Outsource

1. Scribes (ancient Mesopotamia/Egypt/Rome): elites outsource writing, recordkeeping, correspondence.
2. Medieval monastic copyists: outsourcing book reproduction and textual preservation to monasteries.
3. Notaries/public scribes (Middle Ages–modern): drafting contracts, wills, legal forms for the illiterate/busy.
4. Court clerks and chancery staff: rulers outsource administration, decrees, and bureaucratic memory.
5. Town criers: communities outsource public announcements and dissemination of information.
6. Libraries/archivists: institutions outsource long-term memory and document retrieval.
7. Accountants/bookkeepers (Renaissance onward): merchants outsource double-entry bookkeeping and financial control.
8. Tax farming (ancient Rome, Ottoman, French Ancien Régime): states outsource tax collection to contractors.
9. Customs brokers: firms outsource navigating tariffs, import rules, and paperwork.
10. Freight forwarders: companies outsource route planning, shipping coordination, documentation.
11. Shipping insurers/Lloyd’s: merchants outsource risk assessment and pooling for maritime trade.
12. Underwriters/actuaries: organizations outsource probabilistic forecasting (life, fire, shipping).
13. Bailiffs/debt collectors: creditors outsource collections and enforcement.
14. Mercenaries/condottieri: states outsource military capability and tactical expertise.
15. Privateers: governments outsource naval warfare to licensed entrepreneurs.
16. Intelligence networks/spies: rulers outsource information gathering and analysis.
17. Diplomatic envoys: sovereigns outsource negotiation, cultural translation, political judgment.
18. Astrologers at court: elites outsource decision timing/strategy to “expert” star-reading.
19. Oracles (Delphi, etc.): individuals/states outsource high-stakes decisions to religious authorities.
20. Priests/confessors: people outsource moral guidance and personal counsel.
21. Matchmakers: families outsource partner search, vetting, and negotiation.
22. Midwives/doulas: families outsource childbirth expertise and risk management.
23. Physicians: individuals outsource diagnosis and treatment planning.
24. Pharmacists/apothecaries: outsourcing compounding, drug knowledge, interactions.
25. Nurses: patients outsource monitoring and day-to-day clinical judgment.
26. Dentists: people outsource oral health maintenance/decisions.
27. Veterinarians: farmers outsource animal health expertise and interventions.
28. Agronomists/extension agents: farmers outsource crop planning, pest management, soil advice.
29. Weather forecasters/almanacs: sailors/farmers outsource prediction of conditions.
30. Navigators/pilots (age of sail): captains outsource specialized navigation in hazardous waters.
31. Cartographers: states and merchants outsource map-making and geographic knowledge.
32. Surveyors: landowners outsource measurement, boundary setting, technical accuracy.
33. Engineers: organizations outsource design and safety calculations (bridges, rail, factories).
34. Architects: clients outsource spatial planning, codes, aesthetics, structural integration.
35. Master builders/guilds: patrons outsource construction methods and quality control.
36. Project managers: firms outsource coordination, scheduling, risk tracking.
37. Quality inspectors: factories outsource checking and standards compliance.
38. Standards bodies (weights/measures): markets outsource trust and calibration to institutions.
39. Auditors: investors/boards outsource verification and fraud detection.
40. Credit rating agencies: markets outsource assessment of default risk (e.g., bonds).
41. Credit bureaus/credit scores: lenders outsource borrower assessment to centralized scoring.
42. Loan officers/underwriting departments: banks outsource lending decisions from individuals to process.
43. Insurance adjusters: claims handling and judgment outsourced from policyholders to professionals.
44. Lawyers/solicitors: individuals/firms outsource legal interpretation, strategy, drafting.
45. Patent attorneys/agents: inventors outsource claims strategy and filings.
46. Compliance officers: firms outsource interpreting regulations and designing controls.
47. HR recruiters/headhunters: companies outsource candidate sourcing, screening, negotiation.
48. Payroll providers: firms outsource wage calculation, tax withholding, filings.
49. Benefits administrators: companies outsource complex plan management and enrollment.
50. Training/continuing education vendors: companies outsource skill-building programs.
