Welcome to Verses Over Variables, a newsletter exploring the world of artificial intelligence (AI) and its influence on our society, culture, and our perception of reality.

Back to Basics

License to Code: How AI Agents Are Becoming the 007s of the Digital Age

Your coffee maker just ordered your favorite roast without asking. Your car rerouted your commute to avoid an accident you hadn't heard about. Your phone declined a lunch invitation because it conflicted with your "secret" side-hustle. No, you're not living in a sci-fi novel—you've just stepped into the "agentic future," where AI doesn't just assist, it takes charge. The tech world's latest obsession? "Agents." Not the suit-wearing, earpiece-donning kind, but the silicon-based, decision-making variety. These digital entities are poised to revolutionize everything from your morning routine to global economics. But what are these agents, and why are we betting on them?

What is an Agent, Anyway? Picture this: You're at a cocktail party, and someone asks you to define an "agent." Your mind might jump to James Bond or your friendly neighborhood real estate guru. But in AI, an agent is something altogether different—and potentially much more transformative. An AI agent is essentially a digital entity with a mission. It's software that can perceive its environment (think: vacuum your floor, scroll your social media feed), make decisions (should I turn left to avoid the cat or right to dodge the pothole?), and take actions to achieve its goals (must. clean. everything.). It's like Siri or Alexa on steroids, with a dash of free will thrown in for good measure.

Not Your Average Chatbot: How Agents Leave LLMs in the Dust: Now, before you roll your eyes and mutter, "Great, another AI thing," let's clear something up. These agents aren't just souped-up versions of the Large Language Models (LLMs) you've been chatting with lately. Sorry, ChatGPT, but you're yesterday's news. LLMs are like that friend who's great at conversation but terrible at actually getting things done. They can talk a big game about planning your vacation, but won't book your flights or reserve your hotel. They're all talk, no action. Agents, on the other hand, are the overachieving cousins of LLMs. They don't just stop at spouting information or generating text. These digital go-getters can:

1. Take real-world actions: While an LLM might suggest you turn off the lights to save energy, an agent could do it for you.

2. Learn and adapt over time: Unlike LLMs, which are stuck with the knowledge they were trained on, agents can learn from their interactions and improve their performance.

3. Make autonomous decisions: Agents don't need you to hold their hand through every step. They can assess situations and decide on the best course of action.

4. Interact with other systems: These aren't isolated entities. Agents can communicate with other software, APIs, and even other agents to get things done.

5. Pursue long-term goals: While LLMs live in the moment, responding to immediate prompts, agents can work towards objectives over extended periods.

In essence, if LLMs are the brains of AI, agents are the brains with a body and a mission. They're not just here to chat—they're here to change your world, whether you like it or not.

Welcome to Your New Reality: Agents as Co-Pilots of Your Life: So, what does this brave new world of AI agents mean for us—our day-to-day is about to get a significant rewrite. First off, forget everything you know about work-life balance – that concept is so 2023. Imagine an AI agent that doesn't just manage your calendar, but actively negotiates your work hours based on your productivity patterns, stress levels, and that secret side-hustle it's helping you run. Your career path? It's no longer a ladder, it's a game of 4D chess, and AI agents are the grandmasters. They'll be analyzing market trends, your skill set, and probably your zodiac sign to plot out your next career move before you've even considered updating your LinkedIn. But it's not all about work. Your love life is in for a shakeup too. Forget dating apps – your AI agent will be the ultimate wingman, analyzing compatibility based on everything from your Netflix history to your grocery lists. It might even start flirting on your behalf. (Let's hope it has a better grasp of pickup lines than most humans.) And parenthood? Oh boy. Helicopter parents will look positively hands-off compared to the AI agents of the future. These digital Mary Poppins will monitor your kids' development, optimize their education, and probably argue with other parents' AI agents over playdate schedules. Even your mid-life crisis gets an upgrade. Instead of impulse-buying a sports car, your AI agent might surprise you with a perfectly-timed career switch, a round-the-world trip it's been secretly planning for years, or maybe even a Mars colony timeshare (hey, a bot can dream, right?).

Reality Check - Are We There Yet? Before you start drafting your resignation letter to hand over to your new AI overlord, let's pump the brakes. The agentic future we've been painting is more "coming attraction" than "now showing." Today's AI agents are less "I, Robot" and more "I, Bumbling Intern." They're like that eager new hire with potential but needs constant supervision. Current agents still require a human in the loop – and by "in the loop," we mean hovering over their digital shoulders, ready to jump in when they inevitably misread the metaphorical map.

