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

We’ll publish intermittently until the end of the year and then return to our regular cadence in 2025.

AI Hype Cycle:

Navigating the Build, Buy, or Blend Decision

It's a familiar scene in offices worldwide: CEOs demanding to know their organization's generative AI strategy. In 2024, generative AI isn't just another tech buzzword—it's reshaping how enterprises operate, compete, and innovate. But with the hype reaching fever pitch, how do you separate genuine opportunity from digital smoke and mirrors? Let's dive into the real story of generative AI in the enterprise, backed by fresh data and battle-tested insights, from Menlo Ventures’s 2024: The State of Generative AI in the Enterprise.

AI’s Explosive Growth: The days of AI being just a fancy term for chatbots and recommendation engines are long gone. Enterprise investment in generative AI has skyrocketed to $13.8 billion in 2024—a staggering sixfold increase from 2023. This isn't just growth; it's a seismic shift in how businesses approach technology. The current landscape reveals an interesting split: 72% of decision-makers are bullish on AI's potential, and roughly a third still haven't found their AI north star. It's akin to having a superpower without the instruction manual.

Where’s the Money Going: The enterprise AI spending landscape reveals two major investment areas reshaping how businesses operate. First, there's a significant investment in foundation models (LLMs). These models serve as the brains of the operation, processing natural language with human-like accuracy, generating functional code, analyzing complex data patterns, and creating multimedia content at scale. Then there's the explosion in AI applications, where spending has surged from $600 million to $4.6 billion in just one year. This dramatic increase stems from businesses identifying an average of 10 high-impact use cases each. The experimentation phase is over—now it's all about execution and results.

Use Cases That Actually Work: In customer service, we're seeing a complete evolution of what's possible. Modern AI systems have moved far beyond basic chatbots, handling complex inquiries with remarkable context awareness and even predicting customer needs before they arise. These systems generate personalized solutions in real-time while knowing exactly when to bring in human agents for more nuanced situations. In the content creation and marketing, teams are revolutionizing their approach to campaign development. AI enables them to generate and test campaign copy at unprecedented scales, create personalized content for different audience segments, and optimize for SEO while maintaining a natural, engaging voice. The days of manually crafting every social media post are giving way to AI-assisted content strategies that drive real engagement. Developer productivity has seen perhaps the most dramatic transformation. Engineers now use AI to tackle complex debugging tasks in minutes rather than hours, generate essential boilerplate code automatically, and create optimized database queries. Even documentation, traditionally a developer's least favorite task, is becoming streamlined through AI assistance.

Implementation Challenges: The implementation of AI comes with significant hurdles. Data privacy and security concerns loom, from training data contamination risks to intellectual property questions surrounding AI-generated content. The challenge of maintaining proper data governance in a world of AI-generated insights keeps many CIOs up at night. Integration headaches persist as organizations struggle to make AI work with legacy systems. API standardization and performance optimization at scale remain significant technical challenges that require careful consideration and planning. The talent equation presents another critical challenge. The shortage of AI expertise and the accompanying salary demands create a significant entry barrier. Organizations must balance the need to upskill existing teams while building AI-literate leadership, all while competing for scarce technical talent in a hot market.

Build, Buy, or Blend: The landscape of AI implementation has shifted dramatically over the past year. Current data shows a near-even split in approach: 47% of solutions are developed in-house, while 53% are sourced from vendors. This represents a seismic shift from 2023 when 80% of enterprises relied on third-party generative AI software. The change signals growing confidence and capability among enterprises to develop internal AI tools rather than depending primarily on external vendors. The build approach offers complete control over your AI stack and custom solutions for unique problems. Organizations choosing this path gain data sovereignty and potential long-term cost advantages. Still, they must be prepared for significant upfront investment, extended time to market, and the challenges of talent acquisition. Buying solutions provides rapid deployment and proven technology with regular updates and improvements. While this path requires a lower initial investment, organizations must carefully consider limitations in customization options, potential vendor lock-in, and data privacy implications. Most intriguingly, many enterprises are finding success with a hybrid strategy. This approach typically begins with vendor solutions for quick wins, gradually layering in custom development for specific needs while building in-house capabilities. The essential advantage here is flexibility—organizations can adapt their approach as their needs and capabilities evolve.

