Discover why AI automation paradoxically creates more human work. Explore predictions on future work, new job roles, and how to thrive in the AI-driven future.
The AI Paradox: Why Automation Means More Work, Not Less
The conversation between Dan Shipper, CEO of Every, and podcast host Lenny reveals a counterintuitive truth about artificial intelligence and the future of work: despite unprecedented automation capabilities, humans are working more than ever. This paradox challenges the widely held belief that AI will eliminate jobs and render human workers obsolete. Instead, the evidence suggests a more nuanced reality where AI amplifies human capability while simultaneously increasing the scope and complexity of work.
Key Insights
- The AI Automation Paradox: Every automated system requires human oversight and management, meaning automation actually increases human workload rather than eliminating it
- Bifurcated Work Future: Professional work will split into two primary modes—agent-based task delegation through Slack and desktop-based work within AI environments like Codex and Claude Code
- The Rise of Forward-Deployed Engineers: A new essential role emerges: engineers dedicated to building and maintaining AI agent systems, ensuring they function correctly across organizations
- SaaS Is Not Dying: Contrary to predictions, AI adoption increases SaaS usage rather than eliminating it, as agents become power users of software tools
- Skill-Focused Job Security: The jobs that remain are those where humans add creative value beyond commoditized AI capabilities, requiring specialization and novel thinking
- Product Managers and Designers Thrive: These roles are experiencing explosive growth as AI democratizes technical capabilities, allowing PMs and designers to ship products independently
The Automation Paradox: More AI, More Human Work
The central paradox of AI adoption becomes apparent when examining real-world implementation. Dan Shipper's experience at Every, a nearly 30-person company of AI early adopters, demonstrates this principle clearly. The company doubled in size over a year despite having sophisticated AI agents handling significant portions of work. This counterintuitive growth pattern challenges the automation-equals-job-loss narrative.
The reason is straightforward: automation creates new layers of human responsibility. Every agent requires a human guardian—someone who understands its capabilities, limitations, and failures. A data science team at a leading tech company discovered this when their colleagues began using AI to perform analysis. Instead of having fewer data scientists, the team found themselves managing an influx of requests for validation, correction, and quality assessment. The AI generated outputs faster than humans could review them, creating a new job category: output validation and correction.
This isn't a flaw in AI design but rather an inherent characteristic of any powerful automation system. Human managers oversee automated manufacturing lines. Human pilots supervise autopilot systems. The same principle applies to knowledge work. The benchmark systems that measure AI progress often obscure this reality because they measure task completion in isolation, not the full context of organizational deployment and error management.
When GPT-5.5 receives a score of 60 out of 100 on complex code rewrites, compared to senior engineers scoring 80-90 out of 100, the gap appears manageable. However, implementing that AI-generated code requires human engineers to review it, test it, understand the architectural decisions, and catch subtle bugs. The automation hasn't eliminated the senior engineer; it's created a new workflow where senior engineers spend more time evaluating and refining AI output than they would have spent writing code from scratch.
The paradox extends to all knowledge work. Writers now spend time editing AI-generated content. Product managers review AI-suggested feature specifications. Marketing teams validate AI-created campaign materials. Each layer of automation introduces a new layer of human oversight. The net result: more work, not less.
How Work Will Transform: Two Modes of AI Collaboration
The future of work doesn't involve AI replacing workers; instead, it involves a fundamental restructuring of how work happens. Shipper predicts two primary modes of AI-human collaboration will dominate professional environments within the next year.
Mode One: The Super Agent Through Slack
The first transformation involves centralized AI agents that team members interact with primarily through Slack. Rather than individual personal agents (the initial assumption), current technology favors a single, well-maintained company agent. This agent serves as a delegated worker, handling routine tasks, answering questions, and executing predetermined workflows.
The reason for this shift toward super-agents rather than personal agents is critical: AI agents require continuous human care and oversight. A personal agent operating independently quickly degrades in quality, breaking down on complex tasks requiring SSH access, database queries, or nuanced decision-making. Without dedicated human attention, agents become unreliable.
Companies like Shopify and Ramp have already implemented this model successfully. A forward-deployed engineer manages the company agent, ensuring it remains reliable and effective. This individual acts as the bridge between the organization's needs and the agent's capabilities, constantly refining prompts, handling edge cases, and preventing the agent from making costly errors.
