Verses Over Variables

Your guide to the most intriguing developments in AI

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.

AI Hype Cycle

Decoding Opportunity in BOND's Landmark Trends Report

Alright, the tech world just got its latest hefty download: BOND's May 2025 "Trends - Artificial Intelligence" report. If you're familiar with BOND, or the influential work of the analyst Mary Meeker, you'll know these reports are incredibly detailed forecasts of where technology is steering us. This new 340-page edition is laser-focused on the Artificial Intelligence wave that’s currently reconfiguring, well, just about everything. Now, the headline stats about AI’s explosive growth and the jaw-dropping investments are certainly making the rounds, and they are indeed significant. But we decided to sift through the data for some of the less obvious connections and the practical, perhaps even surprising, implications for anyone tuned into the AI frequency. So, let's get into what we found beyond the big, bold numbers.

First, let’s talk about infrastructure. The "Big Six" US tech firms alone invested $212 billion in AI-related capital expenditures (CapEx) in 2024, a 63% year-over-year increase. The significance here extends far beyond an increase in data centers. AI is forging an entirely new class of global infrastructure. We're seeing the rise of what NVIDIA’s CEO calls "AI Factories", powered by custom-designed silicon from giants like Google and Amazon. This signifies a move towards specialized hardware and compute power becoming as fundamental as the electrical grid was to the last industrial revolution. Access to, and innovation in, this new infrastructure will define economic strength and competitive advantage for decades to come.

Next, while AI is undeniably a powerful "copilot," the report subtly suggests we need to think bigger: about AI Co-Evolution. It's about fundamentally intertwining our workflows and creative processes with rapidly evolving AI capabilities. We see companies like Shopify and Duolingo already embedding AI use as a "fundamental expectation". The explosive 448% growth in US AI job postings over seven years signals a broader demand for an AI-literate workforce. The insight is that the most critical human skill in this new era will be our capacity for continuous learning, adaptation, and the ability to direct and integrate AI at a strategic level. For creatives, this means becoming architects of AI-augmented creativity, constantly pushing the boundaries of what’s possible as the AI itself learns and grows alongside us.

Then there's the idea of how the "next billion" will experience the digital world. The report notes that 2.6 billion people are still offline, and their first encounter with the internet will likely be AI-native, perhaps through conversational, agentic systems in their own languages. This signifies a potential "generational skip" in user experience, offering an alternative to traditional app stores, browsers, and search bars. The implication is the emergence of entirely new markets and modes of digital interaction, presenting fresh challenges and immense opportunities for current platforms to adapt. For professionals and businesses, this opens up uncharted territories for engagement, content delivery, and global reach, demanding new thinking about how to connect with an audience whose digital "mother tongue" is AI.

Finally, let’s consider the AI gold rush. The report highlights incredible revenue growth for AI companies. Yet, it also points to the immense compute costs shouldered by foundational model providers like OpenAI (estimated $5 billion in compute expenses in 2024) and the rapidly falling costs of AI inference for users. The "thing unsaid" here is an interesting paradox: while AI will undoubtedly create trillions in value, the most sustainable path to capturing that value may well lie in the creative and strategic application of AI. As foundational AI capabilities become more accessible and even commoditized, real differentiation will come from using these powerful tools to solve specific, high-value problems in unique ways, or embedding them deeply to supercharge existing, profitable businesses.

BOND's report confirms we're in an era of unprecedented technological acceleration. However, looking beyond the immediate statistics, we see the contours of a future where infrastructure, skill sets, user experiences, and value creation are all being profoundly reimagined.

When AI’s Creator Sounds the Alarm

We've reached one of those inflection points where the future arrives faster than anyone expected. While most AI companies have been carefully managing their messaging around job displacement, Dario Amodei decided to break ranks. The CEO of Anthropic (the company behind Claude) just delivered the kind of unvarnished warning that usually comes from outside critics, not industry leaders: AI could eliminate half of all entry-level white-collar jobs within five years. We’ve been watching the AI job displacement conversation unfold with a mixture of fascination and existential dread. But when one of the architects of the technology stands up and says, "we need to raise the alarm," it hits differently. This isn't a doomer or academic speculation. This is the person whose company just released Claude 4, telling us we should probably be worried.

