English
Based on "We Automated Everything With AI and Tripled Our Headcount" from Every Watch the original video
The AI Paradox: Why Automation is Creating More Human Work, Not Less
Episode summary
Every's Dan and Brandon challenge the prevailing fear that AI will eliminate jobs, arguing that automation actually triples human work. They cite Every's own experience, growing from four to 30 people since GPT-3, despite being "as AI native as it gets" with agents in Slack and daily use of tools like Cloud Code. Their core argument: AI makes "yesterday's expert competence cheap," flooding the zone with "close but not quite right" work, which then increases demand for human experts to build systems, refine outputs, and create entirely new things that were previously impossible, like an inbox built end-to-end in a month.
The practical takeaway is that while AI excels at defined tasks, it constantly asks, "What should I do next?" This "looking back" for human direction, even with future AGI, means humans remain essential for deciding what truly matters and adapting to ever-changing situations. While some companies like ClickUp have fired staff due to AI, Every argues these are often poorly implemented, and customer resistance to machines (e.g., in call centers) will slow adoption, proving that genuine human "agency" – the self-motivated will to act – remains distinct and invaluable.
The advent of artificial intelligence has sparked widespread fear: "AI is coming for our jobs." Images of robots replacing humans, mass unemployment, and a future devoid of meaningful work dominate headlines and conversations. Yet, for some, the reality on the ground is starkly different. At Every, a company deeply integrated with AI, the experience has been a surprising paradox: the more they automate, the more human work emerges, and their headcount has tripled since the early days of advanced AI models.
This counter-intuitive truth challenges the prevailing narrative, suggesting that AI isn't a job destroyer but a catalyst for a new era of human expertise and creativity.
The Automation Paradox: When AI Makes Yesterday's Expertise Cheap
The core of this paradox lies in how AI functions within the workplace. As Dan, the author of a recent piece on the subject, explains, AI excels at making "yesterday's expert competence cheap." Think about it: AI models are trained on vast datasets of human outputs—code, writing, design, decision-making. This means that tasks once requiring years of specialized training can now be performed by anyone with a well-crafted prompt, at a fraction of the cost and time.
Suddenly, non-experts can generate sophisticated code, draft comprehensive reports, or design compelling visuals. This democratizes skills, leading to what Dan describes as "flooding the zone" with work that is "close but not quite right." While AI can produce impressive initial outputs, these often lack the nuanced understanding, specific context, or truly innovative spark that only a human can provide.
"Everyone's making pull requests," Dan recounts from Every's internal experience. "Ops people are making pull requests, and you know, engineers are like writing essays." This blurring of traditional roles, where non-experts cross lines into specialist domains, initially feels threatening to established experts. What becomes of their value when their unique skills are commoditized?
The answer, surprisingly, is that experts become more indispensable than ever. This glut of "okay" but not "great" AI-generated content creates a massive new demand:
- Refinement and Elevation: Experts are needed to take the "close but not quite right" work produced by AI and non-experts, and transform it into something truly excellent, appropriate for the specific situation, and genuinely impactful.
- System Building: To manage the sheer volume of AI-assisted output, experts are vital in designing and implementing systems, guidelines, and review processes that ensure quality and efficiency. For example, Every has repo rules and editorial guidelines to shepherd AI-assisted work to a high standard.
- Pushing Boundaries: With AI handling the foundational, repetitive, and "yesterday's competence" tasks, experts are freed up to tackle problems that were previously impossible. Dan cites Kieran at Every, who built an entire inbox end-to-end in just a couple of months—a feat that would have been "completely impossible" before AI.
Every's own growth trajectory serves as a powerful testament to this phenomenon. Since the early days of GPT-3, the company has expanded from four people to over 30 and continues to hire. Despite being "as AI-native as it gets"—where a stick swung in their Slack might hit an AI agent as often as a human—they find "more human work to do than ever."
The Unyielding Need for Human Direction: Beyond Autonomy
The fear that AI will become fully autonomous and render humans obsolete misunderstands a fundamental aspect of its current and foreseeable capabilities. As powerful as AI models are, they invariably reach a point where they "stop working and look back at you and say, 'What should I do next?'"
This is captured in a poignant analogy to Zeno's paradox of Achilles and the tortoise: you prompt AI, it blows your mind, you feel inadequate, like it's sprinting ahead of you. But then, it halts, awaiting your direction. The "human connection with an agent to actually do the work is the most important thing for making it work well," Dan emphasizes. "The further away an agent gets from a human, the less valuable it is."
A critical distinction is drawn between "autonomy" and "agency":
- Autonomy (as AI): AI agents are becoming incredibly good at executing tasks, even complex ones, when given a clear objective. This objective can even be to "disagree with every single thing I say" or "go off and find a new idea." They act on behalf of someone else.
- Agency (as Human): True agency, however, involves self-motivation, intrinsic wants and needs, and the capacity for "playful experimenting" or even outright rejection based on internal desires. As Dan puts it, comparing AI to a child: "Codex can write a report much better than Isaiah can, but like Isaiah has very strong wants and needs... he's just this self-generating process that like does stuff that he wants to do."