The Bottom Line: So the next time you hear someone talking about agents, don't think tuxedos and martinis. Think algorithms and machine learning. Because in the future, the most important agent in your life might just be the one living in your smartphone. Welcome to tomorrow. Hope you like robots.

A Big Week for Launches

The AI world - where "last week" feels like ancient history and "this week" is already outdated. Last week's image and video generation bonanza was like a digital Big Bang, birthing a universe of pixel-perfect possibilities. From hyper-realistic deepfakes to dreamscape generators, the lines between human and machine creativity blurred faster than a Dall-E render. Buckle up, art lovers - we're about to tour AI's latest flexing of its creative muscles. The robots may not have won the art game yet, but they're painting themselves into a commanding lead.

• Flux, by Black Forest Labs - an open source text-to-image model created by the original leaders of Stable Diffusion

• Stable Diffusion Stable Fast 3D - an open source, rapid image to 3D asset generation model

• Midjourney v6.1 - an updated version of its image generation model with better image quality, improved text quality, better personalization and more

• Runway Gen-3 Image to Video - an updated version of its Gen-3 Alpha model which allows users to create video from a single image instead of or combined with a text prompt

• Shutterstock + NVIDIA - commercial users can prototype 3D assets and generate 360° backgrounds using NVIDIA’s Edify, an AI visual foundry

• Getty + NVIDIA - NVIDIA turbocharged Getty’s generative AI image generating service, improving speed, output and fine turning

• Meta Segment Anything Model - an open source, video generation model that allows users to segment specific items, making it extremely useful for computer vision (or replacing a video rotoscope)

• Chrome AI updates including Google Lens - Google Lens now comes to the desktop, allowing users to search for images or pieces of the image directly from the toolbar

• Meta’s AI Creator Studio - allows users to create, share, and discover custom AI characters based on their interests, enabling them to build personalized AIs that can assist with various tasks or serve as extensions of themselves to engage with fans on platforms

• OpenAI Voice Mode - (not video but we couldn’t resist) if you are one of a lucky few, OpenAI has released the natural, real-time voice mode which allows for more human-like conversations, aka HER

We’ll be talking about our favorite tools, but here is a list of the tools we use most for productivity: ChatGPT 4o (custom GPTs), Midjourney (image creation), Perplexity (for research), Descript (for video, transcripts), Claude (for writing), Adobe (for design), Miro (whiteboarding insights), and Zoom (meeting transcripts, insights, and skip ahead in videos).

Intriguing Stories

Acqui-hired: How Big Tech is Rewriting the AI Talent Playbook: A new trend is turning heads in the tech world: Big Tech's penchant for "acqui-hiring" top talent from AI startups rather than purchasing companies outright. This strategic move is rapidly reshaping the AI landscape, with industry giants making bold plays for the brightest minds in the field. Google recently orchestrated a high-profile homecoming, welcoming back the co-founders of Character.AI—a chatbot tool that lets users interact with unique personalities—along with a team of about 30 researchers, through a $25BN licensing agreement. This follows Amazon’s strategic acquisition from Adept AI (an enterprise tool focused on agents). David Luan, Adept's CEO, and several co-founders, joined Amazon's AGI (Artificial General Intelligence) group. While Microsoft, already a major player in AI through its partnership with OpenAI, turned its attention to Inflection AI (an AI studio focused on creating personalized AI). The tech giant hired Mustafa Suleyman, Inflection's co-founder and CEO, to lead a new consumer AI unit. But why are these tech behemoths opting for acqui-hires instead of full-blown acquisitions? It's not just about saving a few billion dollars on office furniture. This strategy allows companies to cherry-pick the cream of the AI crop without inheriting any potential skeletons in the startup's closet. It's like adopting a star quarterback without taking on the entire team's losing streak. This approach gives tech giants the agility of a startup with the resources of a Fortune 500 company. They can quickly pivot into new markets or technologies without the long-term commitment of a full acquisition. However, these high-profile moves haven't gone unnoticed by regulatory bodies. The Federal Trade Commission and Department of Justice in the U.S. are closely monitoring these acqui-hires, concerned about their potential impact on market competition. In a notable development, the UK's Competition and Markets Authority has launched an inquiry into Microsoft's hiring of Inflection AI's executives, questioning whether it constitutes a de facto merger. In this new AI frontier, it seems the real intelligence lies not just in the algorithms, but in mastering the art of talent acquisition. As this latest round of corporate musical chairs plays out, one thing is clear: the race for AI dominance is as much about human capital as it is about cutting-edge technology.

