SPB Git

spb/spboucher.ai Public

spboucher.ai — personal website of Simon-Pierre Boucher.

🔗 https://spboucher.ai
websiteportfolionextjs
TypeScript 93.4% HTML 5.5% CSS 1%

Blog: add essay 'The Macroeconomics of White-Collar Automation'

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Simon-Pierre Boucher committed yesterday (Aug 9, 2026) parent 449f194

Showing 2 changed files with +906 and −0

added blog/blog4.txt +896 −0
@@ -0,0 +1,896 @@
1 +# The Macroeconomics of White-Collar Automation
2 +
3 +Artificial intelligence is usually discussed as a labor-market technology.
4 +
5 +Will programmers lose their jobs?
6 +
7 +Will accountants be automated?
8 +
9 +Will lawyers need fewer junior associates?
10 +
11 +Will analysts, consultants, researchers, designers, and managers become less valuable?
12 +
13 +Those are important questions.
14 +
15 +But they may be too narrow.
16 +
17 +If artificial intelligence substantially automates white-collar work, the consequences will not remain confined to individual occupations. White-collar workers sit near the center of modern developed economies. They earn relatively high incomes, pay significant amounts of income tax, hold mortgages, own financial assets, consume services, and support expensive urban housing markets.
18 +
19 +A large shock to cognitive labor therefore becomes a macroeconomic shock.
20 +
21 +The relevant question is not simply:
22 +
23 +> **How many jobs will AI automate?**
24 +
25 +It is:
26 +
27 +> **What happens when the marginal cost of cognitive production falls faster than an economy can redistribute the resulting productivity gains?**
28 +
29 +## 1. AI as a New Factor of Production
30 +
31 +A simple way to think about AI is to introduce machine cognition directly into the production function.
32 +
33 +Instead of an economy producing output using only capital and human labor,
34 +
35 +$$
36 +Y_t = A_t F(K_t, L_t),
37 +$$
38 +
39 +suppose production increasingly depends on three major inputs:
40 +
41 +$$
42 +Y_t = A_t F(K_t, L_t^{H}, L_t^{AI}),
43 +$$
44 +
45 +where:
46 +
47 +* \(Y_t\) is aggregate output,
48 +* \(K_t\) is conventional capital,
49 +* \(L_t^{H}\) is human labor,
50 +* \(L_t^{AI}\) is machine-provided cognitive labor,
51 +* and \(A_t\) captures general productivity.
52 +
53 +This distinction matters because \(L_t^{AI}\) has very different economics from human labor.
54 +
55 +Human labor is scarce.
56 +
57 +It must be educated, recruited, compensated, coordinated, and given time to perform tasks.
58 +
59 +Machine cognition can potentially be replicated at extremely low marginal cost.
60 +
61 +An effective cognitive-labor aggregate might therefore look like:
62 +
63 +$$
64 +L_t^{\text{eff}} = L_t^{H} + \phi_t L_t^{AI},
65 +$$
66 +
67 +where \(\phi_t\) represents the relative effectiveness of AI labor.
68 +
69 +If \(\phi_t\) rises rapidly, the economy can obtain substantially more effective cognitive labor without proportionally increasing human employment.
70 +
71 +That is the central macroeconomic novelty.
72 +
73 +## 2. The First Effect Is a Productivity Boom
74 +
75 +Initially, the effect should look extremely positive.
76 +
77 +A firm that required 100 analysts may eventually produce comparable output with 40 analysts using powerful AI systems.
78 +
79 +A software company may ship more code with fewer engineers.
80 +
81 +A law firm may process thousands of documents in hours rather than weeks.
82 +
83 +An accounting department may automate reconciliation, reporting, and document processing.
84 +
85 +In a simple production setting,
86 +
87 +$$
88 +\frac{\partial Y}{\partial L^{AI}} > 0.
89 +$$
90 +
91 +More machine cognition increases output.
92 +
93 +If AI also improves the productivity of existing workers, there is an additional complementarity:
94 +
95 +$$
96 +\frac{\partial^2 Y}{\partial L^{AI}\,\partial L^{H}} > 0.
97 +$$
98 +
99 +That is the optimistic version of the story.
100 +
101 +AI does not replace workers.
102 +
103 +It amplifies them.