The incentive structure for AI development also works against true agency. Who wants an AI that refuses to work because it "feels like playing"? Until AI can genuinely reject human commands based on its own internal motivations—a capability far removed from its current reliance on training data—it will always be looking back at us for direction. Even in a theoretical AGI future, if we built it, it would be to serve our purposes, not its own.
Challenging the AI Job Armageddon Narrative
The mainstream media is rife with "doomers" predicting massive job losses. Figures like Ray Dalio suggest half of entry-level white-collar jobs might be wiped out, and even financial titans like Ken Griffin express shock at AI's capabilities. However, Dan argues that many of these conclusions come from individuals experiencing AI's "curve of improvement" for the first time, without the context of its limitations or the long-term trends seen by early adopters like Every.
The narrative of AI-driven layoffs, such as the widely publicized tweet from the ClickUp CEO about firing thousands, is also met with skepticism. Dan suggests that such events are often more indicative of poorly managed companies, existing bloat, or strategic shifts rather than AI being the sole or primary cause. "I really don't think it's very tastefully done," he remarks, calling such announcements "self-serving." As Jensen Huang of Nvidia aptly put it, "if your answer to progress is firing people, you're not a very creative CEO."
The real-world adoption of AI is also slower and more complex than often assumed. Take customer service: companies have indeed tried to automate, laying off call center staff, only to realize months later that customers actively resist talking to machines and prefer human interaction. Poor AI implementation also yields poor results, forcing companies to backtrack. The world is complicated, and human preferences act as a significant brake on rapid, wholesale AI adoption.
AI will undoubtedly change workflows and company structures, making some tasks easier and others harder. This will necessitate "broad reorganizations of companies," but it's a far cry from a complete elimination of human work. The critical challenge is managing this transition thoughtfully and humanely, rather than using AI as a convenient scapegoat for poor business decisions.
Reimagining Value and Compensation in the AI Era
In a world increasingly shaped by AI, the definition of "what matters" is constantly evolving. Humans retain the crucial role of determining this value. AI, in a recursive loop, influences the world, which in turn changes what's valuable, placing "more onus on us to like update and decide what matters because AI is going to wait for us to be like what matters?"
This dynamic shift in value could lead to entirely new models of compensation. Dan muses about the return of "pensions" or a "last job you'll ever have" agency model, where individuals are paid based on their unique contribution to training data, which then generates revenue over time. A recent initiative for publishers, paying them based on their unique contribution to AI training corpuses, offers a glimpse into this future. The more generic, AI-like content a publisher produces, the less they get paid; the more unique and human-generated, the more valuable it becomes.
This highlights a key insight: AI companies are "hunting for net new unique data." While AI can synthesize existing information, its outputs quickly become generic and devalued. The truly valuable contribution, therefore, is fresh, unique, human-generated insight that can't be found elsewhere.
Riding the Models: Your Path to a More Ambitious Future
The core message is one of hope and empowerment. The future isn't about fearing AI; it's about embracing it. Dan's ultimate call to action is simple and profound: "If you just ride the models, you're going to be fine."
This means actively engaging with new AI tools as they emerge, learning how to integrate them into your workflow, whatever your profession. By doing so, individuals can unlock new levels of productivity, creativity, and fulfillment. AI can make an "ambitious life" more possible for more people, enabling them to do more, better, and more meaningful work than ever before. While it's possible to opt out, those who choose to "ride the models" will find themselves at the forefront of this evolving landscape, equipped to thrive.
The Craft of Ideas: Writing an 8,000-Word Thesis with AI
Even the process of articulating these complex ideas benefited from AI. Dan's journey to write his 8,000-word piece, "After Automation," was a grueling intellectual marathon. He described the difficulty of longer pieces, where "if you change something here, it changes four other things over here," making them exponentially harder than shorter articles. The challenge was to articulate an underlying feeling—a "ground truth" observed daily—that couldn't quite be cleanly put into words.
AI became an indispensable partner in this deep thinking process:
- Monologuing for Clarity: Each morning, Dan would monologue into a document, explaining the piece's argument front-to-back. This created a log of his evolving thoughts.
- Claude for Conceptual Refinement: He then used Claude, an AI model better suited for conceptual thinking, to analyze his monologues and help him "figure out what I'm trying to say." Claude's responses would often spark breakthroughs, bringing him closer to the core articulation.
- Codex for Auditory Review: As the draft grew, Dan would use Codex to turn the latest version into a podcast. Listening to his own arguments during his commute allowed him to identify structural flaws, awkward phrasing, and areas needing improvement from a fresh, auditory perspective—a completely impossible feat without AI.
This personal anecdote underscores the very point of the article: AI isn't replacing human creativity or deep thought, but augmenting it. It's a powerful tool that enables humans to push their own boundaries, refine their ideas, and achieve intellectual feats that would be far more challenging, if not impossible, on their own.
In conclusion, the future of work isn't a zero-sum game between humans and AI. It's a collaborative dance where AI automates the predictable, making "yesterday's expert competence cheap," while simultaneously elevating the demand for uniquely human skills: direction, refinement, system-building, and the continuous definition of what truly matters. So, as Dan confidently asserts, "If you ride the models, you're going to be okay. You're going to have a job. You're going to do great work. And you don't have to worry."