer.
The weight of it
Now I'm using every tool my dad gave me, together with AI, to build toward a safer world for people. AI feels like the modern version of being dropped off at that Barnes & Noble, except now I can ask the book a question, and it answers, explains, helps me build, and cites its sources. I have also seen how systems like it get designed for engagement instead of understanding. As the environment changes, people seem checked out, and everything feels so polished with AI, which is why I think the authenticity of storytelling matters more now than ever. People want to feel something, and they want something they can relate to.
Some mornings I wake up wondering whether my thoughts are random, or whether they come from the research I've done and the data I use to make decisions with AI. I reverse-engineered my social media algorithm around what I'm interested in, AI, financial education, economics and politics, because being informed about real-world events matters. Every time I sense I'm leaning one way or another, I try to reset it and see both sides, and since each platform has its own algorithm, I mine signals from them as I build and keep a library of what I find.
I talk to Claude every day with my voice, and it talks back, so we're in a loop of continued refinement toward the outcome I'm trying to engineer, and I had to figure out how to make it better so that it makes me better too. What's left to the person behind it is the judgment and the intention, and what's best for society won't be clear until we see the failures out loud.
When you acquire experience, knowledge and skills in your life, a moral responsibility comes with them. The deeper I got into my career, the more the weight of that responsibility settled on me, and with AI I feel it even more. I'm excited about what AI makes possible, and I want the safeguards to keep up with it for the next generation. I'm partly an enabler of what's coming by building the machine end to end, pushing it as hard as I can and understanding everything around it, because you can't prevent a bad outcome with a tool you haven't built with. So I would rather hold myself to a high standard than use these tools just to get ahead. Staying on the why, written down, keeps me grounded and committed.
I see a cycle like the one that caught my dad, where people are getting left behind, and I think learning and continuing education is a survival skill. After what I've been through, I'm not capable of sitting back and watching it happen. For me, the fulfillment is building a better future and a better community.
Prosperity, to me, means shared learning instead of gatekeeping. Opportunity sits at the bottom of an organization as much as at the top, and AI can open that opportunity or close it, depending on how it's built.
A lot of the people who planted seeds in my life were educators, people who let me learn something useful in both my professional and personal life, the good, the bad, and the ugly. What I learned is not to be so attached to an idea or a decision. Being flexible and open-minded has been better for my own growth, and I think for society too, in this environment of divide.
People need to trust AI more, and the challenge is in how it gets delivered. Some people are enthusiastic, others are turned off, and a growing culture around analog devices adds another layer. I'm trying to keep the positives without the negatives, and I do that by testing the edge cases and evaluating the data. What I want to bring to a team is that verification layer, somewhere trust and verification are the product and the feedback loop with AI gets better with other people's perspectives on how they use it. I look for teams that give me ownership, startups especially, where the room to build is a lot bigger, and larger companies that want people with this mindset.
I can't stop where AI is heading, but I can help put safeguards in place with harnesses for society. Building with AI, I operate with integrity for people like my dad, of any background or age, so that whatever is inevitable, people get a fair share and a voice in the outcome. I do it because of the people I looked up to who taught me, and everything my dad gave up so I could be here.
See the work
- First Day, where the architecture decision above gets built
- Champions of Cash and Garmin vívoactive, the adtech builds where the method started, with the collaboration and teamwork behind it