The advantage of this approach is scalability through quality. One well-maintained super-agent, managed by a skilled engineer, serves an entire organization more effectively than dozens of poorly-maintained personal agents. As AI models improve and require less hands-on maintenance, the system may evolve to include team-level or personal agents, but that evolution depends on meaningful progress in AI autonomy.
Mode Two: Desktop-Based Work Environment (Codex, Claude Code)
The second transformation is more profound: the desktop becomes the primary work environment, not a browser. Tools like OpenAI's Codex and Anthropic's Claude Code function as operating systems for professional work. All tasks—from writing documents to managing emails to analyzing data—happen within these environments.
This shift occurred because of a critical technical realization: when an AI agent runs on your personal computer, it has access to everything you can access. It can use the terminal, browse websites, read your files, and interact with any application. Anthropic discovered that Claude Code, initially positioned as a coding tool, became incredibly powerful for non-technical work when combined with browser capabilities.
Dan Shipper's workflow exemplifies this transformation. He maintains one Codex thread per project. Within each thread, he opens an in-app browser to access his document editor (Proof), email client, and analytics dashboards. Codex runs continuously, watching his work. When he has a task—"Research this topic," "Draft this email," "Gather these documents and create a report"—Codex executes it while maintaining visibility into his work. He remains in control while delegating execution.
This represents a paradigm shift from traditional AI assistance. Rather than integrating AI into every SaaS tool, the future involves running SaaS tools within AI environments. This changes pricing models, user experience, and software architecture fundamentally.
The SaaS Revolution Nobody Expected
The predictions about SaaS's demise due to AI proved incorrect, but for surprising reasons. Shipper is explicitly bullish on SaaS stocks, predicting their significant growth over the coming years. This optimism stems from a counterintuitive understanding: AI adoption increases SaaS usage rather than reducing it.
Here's why: when personal agents operate within Codex or Claude Code, they become power users of existing SaaS tools. An agent using Slack, Google Workspace, Salesforce, or PostHog generates far more interactions, queries, and transactions than a human user. A single human might query a data analytics tool once daily; an agent querying it dozens of times per hour becomes the norm.
This shift transforms SaaS economics. The traditional model involved vendors paying for tokens or computation as users accessed AI features. The new model involves users bringing their own AI (running locally or through their subscription with OpenAI/Anthropic), accessing SaaS tools directly. This removes the token cost burden from SaaS companies while simultaneously increasing their infrastructure demands.
The architectural implications matter too. SaaS tools must become agent-friendly. Clear APIs, consistent user interface patterns, immediate feedback mechanisms, and robust logging become essential. A tool designed for human users—with multiple steps, modal dialogs, and delayed feedback—frustrates agent automation. A tool designed for agent collaboration requires different thinking.
Shipper noticed this with Proof, his Markdown editor. By removing unnecessary formatting options that agents could handle, the tool became simpler. By adding approval systems and transaction logs, the tool became agent-ready. Users still prefer the tool because it now supports agent collaboration.
The economic outcome: SaaS companies maintain or improve margins by eliminating expensive AI features while benefiting from massive increases in usage volume. A SaaS tool used by 100 human users and 10,000 agent instances generates substantially more value than the same tool used by 1,100 human users.
The New Essential Role: Forward-Deployed Engineers
As organizations implement AI agents, a new essential professional role emerges with unexpected demand: the forward-deployed engineer. These individuals are responsible for ensuring AI agents function correctly, adapt to changing requirements, and maintain value delivery across an organization.
The role differs fundamentally from traditional software engineering. Rather than building products, forward-deployed engineers build systems that enable others to use products effectively. At Every, several engineers spend their days in conversation with agents, identifying failures, providing guidance, and constantly refining the systems that manage automated workflows.
This role requires specific personality traits. The forward-deployed engineer must enjoy hands-on work with cutting-edge technology. They must be comfortable with ambiguity, as agent behavior often surprises even their creators. They must combine software engineering expertise with systems thinking, understanding both the technical constraints and organizational workflows.
The forward-deployed engineer position isn't temporary. Even as AI models improve and agents become more autonomous, these roles will expand rather than shrink. Why? Because each new capability introduces new possibilities, new failure modes, and new optimization opportunities. The role evolves, but the need for human care of agents remains fundamental.