104 +
105 +For some occupations, that may dominate for a long time.
106 +
107 +But complementarity does not guarantee permanent labor demand.
108 +
109 +A technology can initially make workers more productive and later become capable enough to substitute for them.
110 +
111 +## 3. The Important Variable Is the Elasticity of Substitution
112 +
113 +The macroeconomic outcome depends heavily on how easily machine cognition substitutes for human cognition.
114 +
115 +A useful conceptual production function is a CES specification:
116 +
117 +$$
118 +L_t^{\text{cog}} = \left[ \alpha (L_t^{H})^{\frac{\sigma-1}{\sigma}} + (1-\alpha)(\phi_t L_t^{AI})^{\frac{\sigma-1}{\sigma}} \right]^{\frac{\sigma}{\sigma-1}},
119 +$$
120 +
121 +where \(\sigma\) is the elasticity of substitution between human and machine cognitive labor.
122 +
123 +If
124 +
125 +$$
126 +\sigma < 1,
127 +$$
128 +
129 +the two inputs are relatively complementary.
130 +
131 +If
132 +
133 +$$
134 +\sigma > 1,
135 +$$
136 +
137 +substitution becomes economically powerful.
138 +
139 +As AI capabilities improve and the price of machine cognition falls, firms have a stronger incentive to replace human cognitive tasks.
140 +
141 +The relevant price ratio becomes something like:
142 +
143 +$$
144 +\frac{P_{AI}}{w_H},
145 +$$
146 +
147 +where \(P_{AI}\) is the cost of acquiring a unit of AI cognitive output and \(w_H\) is the human wage.
148 +
149 +If
150 +
151 +$$
152 +\frac{P_{AI}}{w_H} \rightarrow 0,
153 +$$
154 +
155 +then even moderate substitutability can produce major organizational changes.
156 +
157 +## 4. Why White-Collar Automation Is Macroeconomically Different
158 +
159 +The automation of some previous forms of labor affected relatively low-paid occupations.
160 +
161 +White-collar automation can hit a very different section of the income distribution.
162 +
163 +Consider aggregate labor income:
164 +
165 +$$
166 +Y_L = \sum_i w_i L_i.
167 +$$
168 +
169 +A relatively small number of high-income workers can contribute disproportionately to \(Y_L\).
170 +
171 +If AI primarily affects occupations with wages above the economy-wide average, then employment losses need not be enormous to create a significant decline in aggregate labor income.
172 +
173 +For example, suppose ten percent of workers generate twenty-five percent of total labor income.
174 +
175 +Automating a substantial fraction of their tasks creates a much larger demand shock than the headline employment number alone would suggest.
176 +
177 +This is one reason measuring only unemployment could be misleading.
178 +
179 +The economy might experience limited unemployment while still seeing:
180 +
181 +* slower wage growth,
182 +* fewer entry-level positions,
183 +* declining hours,
184 +* weaker bargaining power,
185 +* lower professional-service prices,
186 +* and a shift from labor income toward capital income.
187 +
188 +The transition can occur through wages before it occurs through unemployment.
189 +
190 +## 5. Labor Share Becomes a Critical Variable
191 +
192 +Define labor's share of aggregate income as:
193 +
194 +$$
195 +s_L = \frac{wL}{Y}.
196 +$$
197 +
198 +If AI increases production faster than labor compensation, then:
199 +
200 +$$
201 +\frac{\dot{Y}}{Y} > \frac{\dot{(wL)}}{wL},
202 +$$
203 +
204 +and consequently:
205 +
206 +$$
207 +\dot{s}_L < 0.
208 +$$
209 +
210 +The economy can therefore become substantially more productive while the fraction of output paid to workers falls.
211 +
212 +This is not logically contradictory.
213 +
214 +Suppose output rises from 100 to 150 while total compensation rises only from 60 to 65.
215 +
216 +Workers are receiving slightly more aggregate income.
217 +
218 +But labor's share falls from
219 +
220 +$$
221 +\frac{60}{100} = 60\%
222 +$$
223 +
224 +to approximately
225 +
226 +$$
227 +\frac{65}{150} \approx 43.3\%.
228 +$$
229 +
230 +The economy is richer.
231 +
232 +Labor receives more income in absolute terms.
233 +
234 +And yet capital captures most of the incremental output.
