Mastering Human-Centric Framing and Review in AI-Automated Work

The AI Sandwich: Mastering Your Role in an Automated World

This conversation with Kieran Klaassen, GM of Cora and creator of Compound Engineering, reveals a critical, non-obvious implication for the future of work: the automation of the "middle" of many processes does not signal the end of human value, but rather a strategic re-focusing. The true advantage lies not in trying to automate everything, but in mastering the human-centric "bread" of the AI sandwich -- the framing of problems and the critical review of AI-generated outputs. This insight is crucial for engineers, product managers, designers, and knowledge workers alike, offering them a roadmap to not just survive, but thrive, by leaning into uniquely human skills that AI cannot replicate. By understanding when to delegate to AI and when to apply human judgment, individuals can unlock greater productivity and find deeper satisfaction in their work.

The Illusion of Full Automation: Where AI Shines and Humans Must Lead

The prevailing narrative around AI often centers on its relentless march toward full automation, creating anxiety about job displacement. However, Kieran Klaassen, through the lens of his Compound Engineering framework, offers a more nuanced perspective: the "work" phase of many AI-assisted processes is, in essence, already solved. Large Language Models (LLMs) excel at executing defined steps, performing deep work over extended periods, and generating outputs based on clear instructions. This capability, while powerful, shifts the human role from direct execution to higher-level strategic engagement. The real bottleneck, and therefore the locus of human value, lies at the beginning and the end of the process.

The Compound Engineering framework, initially developed for AI-driven software development, breaks down work into four key stages: Plan, Work, Review, and Compound. Klaassen highlights that the "Work" phase, where an AI agent performs the core tasks, is remarkably effective when guided by a solid plan. The "Review" phase, traditionally human-led code reviews or quality assurance, is also increasingly supported by AI, refining outputs and catching errors. However, the true strategic advantage emerges from understanding when human intervention is indispensable.

"The beginning and the end the middle can be automated pretty well and driven at some point said oh it's kind of like a sandwich which was like very funny"

This "AI Sandwich" metaphor, coined by Trevin Chow, positions humans as the bread, framing the problem at the start and reviewing the final product. Klaassen argues that attempting to automate these human-centric phases is a fundamental misunderstanding of AI's current capabilities and limitations. The critical insight here is that while AI can execute tasks, it struggles with the higher-order cognitive functions of setting the problem frame and applying nuanced judgment to the output. This distinction is not merely semantic; it represents a significant competitive advantage for those who master it. By offloading the rote "middle" work to AI, humans can dedicate more cognitive energy to the moments where their unique skills are most impactful.

The Art of Framing: Why Humans Hold the Key to Problem Definition

The "Plan" phase, and even more so the preceding "Brainstorm" and "Ideate" steps introduced by Trevin Chow, are where human intelligence remains paramount. Klaassen emphasizes that while LLMs can support ideation, the crucial act of defining the problem's frame--setting its boundaries and context--is a human endeavor. This involves understanding the broader landscape, considering different perspectives, and articulating the core challenge in a way that AI can effectively address.

Klaassen uses an analogy: consider the problem "my knee hurts." An AI might suggest Advil (the immediate solution). However, a human, operating at a higher frame, might recognize the need for stretching, a change in running habits, or even a deeper medical investigation. This ability to shift between frames, to see the problem from multiple angles and at different levels of abstraction, is something LLMs currently struggle with. They operate efficiently within a given frame but lack the intuitive leap to redefine or expand it.

"Humans are very good at flipping and changing frames like that and our job is to set the frame or set the bounds within which we solve the problem and um i think it's very it's going to be very very hard for agents to do that well by themselves"

This human capacity for frame-setting is not just about problem-solving; it's about ownership and authenticity. As Klaassen notes, if you want a creation to be "your own," it needs to be connected to you. This connection is forged in the initial framing and the final review, moments where human intent and judgment are directly applied. The implication for professionals is clear: developing strong framing skills--the ability to ask the right questions, define scope, and set strategic direction--will become increasingly valuable. This is where delayed payoffs manifest; mastering framing now builds a durable advantage that AI cannot easily replicate.

The Polish and the Persona: Elevating AI Outputs with Human Taste

The "Review" phase, while increasingly AI-assisted, still requires a human touch, particularly for achieving true excellence. Klaassen draws a parallel to his background in classical composition. While AI can generate music, it struggles to capture the nuance of a live performance, the emotional resonance of a melody, or the subjective "feel" of a piece. This is where human "taste" and "experience" become critical differentiators.

The ability to critique AI-generated output not just for correctness but for aesthetic quality, emotional impact, and strategic alignment is a uniquely human skill. This "polish" step, as Klaassen describes it, is where work transcends being merely functional to becoming artful. It involves making something "feel great," a subjective evaluation that AI, despite its analytical power, cannot fully replicate. The difference between a good solution and a truly exceptional one often lies in this human-driven refinement.

"The beauty comes in when a human looks at it clicks around and has a feel like ah this is this this doesn't feel good we can polish it even more we can make it even better"

This emphasis on polish and subjective evaluation is where delayed payoffs are most evident. While AI can churn out functional outputs rapidly, the iterative process of refinement, guided by human taste, leads to superior quality and greater user satisfaction over time. This also means that conventional wisdom, which might push for maximum speed and output volume, fails when extended forward. The focus shifts from simply "solving" a problem to "solving it beautifully and effectively," a distinction that requires human judgment. For individuals, leaning into this "polish" phase, developing a discerning eye and an appreciation for craft, offers a pathway to creating work that is both personally fulfilling and competitively superior. It's about finding the parts of work that bring joy and leveraging AI to amplify them.

Key Action Items

• Immediate Action (Next 1-2 Weeks):

  • Identify one recurring task in your workflow that involves step-by-step execution. Experiment with delegating the "work" phase of this task entirely to an AI agent, focusing your effort on crafting a precise plan and a thorough review.
  • Practice articulating the "frame" of a problem you're currently working on. Write down the boundaries, assumptions, and desired outcomes in clear, concise language.
  • Consciously engage in the "polish" phase of a recent AI-assisted output. Go beyond checking for errors and actively seek ways to improve its aesthetic, clarity, or impact.

• Short-Term Investment (Next 1-3 Months):

  • Explore AI tools that support ideation and brainstorming. Use them to generate a wide range of initial ideas, then dedicate significant human effort to selecting, refining, and framing the most promising ones.
  • Seek out feedback on the "feel" or subjective quality of your work, not just its functional correctness. This could involve asking colleagues for their impressions or user testing.
  • Actively look for opportunities to apply your unique "taste" or judgment to AI-generated content. This might involve refining copy, adjusting design elements, or curating information.

• Long-Term Investment (6-18 Months):

  • Develop expertise in a specific domain where nuanced judgment and subjective evaluation are critical. This will position you to guide AI more effectively in complex, artful tasks.
  • Focus on building skills that involve setting strategic frames and understanding complex systems, rather than just executing tasks within predefined frames.
  • Cultivate an appreciation for what brings you joy and energy in your work, and explore how AI can be leveraged to amplify these fulfilling aspects, creating a sustainable and enjoyable career path. This requires patience and a willingness to adapt as AI capabilities evolve.