51. Management consultants: executives outsource diagnosis of organizational problems and strategy options.
52. Strategy “war rooms”/think tanks: governments outsource policy research and long-range analysis.
53. Lobbyists: organizations outsource political persuasion and legislative monitoring.
54. PR agencies: firms outsource messaging, media relations, crisis communications.
55. Advertising agencies: companies outsource market research, creative strategy, campaign execution.
56. Market research firms (surveys/focus groups): firms outsource understanding customer preferences.
57. Branding firms: startups outsource naming, positioning, identity design.
58. Sales outsourcing/rep firms: manufacturers outsource customer acquisition and relationship management.
59. Customer support call centers: companies outsource troubleshooting and user guidance.
60. Virtual assistants/executive assistants: individuals outsource scheduling, reminders, correspondence triage.
61. Switchboard operators (early telecom): users outsource call routing and directory knowledge.
62. Concierges: travelers outsource local planning and reservations.
63. Travel agents: people outsource itinerary design, booking, problem resolution.
64. Event planners: clients outsource logistics, vendor negotiation, scheduling.
65. Caterers: hosts outsource meal planning and execution.
66. Household servants/butlers (historical): elites outsource domestic management and social coordination.
67. Nannies/governesses: families outsource childcare and education.
68. Tutors: students outsource guidance and feedback for learning.
69. Editors: authors outsource clarity, structure, fact-checking, style.
70. Ghostwriters: public figures outsource drafting speeches/books/op-eds.
71. Translators/interpreters: organizations outsource language understanding and cultural nuance.
72. Court reporters: legal system outsources accurate transcription and record.
73. Secretarial pools/typists: offices outsource document production and formatting.
74. Dictation machines + transcriptionists: professionals outsource writing from speech to specialists.
75. Spellcheck/grammar check (software): individuals outsource proofreading and correctness.
76. Style guides/templates: organizations outsource consistency decisions to predefined rules.
77. Photocopiers/printing services: offices outsource document reproduction at scale.
78. IT managed service providers: companies outsource system administration and uptime.
79. Cloud hosting: firms outsource infrastructure capacity planning and maintenance.
80. Cybersecurity firms/SOCs: companies outsource threat monitoring and incident response.
81. Penetration testers: orgs outsource security evaluation and adversarial thinking.
82. Data analysts/BI teams: executives outsource metric definition and interpretation.
83. Data science teams: product groups outsource forecasting, ranking, and experimentation analysis.
84. A/B testing platforms: companies outsource experiment plumbing and statistical routines.
85. Algorithmic trading systems: investors outsource trading decisions to automated strategies.
86. Portfolio managers/financial advisors: individuals outsource asset allocation and planning.
87. Robo-advisors: individuals outsource investment decisions to standardized algorithms.
88. Retirement plan target-date funds: savers outsource rebalancing and glidepath decisions.
89. Estate planners/trust officers: families outsource intergenerational planning and tax strategy.
90. Professional negotiators/buyers: firms outsource supplier negotiation and procurement strategy.
91. Procurement platforms: companies outsource vendor selection processes to systems/workflows.
92. Legal document automation: firms outsource contract drafting patterns to templates/software.
93. Cooks/meal kit services: individuals outsource meal planning and ingredient sourcing.
94. Personal trainers: people outsource exercise programming and adherence coaching.
95. Therapists/coaches: individuals outsource structured reflection, coping strategies, goal planning.
96. Diagnostic labs: clinicians outsource testing and interpretation to specialized labs.
97. Radiologists/pathologists: hospitals outsource image/tissue interpretation to specialists.
98. GPS navigation (consumer): drivers outsource route planning and spatial memory to devices.
99. Calculators/spreadsheets: workers outsource arithmetic and many routine quantitative steps.
100. Search engines: people outsource information retrieval, “where do I find this?” memory, and initial research.