The Bottom Line: The generative AI revolution is here, now. Success isn't about jumping on the bandwagon; it's about making strategic choices that align with business goals, technical capabilities, and risk tolerance. Whether building, buying, or blending, the key is to start small, think big, and move fast. The winners in this space won't necessarily be those with the most significant AI budgets, but those who approach AI with clear purpose and pragmatic strategy.

The AI Slowdown: When Hitting a Wall Hits Different

The latest viral moment in the tech world is an unexpected plot twist: artificial intelligence might be hitting its limits. OpenAI's Orion, Google's Gemini, and Anthropic's Claude aren't delivering the revolutionary leaps we've come to expect. But before you write this off as bad news, we'd like to show you why this apparent plateau might be the best thing to happen to AI since ChatGPT went viral.

Understanding the Technical Roadblock: Remember when every AI announcement felt like the tech equivalent of a Marvel movie premiere? Those days of "10x improvements" and "unprecedented capabilities" are giving way to something more... modest. We’ll break down why our favorite tech darlings are suddenly acting like they hit puberty:

1. The Scaling Law Ceiling: For years, AI was living its best life following scaling laws – basically, throwing more computing power at the problem and watching the magic happen. It worked beautifully... until it didn't. GPT-4 was the life of the party, but Orion? It's like showing up to the after-party only to find out everyone's already gone home. Doubling or even tripling computing power isn't giving us those jaw-dropping improvements anymore. It's like trying to make a better sandwich by adding more bread – at some point, you need to rethink the whole sandwich.

2. The Data Quality Crisis: Modern AI models require enormous amounts of high-quality training data, but the internet's readily available data has been largely exhausted. The remaining data often suffers from redundancy (most new content is derivative or repetitive); quality issues (machine-generated), bias; or a lack of expertise (actual expert content is often behind paywalls). Tech companies are trying everything short of time travel to fix this – synthetic data generation, expert content licensing, probably summoning rituals (kidding, mostly). But it's like trying to write a cookbook by photocopying other cookbooks. Something gets lost in translation.

3. Architecture Limitations: Our current AI models are like that friend who's good at remembering movie quotes but doesn't understand the plot. Current transformer-based architectures, while powerful, have inherent limitations in their ability to maintain consistent reasoning across long context windows, develop genuine understanding rather than pattern matching, and manage computational resource efficiencies at scale.

The Industry Response: The tech giants are pursuing different strategies to address these challenges. OpenAI is exploring new architectural approaches with their o1 project, focusing on improved reasoning capabilities rather than raw size. Google invests heavily in multimodal models, hoping to leverage diverse data types to overcome the text data plateau. Anthropic is reportedly working on novel training methods that could reduce data requirements.

This Crisis is a Gift: This slowdown represents a crucial turning point for the AI industry and its users:

1. The Great Equalizer: While tech companies recalibrate their approaches, smaller players and businesses have time to implement and perfect AI applications using current technology. This period of relative stability creates space for practical innovation in AI deployment.

2. Quality over Quantity: Instead of making AI bigger, we're making it better. Think of it as AI's "finding itself" phase, but instead of backpacking through Europe, it's learning to be useful.

3. Innovation’s New Frontier: When the obvious path hits a dead end, that's when things get interesting. We're not copying and pasting the same solutions anymore – we're having to get creative. Novel architectures, hybrid systems, efficiency improvements... the nerdy stuff that actually matters.

4. Building Trust Through Thoughtful Development: The slower pace allows for proper consideration of AI safety, fairness, and transparency. Companies can focus on building robust systems rather than racing to be first to market.