Companies are beginning to understand this necessity. Major organizations are hiring or promoting people into forward-deployed roles, recognizing that agent systems require dedicated management. This represents genuine job creation—not displacement—driven by AI adoption.
The Shape of Changing Work: New Responsibilities and Opportunities
As AI adoption spreads, the shape of professional work transforms in ways beyond simple acceleration. Several distinct shifts become apparent.
Pull Request Proliferation
The number of pull requests, code changes, and product modifications skyrockets when non-technical people gain agency through AI. An operations specialist can now make database changes. A marketing team member can modify the product interface. An editor can rewrite sections of code.
This creates a management challenge on the opposite end: reviewing, testing, and integrating the flood of contributions. GitHub, which powers Open Source software development, has experienced exponential load increases as agents submit thousands of pull requests daily, all requiring review.
The solution isn't to prevent contributions but to build systems that enable evaluation and integration of this volume. Companies must ask new questions: Which changes align with our product vision? Which should we merge? What should we delete to maintain coherence? Anthropic handles this by maintaining strict design principles and actively deleting features that don't align with their vision, preventing bloat despite massive contribution volume.
Emergence of Generalist Roles
With everyone capable of doing technical work, professional identity becomes confusing. A marketer who designs a feature isn't necessarily becoming an engineer; they're expanding what marketing means. A product manager who writes production code isn't abandoning product management; they're extending their capability.
This creates temporary role confusion as organizations adapt, but ultimately settles into a new understanding where specialization remains valuable despite wider capability distribution. A great marketer who can code brings both marketing insight and technical execution to their role, creating compounding advantage.
The Writing Paradox: AI-Generated Content Becomes Valuable
Perhaps counterintuitively, AI content generation doesn't devalue human judgment—it amplifies it. Shipper's experience at Every demonstrates this: the best outcomes emerge from AI co-authoring with human editorial oversight, not pure human writing and not pure AI generation.
The key distinction separates slop from quality AI-generated content. Slop is content generated quickly without substantive human involvement. It lacks coherence, accuracy, and alignment with the writer's actual perspective. Quality AI-generated content is authored by a human working alongside AI, where the human makes meaningful decisions about direction, accuracy, and voice.
This has practical implications. Strategy documents that might take a human hours to write can be drafted by AI in minutes, with the human then spending time refining rather than creating. Email that humans procrastinate on gets drafted by AI, with the human ensuring tone and specifics are correct. The human remains the creative force but uses AI for execution.
The shift toward acceptance of AI-generated content represents a pragmatic recognition that human time is better spent on thinking and judgment than on transcription and formatting.
Who Succeeds: Product Managers and Designers Rising
The future belongs to specific professional roles, particularly product managers and full-stack designers. These aren't predictions based on theory; they're observations from a company deeply embedded in AI adoption.
Product Managers Unleashed
Product managers traditionally faced friction in implementation. They would write specifications; engineers would implement them based on their own interpretation. Misalignment was common; iteration was slow.
Now, skilled PMs who learn to use AI coding tools become dramatically more powerful. They can author their own implementations, see their designs come to life at their pace, and maintain complete control over the user experience. This doesn't eliminate engineers; it changes the problem they solve.
Rather than building common features, engineers focus on architecture, performance, and complex systems that require specialized expertise. Meanwhile, PMs ship features independently, accelerating the feedback loop between design and market reality.
Shipper points to Marcus, Every's PM for their writing product, as an example. With deep product expertise, solid engineering intuition, and fluency in AI coding tools, Marcus ships features faster than almost anyone on the team. He understands users deeply, speaks to them constantly, and translates their feedback directly into code. This capability makes him extraordinarily valuable.
Companies should expect strong PM demand to increase significantly as organizations recognize how much faster product development moves with AI-fluent product managers.
Designers Becoming Builders
Designers have always faced implementation friction. They create beautiful mockups; engineers implement interpretations. Perfect design rarely makes it into production due to effort constraints or differing priorities.
AI tools like Codex eliminate this friction for designers who learn to build. A designer who can implement their ideas directly becomes exponentially more productive. They're not constrained by engineering capacity or the translator role of specification documents.
