AI Did Not Kill Work. It Exposed the Part We Were Avoiding

Working thesis

AI is not mainly replacing work. It is exposing how much of modern work was already structured to avoid the hard parts: commitment, judgment, taste, accountability, and the slow conversion of effort into depth. The useful frame is not “AI took the job.” It is “AI compressed the execute layer and left people staring at the parts they were using execution to hide from.”

Best counterargument

This thesis can become too clean. Some people will lose jobs because AI automates real tasks. Some firms will use AI to cut labor. Some work really is execution-heavy, and when execution gets cheaper, the people paid mostly for execution are exposed. The danger is turning a labor shock into a tidy moral essay about commitment. The piece has to hold both truths: displacement is real in places, but the broader panic is also hiding a deeper problem with how work was already organized.

Originality check

Queries run:

• "The Three Traps of Modern Work"
• "The Three Traps" "modern work"
• "optionality" "process" "lightness" work commitment essay
• "trapped by optionality" "process" "work"
• "decide execute deliver sandwich" AI compresses execute layer
• "slow productivity" "pseudo-productivity" Cal Newport

Findings:

• The exact title “The Three Traps of Modern Work” did not surface a close match in the search results I checked.
• The ingredients are crowded. Optionality versus commitment is well covered by David Perell and others. Capability traps are established in Repenning and Sterman. AI layoff skepticism and AI-washing are current topics in Narayanan/Kapoor, SHRM, Bloomberg, NYT snippets, and adjacent commentary. Slow productivity and pseudo-productivity are Cal Newport terrain.
• The fresh part is the synthesis: AI compresses execution, which exposes the optionality/process/lightness traps because those traps were ways of avoiding the decision and delivery layers of work.

Classification: fresh synthesis.

Sources and raw signal

External sources checked:

1. Arvind Narayanan and Sayash Kapoor, “Why AI hasn’t replaced software engineers, and won’t,” AI as Normal Technology, June 10, 2026. Contributed the decide-execute-deliver sandwich and the claim that AI compresses the execute layer while decision and delivery resist automation.
2. David Perell, “Hugging the X-Axis,” perell.com. Contributed the optionality and commitment frame: a life without commitment moves horizontally without compounding.
3. Nelson P. Repenning and John D. Sterman, “Nobody Ever Gets Credit for Fixing Problems that Never Happened: Creating and Sustaining Process Improvement,” California Management Review 43, no. 4 (Summer 2001): 64–88. Contributed the capability trap: pressure to work harder crowds out improvement, making future work worse.
4. Roy Maurer, “The AI Layoffs Narrative: Real Transformation, or Scapegoat?” SHRM, May 18, 2026. Contributed the current labor framing: some AI displacement is real, but many cuts are anticipatory, financial, or AI-washed.
5. Cal Newport, Slow Productivity: The Lost Art of Accomplishment Without Burnout (New York: Portfolio, 2024). Contributed the pseudo-productivity frame: visible activity can replace meaningful accomplishment.
6. Bloomberg and NYT search-result checks on AI-washing layoffs. Used as corroborating signal, not primary citation, because extraction/access was limited.

Internal sources checked:

• ~/wiki/reading-notes/synthesis-2026-06-14.md, especially Pattern 1: The Anti-Process Rebellion and Pattern 2: The Weight/Commitment/Lightness Triad.
• Reading notes on Cate Hall, Perell, Sarris, Karlsson, Estrina, Guzey, Altman, and Anu.
• Prior cron output for second-shift-deep-writing-workshop, 2026-06-14.

Draft

The AI layoff story is too clean.

That is the first thing wrong with it. Clean stories are usually selling something. The company says AI made the cut inevitable. The investor hears margin discipline. The employee hears replacement. The public hears future shock. Everybody gets a story that flatters their existing fear.

You can see the shape of it in the corporate language around 2026 layoffs. Cisco announced record revenue and then cut four thousand jobs while talking about restructuring around AI infrastructure. The same day. The same email. The company said it was not about savings. It was about focus.¹ The explanation is not necessarily false. That is what makes it slippery. AI may be part of the story. So are capital allocation, investor appetite, old overhiring, margin pressure, and the pleasure executives get from sounding like they are already living in the future. The phrase “because of AI” can carry all of that without having to confess any of it.

But the more interesting thing about AI and work is not that machines can now do more. That part is obvious. The interesting thing is what remains after they do it.

Arvind Narayanan and Sayash Kapoor have a useful frame for this. Software work, they argue, is not just writing code. It is a decide-execute-deliver sandwich. Decide what should be built. Execute the work. Deliver something that survives contact with users, systems, bugs, integration, politics, maintenance, and accountability. AI is very good at compressing the middle. It makes execution faster. It can write code, generate drafts, summarize documents, build prototypes, and push a half-formed idea into visible shape.²

That sounds like the whole job only if you were already pretending the middle was the whole job.