If you tell me which one (or a small cluster) you want to zoom in on, I can do a historical deep dive focused on what skills atrophied, what new skills emerged, and what institutions/practices mitigated the downsides.

Below are places where historical “outsourcing” strongly supports Husom’s concerns, and places where it pushes back—using analogues from writing, bureaucracy, navigation, management, medicine, etc. The throughline from history is: outsourcing rarely causes a simple “people stop thinking”; it changes which thinking happens, who does it, and what failure modes appear.

1) Personal communication, authenticity, and “deceptive to fake”

What history supports in the post

Husom’s claim: undisclosed LLM help in personal writing violates expectations; phrasing carries meaning; trust erodes.

Strong historical analogues:

• Ghostwriters and speechwriters. For centuries, public figures have outsourced words. The social system adapted by creating role-based expectations: a president is expected to have speechwriters; a friend texting you is not. The key variable has been norms + disclosure. When ghostwriting is hidden in contexts that presume personal authorship, it’s treated as deception.
• Love letters via professional scribes. In many cultures, illiterate (or elite-but-busy) people used scribes for intimate letters. It “worked” socially when it was known/assumed (e.g., literacy constraints) and failed when it violated expectations of personal expression.
• Form letters and secretarial correspondence. Executives have long used assistants to send “their” emails/letters. Recipients often discount warmth/authenticity accordingly—again: expectations matter.

Takeaway: History backs Husom’s emphasis on shared expectations. “It’s fine” or “it’s deceptive” has depended less on the tool and more on whether the audience reasonably believes the sender personally authored the text.

Where history complicates his conclusion

Husom leans toward “keep human-to-human communication free of machine transformation.” The historical counterpoint is that societies often keep authenticity by creating conventions, not by banning mediation:

• Handwritten notes, signatures, voice calls, and later “sent from my iPhone” style markers all served as authenticity signals.
• Today’s equivalent could be disclosure norms (“AI-assisted”), provenance metadata, or social conventions about when AI-polish is acceptable (e.g., workplace HR emails vs. condolences to a friend).

So the lesson is less “mediation destroys communication,” and more: when a medium enables easy impersonation or over-polish, communities invent new authenticity signals—or lose trust until they do.

2) “The thin line”: tool vs replacement

Husom argues the line between spellcheck and LLM authorship is exceptionally thin and current interfaces push people into replacement.

History supports this interface-based concern. Many outsourcing waves created “slippery slopes” when the default workflow moved from assist to substitute:

• GPS navigation: from “help me confirm my route” to “I no longer form a route model at all.”
• Autopilot in aviation: from reducing workload to manual flying skill decay, prompting recurrent training requirements.
• Clinical decision support: from “double-check” to “I trust the system and stop practicing differential diagnosis,” leading to well-known deskilling worries.

Lesson: the design of the tool and default workflow predicts substitution more than the mere existence of the tool. Husom’s call for different interfaces is historically well-grounded.

3) “Functional text” vs “personal author to human audience”

His distinction (code/recipes/docs vs personal writing) maps onto older divisions:

• Notaries / standardized legal forms: society decided certain writing is primarily instrumental; originality/voice is secondary. Outsourcing is expected.
• Literature / personal correspondence: voice and authorship matter; outsourcing triggers backlash unless explicitly framed (co-authored, dictated, edited, etc.).

But the boundary moves. Historically, once an activity becomes high-volume and standardized, it tends to be reclassified as “functional” (e.g., much of business communication). That suggests Husom is right to worry about “inflation of text”—because institutional incentives often push communication into template-land.

4) Bureaucracy arms race: “the weapon exists on both sides”

Husom asks what happens when both parties have word generators.

History says: arms races are real and usually end with filters, gates, or new scarcity signals.

• Email spam → spam filters + reputation systems.
• SEO content farms → ranking algorithm changes + brand authority signals.
• Mass-produced cover letters/resumes (templates) → heavier reliance on referrals, work tests, credential screens, and automated parsing.
• PR/advertising growth → consumer skepticism + regulation + fact-checking + “earned trust” channels.

So the likely equilibrium isn’t “everyone writes more forever”; it’s:

1. a glut,
2. erosion of trust/attention,
3. new filtering institutions (automation, verification, reputational moats),
4. communication channels that re-price “human effort” as a credibility signal.

That supports his worry that “quality drops, volume rises,” and adds: systems tend to respond by shifting the bottleneck elsewhere.