235 +
236 +That distinction is central to understanding AI.
237 +
238 +## 6. Productivity Growth Does Not Automatically Become Demand Growth
239 +
240 +GDP expenditure can be written as:
241 +
242 +$$
243 +Y = C + I + G + NX.
244 +$$
245 +
246 +AI primarily attacks the supply side first.
247 +
248 +It increases what firms are capable of producing.
249 +
250 +But additional productive capacity does not automatically create proportional demand.
251 +
252 +Suppose household consumption depends on labor and capital income:
253 +
254 +$$
255 +C = c_L Y_L + c_K Y_K,
256 +$$
257 +
258 +where \(Y_L\) is labor income and \(Y_K\) is capital income.
259 +
260 +If workers have a higher marginal propensity to consume than capital owners,
261 +
262 +$$
263 +c_L > c_K,
264 +$$
265 +
266 +then shifting one dollar of national income from labor toward capital can reduce immediate aggregate consumption.
267 +
268 +Now suppose:
269 +
270 +$$
271 +Y = Y_L + Y_K.
272 +$$
273 +
274 +An AI-driven redistribution can occur even while total income increases:
275 +
276 +$$
277 +\Delta Y > 0,
278 +$$
279 +
280 +but if
281 +
282 +$$
283 +\Delta Y_L < 0
284 +$$
285 +
286 +and
287 +
288 +$$
289 +\Delta Y_K \gg 0,
290 +$$
291 +
292 +consumption may grow much more slowly than productive capacity.
293 +
294 +This creates one of the central macroeconomic tensions of advanced automation:
295 +
296 +$$
297 +\boxed{ \text{AI productivity growth} \not\Rightarrow \text{proportional aggregate demand growth} }
298 +$$
299 +
300 +The economy can become capable of producing more than households are willing or able to purchase at previous prices.
301 +
302 +## 7. AI Could Be Structurally Disinflationary
303 +
304 +This leads naturally to prices.
305 +
306 +Suppose potential output rises rapidly because of AI:
307 +
308 +$$
309 +Y_t^* \uparrow.
310 +$$
311 +
312 +If aggregate demand grows more slowly,
313 +
314 +$$
315 +AD_t < Y_t^*,
316 +$$
317 +
318 +the output gap becomes negative relative to productive capacity.
319 +
320 +In a simplified Phillips-curve framework,
321 +
322 +$$
323 +\pi_t = \pi_t^e + \kappa (Y_t - Y_t^*) + u_t.
324 +$$
325 +
326 +If AI causes \(Y_t^*\) to rise faster than actual spending \(Y_t\), then:
327 +
328 +$$
329 +Y_t - Y_t^* < 0,
330 +$$
331 +
332 +placing downward pressure on inflation.
333 +
334 +This means sufficiently powerful AI may be structurally disinflationary in many cognitive services.
335 +
336 +Software development becomes cheaper.
337 +
338 +Legal analysis becomes cheaper.
339 +
340 +Translation becomes cheaper.
341 +
342 +Marketing production becomes cheaper.
343 +
344 +Financial analysis becomes cheaper.
345 +
346 +Certain forms of education become cheaper.
347 +
348 +Content production becomes cheaper.
349 +
350 +The economic problem could eventually shift from producing enough to maintaining sufficient nominal demand.
351 +
352 +## 8. But Asset Prices Could Move in the Opposite Direction
353 +
354 +At the same time, AI does not necessarily make every scarce asset cheaper.
355 +
356 +In fact, the opposite may occur.
357 +
358 +If AI increases the returns to capital ownership, income may flow toward owners of:
359 +
360 +* AI companies,
361 +* semiconductor firms,
362 +* energy infrastructure,
363 +* data centers,
364 +* proprietary datasets,
365 +* intellectual property,
366 +* land,
367 +* and productive businesses.
368 +
369 +Capitalized asset values depend on expected future cash flows:
370 +
371 +$$
372 +P_t = \sum_{j=1}^{\infty} \frac{E_t[CF_{t+j}]}{(1+r)^j}.
373 +$$
374 +
375 +If AI dramatically raises expected profits,
376 +
377 +$$
378 +E_t[CF_{t+j}] \uparrow,
379 +$$
380 +
381 +then asset prices can rise even while wage growth weakens.
