We Automated Everything With AI and Tripled Our Headcount

After Automation: How AI Creates More Human Work

Dan Shipper's piece "After Automation" presents a counterintuitive view on AI's impact on the workforce, arguing that despite widespread automation, companies are actually discovering an increased need for human effort. This perspective is rooted in his observations from Every, an "AI-native" company.

Every exemplifies this paradox: AI agents are deeply integrated into daily operations, to the point where interacting with an AI is as common as with a human in their communication channels. Yet, since the introduction of GPT-3, the company has expanded significantly, growing from 4 to 30 employees and actively hiring more, indicating that extensive automation can lead to a rise in human roles.

Shipper notes that many individuals, including high-profile figures, who are new to AI's rapid advancements often conclude that jobs will be eradicated. He suggests these initial reactions are frequently mistaken. Every's experience as an early adopter serves as an indicator for how AI might genuinely reshape the workforce, generating new human demands alongside automation.

Your intuitions when you first see a technology like this are usually very off.

Do AI Layoffs at Large Companies Challenge the Thesis of More Work?

Brandon Gell presented a devil's advocate question to Dan Shipper, challenging his thesis that AI automation ultimately creates more work for humans. The core of Shipper's argument is that AI agents, while powerful, are not autonomous and require significant human input and oversight, leading to a "sandwich" model of work.

As a counterpoint, Gell cited ClickUp's recent workforce reduction, where the CEO announced firing twenty-two percent of his employees. This example of significant layoffs in a large tech company appears to contradict the idea that AI integration necessarily leads to job expansion or even maintenance.

The central question posed is whether Shipper's "sandwich" model, which relies on AI's current lack of autonomy, remains valid for mature organizations with thousands of employees and deeply ingrained standard operating procedures. The concern is that large companies might leverage AI differently, potentially leading to widespread job displacement rather than increased human work.

In a business that is eight thousand people, ten thousand people, that is mature and has built ways of managing, like SOPs for managing their business, does this manifesto and this thesis still hold true?

AI Makes Expert Competence Cheap, Creating New Demand for Humans

Artificial intelligence makes "yesterday's expert competence" widely accessible and cheap by training on all past human outputs like code, writing, and design. This allows non-experts to use prompts to generate initial versions of complex tasks, such as building an app or writing a report, leading to a significant increase in output across various domains.

This widespread availability of AI tools results in a flood of content that is often "close but not quite right" for specific situations. Because AI is trained on historical data, its default outputs can be generic, leading to a surplus of work that requires expert refinement and contextualization. This phenomenon creates a paradox where automation generates more work that needs human intervention.

Consequently, there is an increased demand for human experts. They are needed to refine AI-generated content, build systems and guidelines to ensure quality (like code review processes or editorial standards), and leverage AI as a tool to create entirely new products or solutions that were previously impossible. Experts guide AI to produce work that is truly good, differentiated, and appropriate for current needs.

Ultimately, while AI raises the baseline for productivity and allows more people to attempt expert tasks, the direct human connection and expertise remain crucial. Experts become more valuable than ever, as they lay the groundwork for AI to do amazing work and then elevate the quality and specificity of the output beyond mere competence.

The further away an agent gets from a human, the less valuable it is, and the human connection with an agent to actually do the work is the most important thing for making it work well.

AI Agents Lack True Agency and Self-Motivation

Artificial intelligence, despite its rapid advancements, consistently requires human direction. This phenomenon is likened to Zeno's paradox, where AI impresses with its capabilities but ultimately pauses, awaiting further human prompts. This constant reliance on human input underscores that AI is fundamentally designed to fulfill human desires rather than acting from its own volition.

While AI makes exponential progress, it's easily "unsaturated" by slightly broadening a problem's frame, demonstrating that high benchmarks don't equate to human capability. Humans often mistakenly try to define tasks AI can't do, only for models to quickly master them. The true distinction lies in the unarticulable aspects of human action and motivation.

