Every CEO Dan Shipper recently offered a sharp way to think about AI and work. As models become better at summarizing, drafting, coding, researching, scheduling, and coordinating, the middle of many knowledge workflows starts to look increasingly machine shaped. The work that remains most human gathers at two ends.

At the first end sits intent. Someone must decide what matters, what question deserves attention, what standard counts as good, and which tradeoffs are acceptable. AI can produce options at speed, but the first valuable act is choosing the problem with enough taste and context that the output has somewhere meaningful to go.

At the second end sits accountability. Someone must stand behind the result, explain it to other people, notice when it feels wrong, and carry the consequences when the neat answer fails in the real world. This is where trust, ethics, customer empathy, craft judgment, and organizational memory still matter.

The brutal part of the metaphor is what happens to the middle. A large share of professional work has lived in translation between intent and result. Turn meeting notes into a plan. Turn a plan into copy. Turn a chart into a memo. Turn a bug report into a patch. Turn research into a deck. These tasks once proved competence because they consumed attention. Now they are becoming the natural territory of agents.

Shipper called this shift the allocation economy. The worker becomes less like a maker of every sentence and more like a manager of models, tools, and review loops. Even junior employees may be expected to brief an AI system, compare outputs, refine the brief, and decide when the answer is ready. The skill is no longer only knowledge. It is allocation, taste, sequencing, and review.

His two slice team idea pushes the same thought into company design. If one person with agents can ship what used to require several people, the unit of execution shrinks. The question for a team becomes less about headcount and more about clarity. What should this tiny team own. Which decisions can it make. Which parts need human help from design, growth, legal, or infrastructure. Small teams gain speed only when the human at the center knows what to ask for and what to reject.

This is why the two ends are not abstract. They show up in ordinary work. A product manager uses ChatGPT to turn customer interviews into competing roadmap narratives, then chooses the one that matches strategy. A researcher uses Gemini to compare sources and surface gaps, then decides which claims deserve confidence. A student or engineer uses Miss Formula to convert a photographed equation into usable notation, then checks whether the math still means what the original problem intended.

The pattern is clear. AI is excellent at moving through the middle when the request is legible. Humans create value by making the request worth answering and by judging the answer against reality.

For workers, the practical lesson is uncomfortable but useful. Do not protect the middle simply because it feels familiar. Build strength at the ends. Practice framing problems before opening a tool. Write clearer briefs. Develop sharper taste. Learn to evaluate sources, code, arguments, formulas, images, and numbers. Keep a record of decisions so future agents inherit context rather than noise.

For leaders, the lesson is equally direct. An AI strategy that only asks people to save time will miss the deeper redesign. Teams need new rituals for delegation, review, provenance, and responsibility. They need clear permission to use agents for the middle, and clear standards for the human decisions at the edges.

The future of work may feel like a narrowing path, but it can also become a more demanding craft. The safest human role is neither busy production nor vague supervision. It is the ability to know what should happen, guide powerful tools toward it, and answer for the result when it reaches another person.