Most professional panic right now comes from that confusion. For years, a lot of knowledge work let people hide inside execution. Not because people are lazy. Because execution is legible. It produces artifacts. Decks, tickets, drafts, meetings, pull requests, roadmaps, reports, calendars full enough to feel like proof. If someone asks what you did, you can point to the thing.

Decision is harder to point at. Taste is harder to point at. Judgment is harder to point at. Commitment is almost invisible until it compounds. Accountability usually appears only when something breaks.

So modern work built a culture that over-rewarded visible motion. Then AI arrived and made visible motion cheaper. The hiding place is gone.

I know the trap here because this essay is close to it. A clean frame. A few good lines. A rhythm built to travel. That is not the same thing as weight. If AI makes fake work easier to produce, it also makes fake seriousness easier to perform. The standard cannot be whether a paragraph sounds finished. The standard has to be whether it risks being wrong. So I will be specific about where the danger is real and where the narrative is theater, even when it makes the argument less elegant and harder to share.

The first hiding place is optionality.

David Perell calls it hugging the X-axis: moving horizontally from possibility to possibility without compounding upward into depth. Optionality feels like freedom because nothing has claimed you yet. No craft has trapped you. No audience has disappointed you. No company, city, relationship, thesis, or body of work has forced you to become specific.³

There is a useful early phase where optionality is healthy. You try things. You collect samples. You learn what kind of pain you can tolerate. But optionality becomes poison when it turns into identity. Someone with infinite options often has no obligation to become anything. I have watched people spend years collecting starting points and calling it a career. I have done it myself.

AI makes this worse before it makes it better. It gives optionality an engine. You can prototype ten ideas before lunch. Generate twenty content angles. Rewrite the strategy three ways. Spin up a product mock, a landing page, a research brief, a pitch, a code scaffold, a customer persona, a logo, a plan.

That can be useful. It can also become spiritual junk food. You get the pleasure of beginning without the humiliation of committing.

The second hiding place is process.

Process is not bad. Bad process is bad. Good process protects attention, records decisions, catches mistakes, and keeps groups from eating themselves alive. The problem starts when process becomes a substitute for action. A plan can feel like courage. A framework can feel like judgment. A meeting can feel like alignment. A roadmap can feel like progress, especially before anyone has had to ship the thing.

The danger is that process artifacts are easy to generate and hard to verify. A thirty-page plan can be written in a day. It can also be complete fiction. The person who wrote it may not be the person who has to live with it. The person who approved it may not have read it. The person who executed it may have known it was wrong but needed the cover. When process becomes a language, it becomes a way to distribute responsibility without anyone owning it.

AI is brutal to process theater because it can perform the ceremony instantly. It can write the plan. It can summarize the meeting. It can create the checklist. It can make the project look administratively mature before anyone has made the hard choice.

That means process loses some of its old protective camouflage. If the process artifact can be generated in thirty seconds, the artifact cannot be the proof anymore. The proof has to move somewhere else: to the decision, the taste, the shipped work, the maintenance, the thing that changes because someone took responsibility.

This is where Repenning and Sterman’s capability trap matters. Organizations under pressure tend to work harder instead of getting better. They chase immediate output. They cut the time that would have gone into repair, learning, maintenance, and improvement. Performance gets worse, pressure rises, and the loop tightens. Nobody gets credit for fixing problems that never happened.⁴

AI can become a capability trap accelerant. If it helps a team produce more visible work without improving the system that receives that work, the team may feel more productive while becoming less capable. More tickets close, more drafts pile up, more code merges, but the surface area grows harder to maintain and the debt hides under the speed.

The third hiding place is lightness.

The internet loves light things. Posts, takes, threads, clips, swipeable opinions. Output that travels well because it does not weigh much. Lightness is not evil. Some ideas should be light. A joke should not arrive wearing a backpack. A quick observation does not need a cathedral.

But a life made entirely of light output starts to feel thin. No number of tweets becomes a book. No number of prototypes becomes a company. No number of vibe-coded demos becomes a product unless someone carries the ugly weight of integration, support, distribution, and trust.

This is where Cal Newport’s critique of pseudo-productivity fits. When work lacks clear measures, visible activity becomes a proxy for value. The worker looks busy. The organization feels alive. The machine hums. But the question remains: did anything get heavier? Did any capability deepen? Did any promise become more trustworthy?⁵

AI can flood the zone with lightness. It can create infinite first drafts, infinite summaries, infinite mockups, infinite synthetic movement. The temptation is to confuse lower friction with greater seriousness. But seriousness is not how fast a thing can appear. Seriousness is what survives after appearance stops being impressive.