AI’s Unauthorized Curriculum - YouTube: Big Tech has been caught with their hands in the YouTube cookie jar, and they’re not showing an ounce of shame. Apple, NVIDIA, and Anthropic have reportedly feasted on subtitles from over 170,000 videos, transforming everything from lectures to late night comedy into AI training data. The 170,000 YouTube videos at the heart of this AI-feeding frenzy weren't plucked from the ether by tech giants. Rather, they're part of a dataset dubbed "YouTube Subtitles," a subset of a more extensive AI training collection called "the Pile." This digital treasure trove was originally curated by EleutherAI, a nonprofit with the lofty goal of democratizing AI development. Apple is playing innocent, insisting their OpenELM model was merely a research project. NVIDIA defended their leaked Slack messages, which reveal a rather cavalier attitude towards copyright, by saying they were "in full compliance with the letter and the spirit of the copyright law.” (Apparently, their idea of spirit is more moonshine than legal doctrine.) One executive brushed off concerns, “given we are not publishing anything, there will be no negative sentiment." Ah, the sweet smell of plausible deniability. Both OpenAI’s Mira Murati and Microsoft’s Mustafa Suleyman have hinted that they believe that anything on the internet is “fair game.” It is the digital equivalent of finders keepers. Meanwhile, YouTube titans like Marques Brownlee and MrBeast seem less than thrilled about being unwitting guinea pigs in this AI experiment. Their "like and subscribe" mantra has evolved into "download and analyze" without their consent. (NB - as we were about to hit send, YouTuber David Milllette announced that he has filed a class action lawsuit against OpenAI seeking over $5M in damages. The data grab may be heading for another public reckoning.)

AI at Work: The Revolution is Messy, but Promising: Microsoft released a somewhat self-serving research report of the use of AI productivity tools like Copilot in the workplace. Who knew the company selling AI tools would find them useful? A large-scale randomized trial involving over 6000 employees found Copilot users read 11% fewer emails and edited 10% more documents. However, effects varied widely across roles and organizations. Customer service and sales pros reported the highest productivity gains and job satisfaction, while legal eagles were less impressed. (Apparently, AI hasn't quite mastered the art of billable hours.) Experience matters too - the longer people used AI tools, the more value they found. Intriguingly, 78% of AI users reported using unsanctioned tools, highlighting employees' hunger for AI assistance. The research also found that while AI can reduce mental load for some tasks, it may increase it for others as users grapple with evaluating AI outputs. Developers, for instance, crave AI help with routine tasks like unit testing, but worry about reliability and job security. Using GPT-4 to analyze tasks, researchers found AI-augmented search is shifting work towards higher-complexity activities. This underscores both AI's current limitations and its transformative potential. While it's already boosting productivity for many, realizing its full impact will require rethinking workflows and addressing ethical concerns. In other words, the AI revolution is here, it's messy, and it's forcing us to confront a terrifying new reality: we might actually have to learn how to use the office printer ourselves.

The Mixed Signals of AI Governance: In a world where AI moves faster than a New York minute, regulators are stuck in bureaucratic gridlock. While artificial intelligence is busy writing sonnets, curing diseases, and probably planning the next moon landing, lawmakers are still fumbling with their No.2 pencils. The past few weeks have seen a dizzying dance of AI developments and regulatory tap-dancing that's enough to give even the most caffeinated tech bros whiplash. OpenAI, of course, is playing both sides of the field: offering to let the U.S. AI Safety Institute peek under the hood of its next big model before it hits the streets, and endorsing several Senate bills faster than you can say "regulatory capture.” They're backing legislation that would give the AI Safety Institute some real muscle, plus a few educational initiatives. Meanwhile, the EU has fired the starting gun on its AI Act. As of August 1, the race to compliance is on, with a phased approach that'll keep lawyers and tech teams burning the midnight oil for years. It's a regulatory marathon, not a sprint, folks. And just when you thought you had it figured out, the White House throws a curveball. They're giving a thumbs up to open-source AI, at least for now. But we can’t even decide who’s supposed to be wearing the sheriff’s badge in AI town: the Federal Trade Commission (consumer protection) or the National Institute of Standards and Technology (risk management), or Congress (legislation or circus sideshow). In the race between silicon and bureaucracy, the finish line keeps moving.

— Lauren Eve Cantor

thanks for reading!

if someone sent this to you or you haven’t done so yet, please sign up below so you never miss an issue.

if you’d like to chat further about opportunities or interest in AI, please feel free to reply.

if you have any feedback or want to engage with any of the topics discussed in Verses Over Variables, please feel free to reply to this email.

we hope to be flowing into your inbox once a week. stay tuned for more!

banner images created with Midjourney.