5) Tacit knowledge, repetition, and “boring work builds you”

His jazz anecdote is a strong model: repeated internalization enables fluent generation. Historically, the strongest evidence is from domains where automation removed routine practice:

Supports his point (deskilling is real)

• Navigation: sailors and later drivers relying on tools can lose internal wayfinding models (the “I can’t navigate without GPS” phenomenon).
• Aviation: over-reliance on automation has been linked (in industry discussions and training doctrine) to degraded manual handling and situation awareness, hence mandated manual practice and simulator time.
• Medicine: over-reliance on imaging/labs can reduce physical exam skill and clinical reasoning practice; institutions respond with training standards and checklists.

These match Husom’s claim: you can’t cleanly separate “storage” from “processing” in humans; internalized knowledge changes what you can perceive and generate.

Complicates his point (deskilling is not the only outcome)

Outsourcing also creates new tacit knowledge, often higher-level:

• Double-entry bookkeeping outsourced to accountants didn’t eliminate financial thinking; it shifted merchants/executives toward capital allocation, risk, and strategy.
• CAD tools reduced hand-drafting skill but created tacit knowledge about parametric constraints, simulation, manufacturing pipelines.
• Search engines reduced memorized trivia but increased tacit skill in query formulation, source evaluation, triangulation (when taught).

Lesson: Husom is right that automation can amputate learning-by-doing; the historical fix is not “never automate,” but choose what must remain practiced internally and institutionalize that practice.

6) Outsourcing admin work and losing ownership / judgment

His example: if a chatbot does all project admin “perfectly,” you may lose ownership and the basis for high-level decisions.

This is extremely consistent with management history.

• When leaders outsource too much to executive assistants or middle management, they can lose contact with operational reality (“management by memo”).
• When companies outsource analysis to management consultants, they can gain speed and frameworks but risk hollowing out internal judgment: teams stop building the muscle to diagnose their own systems, and decisions become “because the deck said so.”
• When firms outsource customer contact to call centers, they often lose tacit knowledge about customer pain, which later harms product decisions—prompting some companies to force engineers/PMs to do support rotations.

Lesson: “perfect execution” of low-level tasks can still degrade situational awareness. Historically, organizations correct this by designing feedback loops (rotations, reviews, direct exposure) so decision makers keep grounded models.

7) “Valuable experiences”: planning vacations, writing messages, etc.

Husom’s stance is value-laden (“we treat everything as a chore”), but history shows a recurring pattern:

• Technologies that reduce friction (washing machines, prepared foods, travel agents, package holidays) often redefine what counts as the meaningful part of an activity.
• Some people experience that as liberation; others experience it as hollowing out.

There isn’t a single historical verdict—there are stable subcultures:

• Some preserved “slow” craft as identity (handmade goods, handwritten letters, vinyl records).
• Others embraced convenience and reallocated effort elsewhere.

The historical warning that supports him: once convenience becomes the default, social expectations shift (“why didn’t you just…”) and people who want the “valuable experience” may have to actively defend it. That’s a real dynamic.

8) The “extended mind” dispute

Husom calls Masley’s “arbitrary whether it’s neurons or circuits” absurd, and argues remembering a birthday matters because it reinforces relationships.

History gives ammunition to both sides:

Supports Husom

• Outsourcing memory to calendars can reduce spontaneous recall, which is tied to affect and salience. Remembering unprompted signals care; automated messages can feel empty. Social meaning attaches to “costly signals” of attention and effort.

Pushes back on Husom (the strongest counterexample: writing itself)

The oldest “outsourcing thinking” debate is essentially Socrates (via Plato’s Phaedrus) criticizing writing: external memory would weaken internal memory and understanding. Yet literacy enabled:

• cumulative science,
• complex law and governance,
• long-range coordination,
• deeper multi-generational reasoning.

So “externalizing cognition” isn’t arbitrary, but it also isn’t inherently corrupting. The historical lesson is: external memory can weaken some forms of recall while enabling new forms of thought at scale.

A more precise critique of Masley than Husom’s “absurd” would be: it’s not arbitrary; location matters because internal vs external processes produce different incentives, failure modes, and skill formation. But externalization can still be a genuine part of cognition.