That is why the AI layoff narrative feels both true and fake.

Part of it is true because automation changes labor. Some roles are built around repeatable execution. Some companies will cut people because software can now do enough of the task. Some workers will get hurt. Any theory that skips that is just pundit cologne.

And it is fake because “AI did it” has become a socially acceptable mask for older motives. When a company says a layoff is because of AI, the claim is most credible if it can point to a specific workflow that was automated, a role whose output is now produced without equivalent headcount, and a measurable change in cycle time, cost, or quality. The claim is weakest when AI appears only in the press release while the cuts map onto old problems: margin pressure, overhiring, investor demands, or management fashion. The SHRM reporting on this captures the mess well.⁶ Some displacement is real. Some is anticipatory. Some is financial restructuring dressed in the clothes of technological inevitability. A company can automate customer service in one division while cutting sales staff in another and calling both “AI restructuring.” The phrase covers too much to be trustworthy.

The cleaner question is not whether AI replaces work. The cleaner question is what kind of work remains worth paying humans to do once execution gets cheap.

A lot of people will answer with taste. That is close, but too pretty. Taste without accountability becomes another aesthetic performance. The harder answer is responsibility over the whole sandwich.

Decide what matters. Execute enough to make it real. Deliver it into the world and stay with the consequences.

That full loop is the work.

AI can help with it. It can sharpen a decision, accelerate execution, test alternatives, surface counterarguments, and make the first version less painful. But it cannot make commitment painless. It cannot decide what you are willing to be judged by. It cannot turn a lightweight life into a heavy one unless someone chooses weight.

This is why the people who benefit most from AI may not be the ones with the best prompts. They may be the ones with the highest tolerance for commitment. People who know what they are building. People who can tell the difference between motion and progress. People who can use speed without worshipping it.

The uncomfortable possibility is that AI is making fake work harder to defend, not making work meaningless.

If your job was mostly producing artifacts that signaled progress, AI is a threat. If your organization rewarded process theater, AI will make the theater cheaper and more absurd. If your career was optimized for keeping options open forever, AI will hand you more options than you can metabolize. If your creative life was built around lightweight output, AI will bury you in weightless abundance.

But if your work was already anchored in judgment, commitment, depth, and delivery, AI looks different. It adds pressure to the parts that were never the real work anyway.

The middle got cheaper.

Now we find out who was hiding there.

Notes

Share lines

1. AI did not kill work. It made visible motion cheap.
2. The middle got cheaper. Now we find out who was hiding there.
3. Optionality feels like freedom until it becomes a way to avoid becoming specific.
4. If a process artifact can be generated in thirty seconds, the artifact cannot be the proof anymore.
5. Seriousness is not how fast a thing appears. Seriousness is what survives after appearance stops being impressive.

Revision notes

• The Cisco/AI-layoff example is factually verified: Cisco announced fiscal Q3 2026 record revenue of $15.84 billion on May 13, 2026, and disclosed plans to cut nearly 4,000 jobs the same day, attributing the restructuring to a focus on AI infrastructure.
• The SHRM article by Roy Maurer, “The AI Layoffs Narrative: Real Transformation, or Scapegoat?” (May 18, 2026) is a verified source.
• The Narayanan/Kapoor article, “Why AI hasn’t replaced software engineers, and won’t” (June 10, 2026) is verified. The “decide-execute-deliver sandwich” is their exact framing.
• The Perell essay “Hugging the X-Axis” is verified.
• The Repenning/Sterman paper, “Nobody Ever Gets Credit for Fixing Problems that Never Happened” (California Management Review 43, no. 4, Summer 2001) is verified.
• The Cal Newport book Slow Productivity (Portfolio, 2024) is verified.
• The self-indicting paragraph was added to address the critique that the essay is the thing it critiques.
• The falsifiable claim about when “because of AI” is credible vs. theater was added to replace the slippery “same story can be true in one department and theater in another” construction.
• The June reading notes paragraph was removed to eliminate the leaked scaffolding.
• The fragment-stack rhythm was varied to avoid repetition.
• Proper footnotes with URLs and full bibliographic detail were added to the body of the essay.
• Avoid overclaiming labor impact. Keep the line that displacement is real in some places.
• If publishing, decide whether to keep the direct Narayanan/Kapoor sandwich frame by name or translate it more loosely for non-technical readers.
• Possible alternate title: “The middle got cheaper.”
• Originality classification should remain fresh synthesis, not novel.