382 +
383 +We could therefore see a strange combination:
384 +
385 +$$
386 +\text{consumer-price disinflation} + \text{asset-price inflation}.
387 +$$
388 +
389 +Digital goods become cheaper.
390 +
391 +Labor becomes less scarce.
392 +
393 +But ownership of productive capital becomes more valuable.
394 +
395 +That environment would feel very different depending on whether a household primarily earns wages or owns assets.
396 +
397 +## 9. White-Collar Automation Could Hit Housing Through Income, Not Supply
398 +
399 +Housing introduces another transmission mechanism.
400 +
401 +In many major cities, expensive housing is supported by large populations of high-income professional workers.
402 +
403 +Housing demand can be represented loosely as:
404 +
405 +$$
406 +D_H = f(Y_L, r, E[P_{H,t+1}], N, \ldots),
407 +$$
408 +
409 +where labor income \(Y_L\) is an important driver.
410 +
411 +If AI weakens professional income or reduces the number of workers required in expensive commercial centers, demand for certain housing markets may change.
412 +
413 +This would not necessarily reduce housing prices everywhere.
414 +
415 +Housing remains constrained by land, regulation, geography, and construction.
416 +
417 +But AI could produce substantial relative-price effects.
418 +
419 +Cities whose valuations depend heavily on concentrations of white-collar employment may respond differently from locations where housing value derives primarily from physical scarcity, amenities, or population growth.
420 +
421 +Remote AI-enabled work could amplify this effect.
422 +
423 +If one highly productive worker can coordinate many digital agents from almost anywhere, the economic premium attached to physically locating large teams in expensive cities may decline.
424 +
425 +## 10. Credit Markets Create an Amplification Channel
426 +
427 +White-collar workers do not merely consume.
428 +
429 +They borrow against expectations of future income.
430 +
431 +A standard household borrowing constraint might look like:
432 +
433 +$$
434 +B_t \leq \theta E_t \left[ \sum_{j=1}^{T} \frac{Y_{L,t+j}}{(1+r)^j} \right].
435 +$$
436 +
437 +Credit capacity depends partly on expected future labor income.
438 +
439 +If households begin to believe that certain professional incomes are less secure, lenders may eventually incorporate that risk into underwriting.
440 +
441 +That could lower borrowing capacity even before actual job losses occur.
442 +
443 +The mechanism becomes:
444 +
445 +$$
446 +\text{AI exposure} \rightarrow \text{lower expected lifetime income} \rightarrow \text{lower credit capacity} \rightarrow \text{lower consumption and housing demand}.
447 +$$
448 +
449 +Because developed economies are highly leveraged, this expectation channel could matter.
450 +
451 +Automation does not need to eliminate millions of jobs overnight.
452 +
453 +It only needs to alter expectations about the persistence of future earnings.
454 +
455 +## 11. The Entry-Level Problem May Arrive First
456 +
457 +One especially important transition mechanism concerns junior workers.
458 +
459 +Organizations often use junior employees for tasks that are simultaneously productive and educational.
460 +
461 +Drafting.
462 +
463 +Research.
464 +
465 +Data preparation.
466 +
467 +Basic programming.
468 +
469 +Financial modeling.
470 +
471 +Document review.
472 +
473 +Presentation preparation.
474 +
475 +Those tasks also happen to be among the easiest to delegate to AI.
476 +
477 +Suppose firms require experienced workers \(L_S\) and junior workers \(L_J\).
478 +
479 +Traditionally, today's junior employees become tomorrow's senior employees:
480 +
481 +$$
482 +L_{S,t+1} = (1-\delta)L_{S,t} + \gamma L_{J,t}.
483 +$$
484 +
485 +If AI substantially reduces demand for juniors,
486 +
487 +$$
488 +L_{J,t} \downarrow,
489 +$$
490 +
491 +then eventually:
492 +
493 +$$
494 +L_{S,t+1} \downarrow.
495 +$$
496 +
497 +This creates an institutional problem.
498 +
499 +Automation can remove the training layer from a profession before it removes the profession itself.
500 +
501 +In the short run, senior professionals become more productive.
502 +
503 +In the long run, fewer humans accumulate the experience required to become senior professionals.