9) “Most environments are designed to minimize thinking”

Husom says we’d adapt quickly; only changing environments cause extra thinking.

History suggests a middle view:

• Stable affordances (signage, standardized controls, road rules) do reduce cognitive load and accidents. That’s why they persist.
• But that reduction can create skill fragility: when the environment deviates (unusual airport signage, road closures, tool failures), some people lack the internal model to cope.

So: yes, humans adapt to stable designs—but designs also shape what skills remain practiced. This again matches the deskilling/reskilling pattern.

10) Critique of the “lump of cognition” framing

Husom agrees there’s no fixed pool of thinking, but argues “what we think about matters,” and that replacing boring tasks can remove important formation.

History supports his correction. Division of labor and outsourcing often increase total output while:

• making work more fragmented,
• reducing craft mastery,
• shifting agency upward,
• creating alienation or dependency.

The “lump of labor” analogy is instructive in exactly the way he uses it: new work appears, but it isn’t automatically meaningful or socially beneficial.

Where Husom may overreach is implying a near-universal weakening: historical outcomes depend heavily on:

• whether people still practice the core skill somewhere (training, hobby, rotations),
• whether institutions reward understanding or just output,
• whether the outsourced system is reliable and aligned (safety-critical vs low stakes),
• whether authenticity norms are clear.

A synthesis: what history suggests we should expect with LLMs

If LLMs become ubiquitous, historical outsourcing patterns predict:

1. Text inflation + trust erosion in channels where authorship mattered, until new norms/verification arise.
2. Deskilling in “routine cognition” (drafting, summarizing, first-pass reasoning) unless practice is intentionally preserved.
3. Reskilling and new tacit knowledge (prompting, evaluation, orchestration, taste, domain judgment, provenance checking).
4. Bottleneck migration: as generating words becomes cheap, scarcity shifts to attention, credibility, and demonstrated competence.
5. Institutional responses: training requirements, audits, disclosure norms, and “human-only” rituals for high-trust contexts.

If you want to pick one historical analogue to deep-dive (e.g., GPS, calculators in education, speechwriters/ghostwriting, management consulting, autopilot in aviation, or scribes/writing), I can map it point-by-point to likely LLM outcomes and the mitigations that actually worked.

History suggests software teams will do best with AI code agents if they treat them less like “automatic programmers” and more like a powerful form of outsourcing + automation—which reliably shifts bottlenecks, changes skill formation, and creates new failure modes unless teams add guardrails.

Below are the most consistent historical patterns (from prior “automation waves” like compilers/IDEs/codegen, offshoring/contractors, spreadsheetization of finance, autopilot in aviation, and safety-critical checklists), translated into concrete implications for AI-in-dev.

1) Expect the bottleneck to move: generation gets cheap; verification gets expensive

Pattern: When production becomes easier (printing presses, templates, spreadsheets, code generators), organizations don’t just “do the same work faster”; they produce more, and the limiting factor becomes review, trust, and integration.

In software with agents:

• You’ll likely ship more diffs/PRs, but spend more time on:
• code review quality,
• test adequacy,
• architecture coherence,
• debugging “looks right but isn’t” failures,
• dependency/security review.

What to do:

• Invest early in fast tests, CI, linting, type checking, and reproducible builds.
• Set explicit “definition of done” that makes review cheaper (tests required, no silent behavior changes, performance baselines, threat model notes).

2) Outsourcing works when responsibility stays internal

Pattern: With consultants/contractors, teams succeed when they keep clear internal ownership. When they outsource and relinquish understanding, they get “hollowed out.”

In software with agents:

• If “the agent wrote it” becomes an accepted explanation, you’ll accumulate unowned code and fragile systems.

What to do:

• Require a human owner per change who can explain:
• why it’s correct,
• how it fails,
• how to operate/roll back.
• Treat agent output like code from a new contractor: acceptable only with internal comprehension and supportability.

3) Deskilling is real—so preserve “manual flying” time

Pattern: Aviation automation improved safety but introduced skill decay, so the industry created recurrent training and manual practice.