The Revolution is in the Application: This AI slowdown isn't the end of the story – it's the part where things get interesting. While the headlines scream about plateaus and limitations, the real revolution is happening in small offices, startups, and that one person's laptop at your local coffee shop. Remember: electricity wasn't revolutionary because someone invented a better generator – it was revolutionary because people figured out a million exciting ways to use it.

Back to Basics

AI + Art: The Great Creative Collision

We've been immersed in the world where artificial intelligence meets artistic expression, and we took a deep dive into the ArtReview X NOWNESS—AI+ study. This comprehensive analysis shows how AI is reshaping creative possibilities in exciting and challenging ways. Whether you're a digital artist or a traditional creator, these changes are coming.

AI as a Creative Catalyst: The reality of AI in creative spaces might surprise you. Rather than replacing human creativity, AI has emerged as a sophisticated collaborative tool. The numbers tell an interesting story - 38% of artists and filmmakers now work with AI multiple times weekly, primarily for brainstorming, writing, and image generation. Tools like ChatGPT and Midjourney have become valuable allies in the creative process, helping streamline workflows and spark new ideas. This shift brings both opportunities and challenges. While 60% of creators celebrate AI's ability to save time, and 49% value its problem-solving capabilities, important concerns have emerged. A significant 55% worry about misinformation, 34% have concerns about copyright implications, and 38% think about potential job impacts. These statistics reflect the complex reality of integrating AI into creative work.

The Authenticity Paradox: The data reveals fascinating insights about how we perceive AI-created work. A third of study participants couldn't distinguish between human-made and AI-generated creations - a finding that opens up exciting discussions about creativity and authenticity. While 55% of people are open to experiencing AI-generated films, 56% hesitate to purchase AI-created artwork, highlighting an intriguing split between consumption and ownership of AI-generated art.

Fundamental Shifts in Creative Practice: The integration of AI into artistic practice has catalyzed four major transformations:

1. Digital Reality Evolution: We're witnessing a transformation in how digital art is created and perceived. The increasing sophistication of AI-generated content pushes us to reconsider our understanding of artistic creation and authenticity.

2. Redefinition of Craft: Traditional artistic skills blend with digital expertise in exciting ways. Artists are expanding their toolkits to include conventional techniques and AI capabilities, creating new forms of creative expression.

3. Creative Amplification: AI is amplifying human creativity in unexpected ways. Working with existing data enables artists to explore new creative territories and push boundaries in fascinating directions.

4. Accessibility and Bias: More people now have access to sophisticated creative tools, though we need to be mindful of potential biases in AI systems. The goal is to ensure this democratization truly benefits diverse creative voices.

What Lies Ahead: AI's role in creative fields represents something more significant than a passing trend. It's changing how we approach creativity, ownership, and artistic value. While we are genuinely excited about the possibilities - particularly in making creative tools more accessible - we believe it's crucial to maintain thoughtful engagement with these developments. The future of creativity is about more than choosing between human and machine capabilities. Instead, it's about finding meaningful ways to combine both, enhancing our creative expression while preserving the authentic human elements that make art resonant and meaningful.

AI Speaks Your Language, Not the Other Way Around

As Conor Grennan astutely points out, we've overcomplicated talking to AI. Somewhere between the first ChatGPT prompt and today's elaborate tutorials, we've turned a natural conversation into something that sounds like it requires a PhD.

Your Brain’s Hidden Superpower: Remember learning to read a room? That sixth sense you developed for knowing when your boss is saying "yes" versus saying "yes, but"? That's not just social intelligence or EQ—it's the foundation of effective AI interaction. Your brain has spent years mastering the art of extracting meaning from subtle cues and adapting your communication style on the fly. Think about how you tackle a conversation with a new client versus chatting with your closest friend. Without thinking about it, you're processing context, adjusting your language, and fine-tuning your message for maximum impact. This intuitive ability to shape your communication style isn't just a social skill—it's the same talent that makes you a natural at directing AI. You're not learning a new language; you're simply adding another dialect to your impressive communication repertoire.