Equally important: design sensibility remains scarce. As everyone can ship AI-generated interfaces, the ability to create visually coherent, delightful interactions becomes more valuable, not less. The tools democratize execution but can't replicate the judgment about what's beautiful or intuitive.
Designers who master AI tools become micro-entrepreneurs within organizations, shipping products at their own pace. Some will remain in companies, creating tremendous value; others will recognize they can build independently and start their own ventures.
Practical Strategies: How to Thrive in the AI-Transformed Workplace
For individuals concerned about job security or wanting to maximize opportunity in this transition, Shipper's advice is elegant: "Ride the models."
This means adopting new models immediately when they're released, understanding their capabilities through hands-on experimentation, and consistently finding ways to apply them to work that matters to you. It's not about fear or job security; it's about expansion.
When GPT-5.5 releases, try it. When new versions of Codex emerge, experiment. When new agents become available, test them on problems you care about. This consistent engagement creates compounding advantage over time.
Practically, this means:
For Technical Workers: Master tools like Codex, Claude Code, and Cursor. Understand how to use agents effectively. Build systems that leverage agents while maintaining quality. Become a forward-deployed engineer who understands both AI capabilities and organizational needs.
For Non-Technical Workers: Start using AI tools for your specific work. If you're a designer, learn to code with AI assistance. If you're a PM, get comfortable giving technical direction to agents and understanding code. If you're an analyst, experiment with having agents perform analysis while you focus on interpretation.
For All Workers: Make tools agent-friendly if you build them. Understand that most professional software will soon be used by both humans and agents simultaneously. Design accordingly.
Across Organizations: Build systems that enable agents while maintaining quality. Implement review processes for agent-generated work. Create feedback loops where agents learn from corrections. Hire forward-deployed engineers who care about agent system quality.
The common thread: engagement with current tools matters more than fear about future displacement. Those who actively experiment with AI maintain the flexibility to adapt as capabilities evolve.
The Reality of Job Transformation, Not Elimination
The "AI jobs apocalypse" narrative, while compelling and useful for some rhetorical purposes, doesn't match observable reality. Every, a company where everyone actively uses the latest AI tools and is deeply embedded in the emerging future of work, doubled headcount in a year. This isn't an anomaly; it reflects broader market dynamics.
What's actually happening is restructuring. Some job categories decline while others expand dramatically. Middle management roles, administrative positions focused on routine tasks, and roles built entirely around coordination may shrink as agents handle coordination work more efficiently.
Simultaneously, demand spikes for roles that involve judgment, creativity, oversight, and synthesis. The forward-deployed engineer role that didn't exist five years ago is becoming essential. PM and designer roles, which were expected to shrink, are expanding.
The crucial distinction: you don't avoid job displacement by hiding from AI. You avoid it by becoming part of the future, learning to work alongside AI systems, and understanding how your role transforms rather than vanishes.
Shipper notes that benchmark progress suggests significant capability increases are coming. Yet benchmarks measure isolated tasks, not the full context of organizational work. They don't capture the human judgment required to stitch together AI outputs into coherent strategy. They don't measure creativity or the ability to recognize which problems matter.
This is why he's simultaneously "extremely AI-pilled and very bullish on humans." The AI revolution isn't displacing humans; it's transforming human work in ways both challenging and opportunity-rich.
Conclusion
The AI paradox—more automation creating more human work—reveals a deeper truth about how intelligent systems function within organizations. Automation amplifies human capability while simultaneously creating new categories of human responsibility. The future of work involves not AI replacing humans but humans and AI collaborating in newly structured relationships.
Success in this future requires active engagement rather than passive concern. Those who experiment with emerging tools, stay curious about new models, and adapt their workflows gain exponential advantage. Those who resist change or hope automation will somehow go away will find themselves increasingly irrelevant.
The most exciting aspect of this transition is its accessibility. Advanced AI systems are available to anyone with a computer and an internet connection, not reserved for large corporations or technical experts. This democratization means individuals worldwide have opportunity to participate in defining how AI is used, discovering applications their creators never imagined, and building value that wouldn't exist otherwise.
The horizon doesn't bring dragons or utopia—it brings complexity, opportunity, and profound change. Riding the models means learning to navigate that complexity with curiosity and adaptability, ensuring you remain part of the future rather than displaced by it.
Original source: The AI paradox: More automation, more humans, more work | Dan Shipper
powered by osmu.app