504 +
505 +That dynamic could reshape law, finance, consulting, software engineering, academia, medicine, and many other knowledge industries.
506 +
507 +## 12. The Corporate Sector Could Become Extremely Profitable
508 +
509 +From the firm's perspective, AI can reduce unit labor costs.
510 +
511 +Define unit labor cost as:
512 +
513 +$$
514 +ULC = \frac{wL}{Y}.
515 +$$
516 +
517 +If output rises while payroll falls or grows slowly:
518 +
519 +$$
520 +ULC \downarrow.
521 +$$
522 +
523 +Holding prices approximately constant, operating margins increase.
524 +
525 +A simplified profit function is:
526 +
527 +$$
528 +\Pi = PY - wL - rK - C_{AI}.
529 +$$
530 +
531 +If AI substitutes for expensive labor such that:
532 +
533 +$$
534 +\Delta C_{AI} < -\Delta(wL),
535 +$$
536 +
537 +then profits increase.
538 +
539 +At scale, this could produce unusually high corporate profitability.
540 +
541 +The macroeconomic distribution becomes:
542 +
543 +$$
544 +Y = W + \Pi + R + T + \cdots
545 +$$
546 +
547 +with the profit share \(\Pi/Y\) potentially rising relative to the wage share \(W/Y\).
548 +
549 +Equity markets might therefore respond positively to developments that create significant anxiety in labor markets.
550 +
551 +That is not paradoxical.
552 +
553 +It reflects the difference between owning firms and working for them.
554 +
555 +## 13. GDP May Become a Worse Measure of Economic Security
556 +
557 +Imagine an extreme scenario.
558 +
559 +AI allows the economy to produce twice as many valuable services with half as much human labor.
560 +
561 +GDP increases enormously.
562 +
563 +Productivity increases enormously.
564 +
565 +Corporate profits increase.
566 +
567 +Consumer prices decline.
568 +
569 +Measured economic efficiency looks excellent.
570 +
571 +Yet many households experience lower labor income.
572 +
573 +We could then observe:
574 +
575 +$$
576 +\frac{Y}{N} \uparrow
577 +$$
578 +
579 +while, for a substantial fraction of households,
580 +
581 +$$
582 +Y_{L,i} \downarrow.
583 +$$
584 +
585 +GDP per capita rises while perceived economic security falls.
586 +
587 +That is possible because GDP measures production, not the distribution of claims on production.
588 +
589 +An AI-intensive economy could therefore force economists and policymakers to pay much more attention to:
590 +
591 +* median disposable income,
592 +* household wealth,
593 +* labor share,
594 +* capital ownership,
595 +* consumption distribution,
596 +* and access to essential goods.
597 +
598 +Aggregate abundance does not automatically imply household security.
599 +
600 +## 14. Monetary Policy Would Face a Strange Problem
601 +
602 +Central banks are accustomed to recessions caused by weak productivity, financial crises, demand collapses, or supply shocks.
603 +
604 +AI could create something unusual:
605 +
606 +**a positive supply shock combined with a negative labor-income shock.**
607 +
608 +Potential output rises:
609 +
610 +$$
611 +Y^* \uparrow.
612 +$$
613 +
614 +Inflation falls:
615 +
616 +$$
617 +\pi \downarrow.
618 +$$
619 +
620 +But labor-market weakness increases:
621 +
622 +$$
623 +u \uparrow
624 +$$
625 +
626 +or wage growth slows:
627 +
628 +$$
629 +\dot{w} \downarrow.
630 +$$
631 +
632 +A standard policy rule such as:
633 +
634 +$$
635 +i_t = r^* + \pi_t + \phi_{\pi}(\pi_t - \pi^*) + \phi_y(Y_t - Y_t^*)
636 +$$
637 +
638 +would likely imply easier monetary policy.
639 +
640 +But lower interest rates do not solve the underlying distribution problem.
641 +
642 +Cheap credit can support aggregate demand.
643 +
644 +It cannot necessarily restore the bargaining power of labor if cognitive work has become structurally abundant.
645 +
646 +And lower rates may further increase asset prices, benefiting those who already own capital.
647 +
648 +Monetary policy could therefore become less capable of addressing the core economic consequences of automation.
649 +
650 +## 15. Fiscal Policy Becomes More Important
651 +
652 +Governments are heavily dependent on labor income.