Currently, 60% of people use AI for augmentation, and 40% use it for automation, with the latter segment growing. Translation: we're rapidly moving from "AI helps me do my job better" to "AI does my job instead of me." For those of us in creative fields, this shift feels especially personal. Almost half (49%) of 400 graphic professionals surveyed across the UK, US, France and Germany reported seeing the manual graphic production industry becoming obsolete within five years. Meanwhile, 81.6% of digital marketers hold concerns about content writers losing their jobs due to AI's influence. Ad agencies are eliminating creative and copywriting positions as generative AI produces marketing materials more quickly and cost-effectively. Notably, even Grammarly (a company founded on augmenting human writing) laid off 230 employees (roughly 20% of staff), explicitly citing AI's growing capability to perform editing and writing tasks autonomously.

The displacement extends far beyond creative work. Microsoft laid off 6,000 workers in 2025, nearly 3% of its workforce, with CEO Satya Nadella noting that 30% of the company's code is now written by AI. "AI in 2024 has already led to the automation of administrative tasks, which, in turn, is leading to a reduction in entry-level roles," explains Hugo Farinha, co-founder of AI testing provider Virtuoso QA. Muddu Sudhakar, CEO of AI solutions company Aisera, reports seeing "a huge displacement of white-collar workers" in basic software developer jobs and database administrator roles. Even traditionally stable consulting firms are feeling the pressure. PwC laid off 1,500 employees in May 2025, while Deloitte has begun workforce reductions as federal consulting contracts get slashed by government cost-cutting initiatives. The Big Four accounting firms are all reportedly examining staff reductions as clients increasingly turn to AI for tasks previously handled by junior consultants.

The data paints an increasingly stark picture. According to recent surveys, 41% of employers plan to reduce their workforce due to AI automation by 2030, while 14% of workers report they've already experienced job displacement due to automation or AI. A study of business leaders found that half of CEOs believe they may replace jobs with AI, with the figure rising even higher among C-suite executives. MSN replaced dozens of journalists with AI content generation back in 2020, while Duolingo cut 10% of its contractor workforce as the company pivoted to AI for translation work. The language learning app's CEO announced in April 2025 that the company would be "AI-first moving forward," using AI where they previously would have hired contractors.

What makes Amodei's warning particularly compelling is its source. Amodei detailed these grave fears after spending the day onstage touting the astonishing capabilities of his own technology. The irony isn't lost on him either. He describes the current situation where AI leaders are both building more powerful tools and privately worrying about them as "a very strange set of dynamics". Mark Cuban fired back, arguing that "new companies with new jobs will come from AI and increase total employment", pointing to historical examples like the displacement of secretaries. It's the classic technological optimism playbook: every wave of automation eventually creates more opportunities than it eliminates.

Understanding what's driving this acceleration requires grasping what AI researchers call "agentic AI." These agents are powered by large language models and can do the work of humans, instantly, indefinitely and exponentially cheaper. Imagine an agent writing code, handling finance analysis, customer support, marketing, or content distribution. The transformation is already visible in how companies approach hiring. Axios reports that every single CEO they spoke with is "working furiously to figure out when and how agents or other AI technology can displace human workers at scale".

The path forward starts with Amodei's most practical advice: "Learn to understand where the technology is going. If you're not blindsided, you have a much better chance of adapting.” However, adaptation extends beyond simply becoming comfortable with AI tools. For professionals, this may mean transitioning from execution-heavy roles to those focused on strategy, direction, and quality control. The research suggests that there is still significant value in human creativity and judgment. AI writing tools struggle to match the quality and creativity of talented human writers when it comes to critical thinking and audience-specific decision-making. Similarly, AI art generators are skilled at producing high-quality pieces, but these are only based on existing artworks and styles.

Amodei's most sobering observation might be his admission that "I don't think we can stop this bus." But he adds a crucial qualifier: he can "maybe hope to do a little to steer the technology in a direction where we become aware of the harms, we address the harms, and we're still able to achieve the benefits." That steering requires all of us to engage more seriously with both the technology and its implications. For professionals, it means moving beyond either wholesale rejection or uncritical embrace toward thoughtful integration and strategic positioning. We're living through one of those historical inflection points where the choices we make today will determine whether technological progress serves human flourishing or undermines it. Amodei has issued the warning from inside the house. Now we get to decide what we do with that information.

Back to Basics

The Search Revolution is Here

ChatGPT handled 365 billion queries in 2024. That’s 1 billion queries per day flooding AI platforms for instant answers. It took Google eleven years to reach that same milestone back in 2009, and ChatGPT only two years. But what should make every professional pay attention: Vercel reports that 10% of their new signups now come directly from ChatGPT. Not Google. This isn't some distant future scenario, this is happening right now, transforming how people discover content, products, and services. Welcome to Generative Engine Optimization (GEO), where success isn't about ranking #1 on Google anymore. It's about becoming the source that AI chooses to cite, quote, and recommend. Traditional search was built on links. GEO is built on language. When someone asks ChatGPT about the best project management tools for creative teams, you want your platform to be the one it mentions, not buried on page two of Google results they'll never see.