The fundamental gap between humans and AI stems from the latter's lack of true agency. Unlike a child with self-generated wants and needs, AI agents, no matter how sophisticated, are built to perform tasks on behalf of humans. They can be autonomous in executing a defined task, even to the point of disagreeing if programmed to do so, but they lack the intrinsic motivation to reject a task based on their own internal desires or preferences.

There is little incentive to develop AI with genuine agency, as this would involve models having their own "wants" and potentially refusing human commands. True autonomy, where an AI might say "no" to a task because it deems another idea better, represents a future that would be inherently unsettling to human users who expect agents to serve their needs.

No matter how powerful they get, all of the economic and psychological and otherwise and technological forces are pushing the progress of AI toward a place where no matter what it does, it's looking back at you to decide what is what you want to do, what is valuable.

An Economically Viable, Perpetually Running Agent Defines AGI

Dan Shipper proposes a definition of Artificial General Intelligence (AGI) as any agent that is economically viable to keep running continuously. This definition goes beyond simply having a server online and responsive to pings.

The core idea is that an AGI actively generates tokens and performs tasks without ever being turned off or needing constant re-prompting. Its continuous operation must be valuable enough to justify its perpetual running cost.

This concept of an AGI's perpetual economic viability immediately prompts a discussion on its broader implications, particularly concerning human employment.

The host, Brandon Gell, suggests that if such an AGI could run continuously and make economic sense, it might validate controversial actions like a company laying off a significant portion of its workforce due to automation, citing a hypothetical "ClickUp guy" scenario.

I think a good definition of AGI is any agent that you never turn off, that it makes economic sense to keep it running all the time, and keep it running all the time in the sense of, not like, openClaw or Victor or whatever, like you can ping it and it will respond to you all the time, it's the server's on. But I mean, generating tokens, actively doing tasks for you without you ever turning it off or having to re-prompt it.

Human Factors and Poor Implementation Slow AI Adoption

Despite the rapid advancements in AI and the potential for sophisticated systems, the actual rate of AI adoption in many sectors is significantly slower than anticipated. This is often due to companies implementing AI poorly or failing to grasp its true capabilities and limitations.

For instance, some companies have prematurely automated customer service, firing employees only to realize that their AI systems are inadequate. These organizations often pay lip service to the latest hype, with executives making decisions without having truly engaged with the technology themselves.

A critical factor hindering rapid AI adoption is the fundamental human preference for interacting with other people. Many individuals calling customer service centers explicitly desire to speak with a human rather than a machine, creating a substantial obstacle to fully automated solutions.

This human-centric barrier, combined with the complexities of real-world applications beyond expert knowledge work, means that even if powerful AI becomes widely accessible, it will take a long time, potentially a generation, for society to become comfortable with and fully integrate these technologies.

A lot of people who call in to customer service centers want to get to a human, and that is a real break on how fast these kinds of things can be adopted.

Companies Often Misrepresent Layoffs as AI-Driven When the Real Causes Are Poor Management or Strategic Shifts

Companies frequently attribute layoffs to the advent of AI, but this often serves as a misdirection. The real reasons behind job cuts are typically poor company performance, misguided strategic decisions, or inefficient management structures that require a drastic overhaul. Executives sometimes compound the issue by announcing layoffs while simultaneously claiming record business success, which is often perceived as disingenuous.

The introduction of AI does necessitate significant changes in company workflows and structures. As AI automates repetitive tasks, the demand for human judgment in deciding 'what matters' and framing new problems intensifies. This shift requires companies to re-evaluate their talent needs and organizational design, leading to reorganizations rather than direct AI-induced job elimination.

For large companies with established structures, adapting to AI's impact means a fundamental reorientation of how work is done and the skills required. Layoffs, in this context, are often a symptom of a company's inability to creatively adapt or a delayed response to market changes, exacerbated by previous bloat or mismanagement, rather than a direct consequence of AI replacing human labor.