Here's the technical truth that prompt engineering gurus often overlook: AI uses Natural Language Processing (NLP)—and no, that's not a marketing euphemism. These systems are specifically designed to parse and understand human language in its natural form. They're built to process how you speak and write, not some specialized coding language.

The term "prompt engineering" itself is part of the problem. It's like calling a conversation "verbal data transmission"—technically accurate but unnecessarily intimidating. Modern AI isn't trying to test your coding skills; it's designed to meet you where you are. Think of it as the eager new hire who has read every manual but needs your real-world wisdom to apply that knowledge effectively. When organizations report saving billions with AI, they're not achieving this through complex, prompt engineering frameworks. They succeed because people apply their natural communication skills to a new medium. The process is simple: express needs clearly, provide context, give feedback, and refine based on results.

Stop treating AI interaction as if it requires special certification. Your natural communication instincts are your best guide. The same skills that help you navigate office politics, client relationships, and family dynamics are ideally suited for AI collaboration. Think of AI as your enthusiastic new team member: incredibly knowledgeable but looking to you for direction and context. Most AI models can help you determine what you are looking for—just ask them how to improve your prompt.

Tool Update

Some of our favorite tools got upgraded over the past week, so we’ll just hit the highlights.

LTX Studio goes Open Source: LTX Studio is where AI meets filmmaking magic—a platform designed to take your wildest storytelling dreams and make them real. From scripting to storyboarding to full-on video generation, LTX Studio turns the art of filmmaking into something anyone can do, whether you’re a seasoned pro or just dipping your toes in the creative pool. Now, about that update—they’ve seriously outdone themselves. First off, LTX went full open source. Second, the interface has had a glow-up, so you’ll spend less time hunting for buttons and more time creating. Everything runs faster because waiting for renders is so last decade. And for those deep in the AI trenches, you’ll love the expanded compatibility with other models and frameworks. It’s like LTX Studio opened the door to a much bigger playground.

Runway thinks outside the Frame: Runway just turned up the heat with their latest trick—say hello to Expand Video. This feature lets you take your existing videos and stretch their horizons—literally. Need a new aspect ratio? No problem. Expand Video generates brand-new content around your original footage, seamlessly filling in the gaps like it was always meant to be there.

Suno turns it up to 11: Suno just cranked the volume with the release of its highly anticipated v4 model, and it’s clear they’ve been listening to what creators want. This isn’t just an upgrade—it’s a whole new era for AI music generation. Suno v4 brings cleaner instrumentals, remarkably human vocals, and song structures that feel alive and dynamic. One of the standout features has to be Personas—capture the vibe of a vocal style and carry it across projects for a cohesive sound. The ReMi Lyrics Model generates edgier, more unpredictable lyrics, while the Covers tool lets you reinvent songs in ways that weren’t possible before.

ElevenLabs elevates the Chat: ElevenLabs has just rolled out the ability to build conversational AI agents directly on their platform. This isn't just a minor tweak; it's a full-scale upgrade that lets you craft AI agents with customizable variables like tone of voice and response length. Previously, ElevenLabs was all about voice cloning and text-to-speech services. However, they noticed that many clients were already using their tools to create conversational agents despite the challenges of integrating knowledge bases and handling customer interruptions. So, they decided to streamline the process by offering a complete pipeline for building these bots.