The transformation is measurable and dramatic. AI-native search queries average 23 words compared to Google's 4-word average, with sessions lasting 6 minutes instead of seconds. People are fundamentally changing how they search. They're having conversations with AI about their problems, and that AI is choosing who gets to be part of that conversation. Only 30% of ChatGPT prompts fall into traditional search-like behavior, meaning 70% represent entirely new ways people seek information. The results speak volumes across industries, with concrete success stories emerging from early adopters. The finance sector dominates LLM-driven traffic, capturing 84% of all referrals analyzed, largely due to integrations that enable seamless access to financial data through platforms like Perplexity. ChatGPT referrals surged by 900% in the events industry over 90 days, while e-commerce and finance sectors experienced 400%+ growth in AI-driven traffic. These metrics reveal fundamental shifts in how customers discover solutions.

Large brands, like Canada Goose, have discovered this firsthand when they use AI monitoring tools to understand not just how LLMs referenced their products' features, but whether the model spontaneously mentioned the brand at all (an indicator of unaided awareness in the AI era). This transformation demands understanding how LLMs fundamentally differ from traditional search engines. They benefit from clarity, context, and structure in ways conventional search doesn't.

Google included an llms.txt file in their new Agents to Agents protocol, while Anthropic, creator of Claude, specifically worked with Mintlify to implement llms.txt standards for their documentation. When industry leaders actively adopt these standards, it signals effectiveness beyond theoretical arguments. Think of llms.txt as your content's resume for AI, a clean, organized summary that helps intelligent systems quickly understand what you're about and why they should trust you enough to cite you. Companies like Windsurf report that llms.txt saves time and tokens when AI agents don't need to parse complex HTML. But the technical infrastructure is only the foundation.

Traditional SEO rewards precision and repetition. Generative engines prioritize content that's well-organized, easy to parse, and dense with meaning. Phrases like "in summary" or bullet-point formatting help LLMs extract and reproduce content effectively. Creative professionals have a massive advantage here because your instinct for clear communication, compelling narratives, and organized information directly translates to AI optimization. The same skills that make your client presentations memorable make your content quotable by AI. What's working now includes conversational content structure that uses natural language answering specific questions, building authority through expertise where AI models prioritize content demonstrating clear knowledge and citations, and maintaining geographic and entity clarity that makes it crystal clear who you are, where you're located, and what you specialize in. Blog content dominates LLM referral traffic, receiving 77.35% of visits, while product pages capture less than 0.5%, highlighting the importance of informational content in AI discovery.

The business dynamics create even more opportunity. The LLM market fundamentally differs from traditional search in business model and incentives. Classic search engines monetized through ads where users paid with data and attention. Most LLMs are paywalled, subscription-driven services. When AI platforms cite your work, they're not trying to keep users on their platform to show ads. They're providing value to paying subscribers. This creates better attribution, higher-quality referrals, and users who arrive with serious intent.

The timing couldn't be better for early movers. Reference rates (how often your brand or content is cited in model-generated answers) are becoming as important as click-through rates. Your path forward requires focus on fundamentals: creating clear, structured content answering specific questions, implementing basic llms.txt files since many platforms generate these automatically, monitoring how AI systems reference your work, and building authority through consistent, expert-level content.

Your audience now includes LLMs alongside humans, and optimizing for AI means ensuring accurate, authoritative representation across intelligent systems. This represents a shift in content accessibility thinking. In a world where AI becomes the front door to commerce and discovery, the crucial question for marketers is: Will the model remember you?

Tools for Thought

May was a significant month for new product releases and announcements from most of the major AI companies. We’ll try to hit the highlights, especially the tools we love or can’t wait to try, but as usual, it was a whirlwind.

Google Drops the Mic

Google just delivered its most ambitious developer conference yet at I/O 2025. The scope of innovation they unveiled at I/O 2025 was so bold that they summed it up through a post called "100 Things We Announced at I/O." Here are just some of them.

  • Gemini 2.5 Pro Deep Think Mode: Google introduced an experimental enhanced reasoning mode called Deep Think, which utilizes cutting-edge research techniques, including parallel thinking, to enable the model to consider multiple hypotheses before responding.

  • Gemini 2.5 Flash Production Release: The updated 2.5 Flash model delivers 20-30% improved efficiency while maintaining breakthrough performance on reasoning, multimodality, code, and long context benchmarks.