In software with agents:

• Likely atrophy areas: routine implementation, debugging muscle, reading unfamiliar code, API fluency, mental models of the system.

What to do (pragmatic, not ideological):

• Bake in “manual reps”:
• some tickets/class of work done without agent assistance,
• periodic debugging/incident rotations,
• code-reading sessions.
• Evaluate people on systems understanding, not just throughput.

4) Standardization helps—until it erases local context

Pattern: Templates/forms increase speed but can flatten nuance; organizations respond by introducing structured inputs (forms) plus human judgment gates.

In software with agents:

• Agents are strongest when you can constrain the problem: clear patterns, bounded scope, good examples, stable APIs.
• They are weakest in “high-context” code: tricky concurrency, security boundaries, distributed systems behavior, complex domain invariants.

What to do:

• Start adoption in high-constraint zones:
• scaffolding, migrations with known patterns,
• test generation + refactors,
• internal tooling,
• “boring” glue code behind strong tests.
• Be conservative in high-context zones unless you have excellent specs/tests.

5) When output is cheap, “authenticity” is replaced by provenance

Pattern: When text becomes easy to mass-produce (printing → spam → SEO), systems shift to provenance signals (reputation, signatures, verification).

In software with agents:

• You’ll want to know: what was generated, from which prompt/context, using which model, with which dependencies.

What to do:

• Track provenance in a lightweight way:
• label PRs as AI-assisted,
• keep prompts for non-trivial changes (especially security-sensitive),
• record tool/model versions for audits.
• This isn’t about policing; it’s about later debugging, compliance, and incident response.

6) Quality control must become more formal, not less

Pattern: Mass production drove the invention of QA, standards bodies, audits, checklists.

In software with agents:

• “Looks plausible” code increases the risk of subtle bugs and security issues.

What to do:

• Strengthen gates that are hard for plausible text to fake:
• property-based tests, fuzzing,
• differential testing vs reference implementations,
• static analysis/SAST and dependency scanning,
• performance regression tests,
• threat modeling for sensitive components.
• Use agents to help write these tests, but don’t waive them.

7) Agents change team structure: fewer pure implementers, more integrators

Pattern: With spreadsheets and analytics tools, value moved from arithmetic to modeling, interpretation, and governance.

In software with agents:

• The scarce skills shift toward:
• architecture,
• product-domain modeling,
• writing good specs,
• evaluating tradeoffs,
• debugging and incident handling,
• security and reliability engineering.

What to do:

• Train for “spec → evaluate → integrate” workflows:
• teach people to write crisp acceptance criteria,
• require design notes for non-trivial changes,
• emphasize reading code and reasoning about behavior.

8) The predictable failure mode: local optima and incoherent systems

Pattern: Outsourced piecework (or heavy templating) can optimize each part while degrading the whole; companies respond with stronger architecture governance.

In software with agents:

• You can get a lot of individually “fine” modules that don’t compose well: inconsistent patterns, duplicated logic, leaky abstractions.

What to do:

• Enforce architecture constraints:
• approved patterns, reference implementations,
• “one way to do X” docs,
• library-first reuse,
• periodic refactor budgets.
• Use agents deliberately for refactors after you’ve chosen the target pattern.

9) Rollout lesson: pilots, not big-bang adoption

Pattern: New operational technologies succeed with staged adoption + feedback loops (instead of mandate-first).

What to do:

• Run a 4–8 week pilot with:
• clear use cases,
• baseline metrics (lead time, escaped defects, review time, incident rate),
• qualitative feedback (does it reduce toil or just move it?).
• Expand only where metrics don’t degrade quality/operability.

A simple “historically informed” operating rule

Treat an AI agent as a very fast junior developer with perfect confidence and unknown training data:

• great at drafts and variations,
• weak at truth, context, and long-tail correctness,
• requires strong supervision, tests, and clear standards.

If you tell me your team context (language, size, deployment frequency, regulated or not, and whether you already have strong CI/test coverage), I can propose a concrete adoption policy: which tasks to allow/forbid, what review checklist to add, and what metrics to watch so you catch hollowing-out early.