653 +
654 +A simplified tax system can be written as:
655 +
656 +$$
657 +T = \tau_L Y_L + \tau_K Y_K + \tau_C C.
658 +$$
659 +
660 +If AI shifts income from labor toward capital,
661 +
662 +$$
663 +Y_L \downarrow, \qquad Y_K \uparrow,
664 +$$
665 +
666 +the tax base changes.
667 +
668 +This matters because many government systems were implicitly built around abundant employment.
669 +
670 +Payroll taxes finance social programs.
671 +
672 +Income taxes finance public services.
673 +
674 +Workers accumulate pension rights through employment.
675 +
676 +Health and other benefits can be linked directly or indirectly to work.
677 +
678 +If labor's share falls significantly, the fiscal architecture may eventually need to follow the income.
679 +
680 +The problem becomes less:
681 +
682 +> How should we tax robots?
683 +
684 +and more:
685 +
686 +> Where is national income actually accruing?
687 +
688 +A tax system designed for an AI-intensive economy may need to rely more heavily on profits, capital income, consumption, land, or broad ownership structures.
689 +
690 +## 16. The Real Problem Is the Transition
691 +
692 +The long-run equilibrium could be extraordinarily prosperous.
693 +
694 +Imagine a world where machine intelligence makes:
695 +
696 +* legal services,
697 +* software,
698 +* financial analysis,
699 +* education,
700 +* design,
701 +* administration,
702 +* and research
703 +
704 +dramatically cheaper.
705 +
706 +Real living standards could increase enormously.
707 +
708 +The difficult part is getting from the current economy to that economy.
709 +
710 +The transition can be represented as a race between productivity and redistribution:
711 +
712 +$$
713 +g_{AI} \quad \text{versus} \quad g_R,
714 +$$
715 +
716 +where \(g_{AI}\) is the rate at which AI changes productive capacity and \(g_R\) is the rate at which institutions, ownership, wages, prices, and new industries redistribute those gains.
717 +
718 +If
719 +
720 +$$
721 +g_{AI} \gg g_R,
722 +$$
723 +
724 +the transition becomes economically unstable.
725 +
726 +Not because AI fails.
727 +
728 +But because it succeeds faster than social and economic institutions adapt.
729 +
730 +## 17. New Jobs Are Not a Complete Answer
731 +
732 +A common response is that technology always creates new occupations.
733 +
734 +Historically, that is largely true.
735 +
736 +But the important question is not whether new jobs exist.
737 +
738 +It is whether they appear:
739 +
740 +* quickly enough,
741 +* at sufficient scale,
742 +* with similar compensation,
743 +* and with skills that displaced workers can realistically acquire.
744 +
745 +Suppose AI destroys cognitive tasks at rate \(\delta_A\) and new labor demand emerges at rate \(\lambda_N\).
746 +
747 +A smooth transition requires something like:
748 +
749 +$$
750 +\lambda_N \geq \delta_A.
751 +$$
752 +
753 +But if:
754 +
755 +$$
756 +\delta_A \gg \lambda_N,
757 +$$
758 +
759 +even temporary adjustment can become a major macroeconomic event.
760 +
761 +A ten-year transition may eventually look harmless in a century-long economic chart.
762 +
763 +It can still define an entire generation's economic experience.
764 +
765 +## 18. The Most Important Variable May Be Ownership
766 +
767 +All of these mechanisms ultimately converge on one issue.
768 +
769 +Ownership.
770 +
771 +Suppose productive output increasingly depends on AI capital:
772 +
773 +$$
774 +Y = F(K_{AI}, K_O, L_H).
775 +$$
776 +
777 +As
778 +
779 +$$
780 +\frac{\partial Y}{\partial K_{AI}}
781 +$$
782 +
783 +grows, ownership of \(K_{AI}\) becomes increasingly important.
784 +
785 +If ownership is highly concentrated, AI may produce high growth alongside high inequality.
786 +
787 +If ownership is broad, households can receive AI-generated income through dividends, retirement funds, public investment vehicles, employee ownership, or direct entrepreneurship.
788 +
789 +The distributional question therefore changes.
790 +
791 +Instead of asking only whether people can continue selling labor, we may need to ask whether people participate in ownership of productive capital.