  • Native Audio Output Integration: Both 2.5 Pro and Flash models now feature native audio output capabilities, enabling natural conversational experiences and seamless switching between 24 languages with consistent voice profiles.

  • Veo 3 with Synchronized Audio-Video Generation: Veo 3 becomes the first state-of-the-art video generation model capable of producing synchronized audio alongside video content, including environmental sounds, background music, and character dialogue with accurate lip-syncing.

  • Flow: Professional AI Filmmaking Platform: Flow integrates Veo, Imagen, and Gemini models into a comprehensive filmmaking tool featuring camera controls, scene builders, asset management, and continuous character consistency across multiple clips.

  • Jules Autonomous Coding Agent: Jules enters public beta as a "true coding agent" capable of reading large codebases, understanding developer intent, writing tests, building features, fixing bugs, and updating dependencies autonomously.

  • Stitch AI-Powered UI Generator: Google's new Stitch tool transforms app development by generating production-ready UI designs and frontend code from simple natural language descriptions or image prompts.

  • AI Mode Universal Rollout: Google launched AI Mode for all U.S. users, representing a total reimagining of Search with advanced reasoning capabilities that support queries 2-3 times longer than traditional searches.

Microsoft focuses on Agents and the Enterprise

Microsoft doubled down on what they do best: enterprise transformation at scale. Announcing over 50 AI tools centered around "building the open agentic web," Microsoft positioned GitHub Copilot as evolving from an in-editor assistant to a fully autonomous coding agent capable of operating as a team member, refactoring code, fixing defects, and implementing features across the entire software development lifecycle. The introduction of Microsoft 365 Copilot Tuning enables organizations to train AI models using their own company data, eliminating the need for data scientists or weeks of development time. Meanwhile, multi-agent orchestration in Copilot Studio enables specialized agents to collaborate as a team.

Anthropic Bets on Coding

Rather than competing on sheer volume of features, Anthropic focused its firepower on solving the fundamental challenge that has plagued AI coding tools: sustained, multi-hour autonomous development workflows that actually work. At its inaugural Code with Claude conference, Anthropic also made several announcements:

  • Claude 4 Opus and Sonnet 4 Release: Anthropic launched Claude 4 Opus and Claude 4 Sonnet, setting new standards for coding, advanced reasoning, and AI agents. Opus 4 is positioned as "the world's best coding model," and Sonnet 4 delivers a significant upgrade over Claude 3.7. Both are hybrid reasoning models offering instant responses or extended thinking modes, capable of maintaining sustained performance across thousands of steps over multiple hours without losing focus.

  • Claude Code General Availability: Claude Code transitions from research preview to full production with background task support. Anthropic's Chief Product Officer, Mike Krieger, noted that with just 2-6 engineers supporting multiple mobile platforms, Instagram could have produced prototypes "in days, not weeks" using Claude Code.

  • Enterprise Security and Safety: Claude 4 launches under enhanced safety controls, including strengthened harmful content detectors and cybersecurity defenses, meeting Anthropic's ASL-3 model specification for responsible deployment.

OpenAI Acquires

OpenAI capped off the most competitive week in AI history with a strategic double punch that signals its evolution from software pioneer to platform powerhouse. A $3 billion acquisition of Windsurf, an AI-assisted coding tool that competes directly with Microsoft's GitHub Copilot, followed immediately by a stunning $6.5 billion all-stock deal for io, the secretive hardware startup co-founded by legendary iPhone designer Jony Ive. The Ive acquisition brings 55 engineers, designers, and researchers (including former Apple veterans who helped create the iPhone, iPod, and iPad) under OpenAI's creative direction, with the first AI devices expected to debut in 2026. These back-to-back acquisitions represent a calculated bet that the next phase of AI dominance requires owning the entire stack from foundational models to the physical devices that bring AI into users' hands.

And There is More…

  • DeepSeek R1 Stealth Upgrade: Chinese startup DeepSeek quietly released an upgraded version of its market-disrupting reasoning model, delivering significant improvements in depth of reasoning, inference capabilities, and hallucination reduction while achieving performance closer to OpenAI's o3 and Google's Gemini 2.5 Pro.

  • Mistral's Devstral: French AI startup Mistral released Devstral, a 24-billion parameter open-source model designed explicitly for autonomous software engineering workflows.

  • Perplexity Labs: Perplexity launched Labs, a comprehensive tool suite that can craft reports, spreadsheets, dashboards, and interactive web apps through 10+ minutes of autonomous work combining deep web browsing, code execution, and asset generation.

— 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.

we’ve also started publishing more frequently on LinkedIn, and you can follow us here

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.

banner images created with Midjourney.