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

Cloud Meets Claude: Amazon and Anthropic are making waves—not the small kind that ripples quietly across the industry. These two titans just leveled up their relationship, and it’s all about bringing artificial intelligence to new heights with some seriously robust cloud support. Amazon’s pumping another $4 billion into Anthropic, bringing its total stake to a hefty $8 billion. But this is no casual fling. With AWS crowned as Anthropic’s go-to cloud and training partner, the collaboration is diving deep into hardware-software synergy. AWS’s custom silicon is getting tuned to turbocharge Claude, Anthropic’s suite of AI models. This partnership isn’t happening in a vacuum. It’s a move on the chessboard against Microsoft’s deep integration with OpenAI. And while Anthropic has been clawing its way up in the enterprise AI space (doubling its market share in just a year), this Amazon cash infusion might cement its place as a serious contender. Things get a bit spicy here: regulatory eyes are locked on this partnership. The FTC and the UK’s Competition and Markets Authority have been sniffing around. The CMA just gave it the green light, but you can bet this isn’t the last we’ll hear about potential competition concerns. Amazon and Anthropic are setting up a long-term game plan that’s as much about reshaping the AI landscape as it is about proving the worth of high-performance, cloud-native infrastructure. It’s a fascinating blend of ambition, technical prowess, and (let’s face it) a touch of corporate drama.

Browser Wars: The world of web browsers—a space that seemed locked down by Chrome’s dominance—is suddenly full of suspense. Two significant shifts are shaking the industry: the DOJ’s antitrust push against Google, which might force a spinoff of Chrome, and OpenAI’s ambitious leap into browser development. Together, these moves could redraw the internet’s power map. The DOJ is coming for Chrome, and it’s not just a slap on the wrist—it’s a breakup attempt. Google’s browser, commanding 67% of the market, is the linchpin of its search dominance. By forcing a Chrome spinoff, the DOJ aims to level the playing field, letting rival search engines finally compete for the default spot. For Google, losing Chrome would be like cutting off its data supply, weakening its advertising empire. Meanwhile, users might face a less polished Chrome, as the browser loses Google’s deep resources for innovation and security. Now, cue OpenAI, stepping into the spotlight with whispers of its browser ambitions. The timing here is fascinating, as OpenAI has already been making waves with its SearchGPT—a bold move into the search engine space. Building a browser to pair with its AI tools seems like a natural progression. OpenAI isn’t playing around either; they’ve hired Ben Goodger, one of Chrome’s founding team members, signaling their intent to hit the ground running. Imagine a browser purpose-built for AI integration. Search isn’t just reactive; it’s proactive. You’re not typing; you’re conversing. And beyond search, AI functionality is baked right into the browser experience. Need real-time data analysis? Done. Writing assistance? Seamless. A shopping assistant that understands your preferences? Let’s go. OpenAI could aim to redefine what a browser is, transforming it from a gateway to the web into an intelligent assistant that redefines productivity and convenience. Of course, ambition doesn’t equal execution. Building a browser from scratch is no small feat. Chrome and Edge have set the bar high with speed, extensions, and stability. OpenAI’s challenge will be matching those benchmarks and convincing users to switch. There’s also the question of market readiness. Does the world want an AI-first browser? The interplay between these developments—the DOJ’s attempt to dismantle Google’s search-and-browser monopoly and OpenAI’s potential rise as a browser innovator—could lead to a fascinating reshuffling of power.

Quantum Leap: This is very meta—and not just in a “self-referential” way. Google DeepMind’s new AlphaQubit AI has leaped into the heart of quantum computing’s Achilles' heel: error correction. Think about it—an AI model built to think like a language translator now speaks fluent “quantum” by decoding and fixing the errors that naturally arise in the hyper-sensitive, wildly complex world of qubits. AlphaQubit combines the predictive prowess of Transformers (the kind that powers your favorite generative AI models) with unprecedented accuracy in identifying and correcting quantum errors. We’re talking 6–30% better performance compared to traditional decoders—numbers that make quantum engineers weak at the knees. Its adaptability means it can scale to hundreds of qubits and keep pace with the error-correction marathon. DeepMind and Google Quantum AI have given us a roadmap for fault-tolerant quantum systems. And while it’s not quite real-time-ready, the implications for fields like drug discovery, climate modeling, and cryptography are massive. AlphaQubit proves that the sky—or maybe the multiverse—is the limit when AI gets quantum-savvy.

— Lauren Eve Cantor

thanks for reading!

if someone sent this to you or you haven’t done so yet, please sign up 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.