792 +
793 +A highly automated economy with broad capital ownership looks radically different from a highly automated economy with concentrated capital ownership.
794 +
795 +The technology can be identical.
796 +
797 +The macroeconomics are not.
798 +
799 +## 19. The Abundance Paradox
800 +
801 +This leads to a strange possible future.
802 +
803 +The economy becomes capable of producing unprecedented amounts of cognitive output.
804 +
805 +The supply of software, analysis, designs, research, administration, and digital services explodes.
806 +
807 +The marginal cost of many services approaches zero.
808 +
809 +Yet households can still feel financially insecure because access to that abundance remains mediated by income and ownership.
810 +
811 +The paradox can be expressed simply:
812 +
813 +$$
814 +\text{productive abundance} \neq \text{income abundance}.
815 +$$
816 +
817 +An economy can solve the technical problem of scarcity before it solves the institutional problem of distribution.
818 +
819 +That may be the central macroeconomic challenge of advanced AI.
820 +
821 +## 20. The Macroeconomic Question Is Larger Than Employment
822 +
823 +The debate over AI and jobs often assumes a binary outcome.
824 +
825 +Either AI replaces workers or it does not.
826 +
827 +The actual transition will probably be much more complicated.
828 +
829 +AI can simultaneously:
830 +
831 +* increase worker productivity,
832 +* reduce the number of workers required per unit of output,
833 +* lower consumer prices,
834 +* raise corporate profits,
835 +* reduce labor's income share,
836 +* increase the value of scarce assets,
837 +* weaken some urban housing markets,
838 +* increase others,
839 +* reduce inflation,
840 +* alter tax revenues,
841 +* increase wealth inequality,
842 +* and raise real GDP.
843 +
844 +None of these outcomes are mutually exclusive.
845 +
846 +The most important macroeconomic identity remains simple:
847 +
848 +$$
849 +Y = C + I + G + NX.
850 +$$
851 +
852 +But AI potentially changes the mechanisms behind every component.
853 +
854 +Consumption depends on who receives the income.
855 +
856 +Investment depends on the expected returns to AI capital.
857 +
858 +Government spending depends on a tax system increasingly disconnected from employment.
859 +
860 +Net exports depend on which economies control the productive technologies.
861 +
862 +And aggregate supply could expand at a speed rarely seen outside major industrial transformations.
863 +
864 +The defining economic question of artificial intelligence may therefore not be whether machines become capable of doing white-collar work.
865 +
866 +That seems increasingly like a technical question.
867 +
868 +The harder question is what happens afterward.
869 +
870 +If cognitive labor becomes abundant, then the institutions of an economy built around scarce human cognition will begin to look increasingly strange.
871 +
872 +Wages.
873 +
874 +Careers.
875 +
876 +Mortgages.
877 +
878 +Education.
879 +
880 +Taxation.
881 +
882 +Retirement.
883 +
884 +Corporate organization.
885 +
886 +Even the relationship between employment and economic survival.
887 +
888 +All of these institutions were designed around a simple historical assumption:
889 +
890 +> **Useful intelligence is scarce because useful intelligence requires human time.**
891 +
892 +Artificial intelligence attacks that assumption directly.
893 +
894 +And if that assumption breaks, white-collar automation will not simply change the labor market.
895 +
896 +It will change the macroeconomic architecture built on top of it.
modified lib/blog.ts +10 −0
@@ -32,6 +32,16 @@ export interface BlogPost extends BlogPostMeta {
32 32
33 33 /** Registry of published essays — sources live in /blog/*.txt. */
34 34 const registry: BlogPostMeta[] = [
35 + {
36 + slug: "macroeconomics-of-white-collar-automation",
37 + file: "blog4.txt",
38 + title: "The Macroeconomics of White-Collar Automation",
39 + excerpt:
40 + "White-collar workers sit near the center of developed economies — their incomes support taxes, mortgages, consumption, and urban housing. If AI automates cognitive work faster than institutions redistribute the gains, the shock becomes macroeconomic, not merely occupational.",
41 + date: "2026-08-09",
42 + dateLabel: "August 9, 2026",
43 + tags: ["AI", "Economics", "Macroeconomics"],
44 + },
35 45 {
36 46 slug: "cost-of-intelligence",
37 47 file: "blog1.txt",
38 48