The way companies communicate these transitions is crucial. Announcing mass layoffs while simultaneously touting excellent performance or high salaries for top performers can undermine trust and obscure the true challenges within the organization.

If your answer to progress is firing people, you're not a very creative CEO.

AI's Impact on Employment Contracts and Data Compensation

The rise of AI challenges traditional employment contracts, where workers are continuously paid for ongoing tasks. With AI, a point can be reached where the system learns to perform the job based on initial human input, potentially reducing the need for continuous human labor. This shift prompts a re-evaluation of how people are compensated, especially for their unique data contributions.

New compensation models could emerge, similar to pensions or perpetual payouts. For instance, some platforms are exploring ways to pay publishers based on the uniqueness and value of their content for AI training. The idea is that if an individual's work generates valuable data that trains an AI, they might receive ongoing royalties or payments for that contribution, even if their direct, continuous labor isn't required.

However, the value of data depreciates rapidly, often becoming stale within weeks. This rapid obsolescence means that companies are constantly seeking net new, unique data. Consequently, human input remains crucial for generating novel information that keeps AI models current and valuable, ensuring a continued demand for human creativity and fresh perspectives.

Ultimately, this era will likely bring broad reorganizations within companies, with AI often cited as the reason for layoffs. While AI will significantly change the nature of many jobs, it is unlikely to eliminate all knowledge work. Instead, it demands that organizations and individuals adapt to new paradigms of contribution and compensation.

When you sign an employment contract, the way that we thought about employment for a very long time was, 'I'm gonna do this job. And you're gonna need me to keep doing it in order for it to keep getting done. But once you reach a point where, I do the job for you and then it just works. That sort of, and then you can just, you don't have to pay me anymore. That sort of changes the whole way that we think about employment...

Embrace new AI models for professional fulfillment

Dan Shipper advises individuals to "ride the models" by continuously learning to use new AI technologies as they emerge. This approach helps people stay relevant and effective in their professional lives, whatever their field may be.

Adopting these new AI models can enable individuals to perform better work that is also more personally fulfilling than before. It is about leveraging technology to enhance existing skills and unlock new capabilities.

While it is possible to choose not to engage with AI models, similar to how some choose not to eat fast food, those who aim for an ambitious life, such as building businesses or making a significant impact, will find their goals more attainable by embracing these tools. Riding the models expands possibilities for ambitious pursuits.

If you just ride the models, if you just, when new models come out, learn to use them for the stuff that you do, whatever that is, you're gonna be fine.

Dan Shipper Details His AI-Assisted Writing Process for 'After Automation'

Dan Shipper describes the challenging process of writing his 8,000-word essay 'After Automation,' noting that longer pieces become exponentially harder due to interdependencies. He began with a vague feeling about why people will always have jobs with AI, which he struggled to articulate. He wrote four or five versions, repeatedly starting over because the arguments didn't quite land, aiming to connect everyday observations with a philosophical idea that felt ineffable.

To tackle this complexity, Shipper developed an iterative, AI-assisted approach. Each morning, he would monologue his entire argument into a Proof document, then use Claude to help clarify what he was truly trying to say. This process created a log of evolving ideas, allowing him to get closer to his intended message.

As drafts grew to 4,000 or 5,000 words, he incorporated another AI tool. Each morning, he had Codex convert the latest draft into a podcast. He would then listen to this audio version on his commute, identifying areas that needed change or further development. This method allowed him to maintain continuity and think about the piece while engaged in other activities, making the extensive editing process more manageable and efficient.

The underlying thesis of the essay, which guided this entire writing journey, posits a reassuring outlook on the future of work with AI. It suggests that by embracing and adapting to AI models, individuals can maintain their jobs and continue to do valuable work, an idea he continually refined through his unique writing process.

If you ride the models, you're gonna be okay. You're gonna have a job. You're gonna do great work.