How Knowledge Work Changes | AI Wars of 2026
For more than a century, office technologies have changed how people find information, coordinate decisions, produce artifacts, and preserve organizational memory. The work did not simply disappear: its cognitive burden moved.
Title: How Knowledge Work Changes
Eyebrow: Work, technology, and cognitive investment
Published: 2026-08-30T06:35:38-04:00
Updated: 2026-09-28T10:58:09-04:00
Lede: For more than a century, office technologies have changed how people find information, coordinate decisions, produce artifacts, and preserve organizational memory. The work did not simply disappear: its cognitive burden moved.
Big Idea: As routine cognitive friction becomes cheaper, a high-performing worker can invest more attention in craft, critical thinking, learning, judgment, relationships, and accountable outcomes.
Evidence Boundary: This is a qualitative history of work systems, not a productivity time series. A milestone records a documented change in work practice; it does not establish universal adoption, a single causal productivity effect, or the elimination of an occupation.
Milestones
Date: 1850s-1880s
Phase: Work is formalized
Title: Managerial bureaucracy turns local knowledge into organizational reports
Work Change: Large, distributed enterprises built departmental hierarchies, formal accounting, schedules, standardized forms, and recurring operational reports.
Cognitive Shift: Information moved from personal recollection into procedures and reports, giving managers visibility across distance while creating a distinct administrative profession.
Caveat: This pattern is most clearly documented in large U.S. enterprises; it did not describe every small firm or professional practice.
Date: 1874-1910s
Phase: Organizational memory scales
Title: Typewriting and filing make office memory reproducible and retrievable
Work Change: Standardized correspondence, carbon copies, and indexed files made records easier to produce, retain, and retrieve at scale.
Cognitive Shift: Composition separated from transcription, while files externalized knowledge that had depended on a particular clerk's memory.
Caveat: The technology did not simply remove clerical work; clerical employment expanded and its gendered division of labor carried unequal status and autonomy.
Date: 1876-1920s
Phase: Coordination becomes synchronous
Title: The telephone compresses the time needed to coordinate
Work Change: Offices could clarify orders, negotiate, check availability, and coordinate distant branches by live voice instead of waiting for correspondence or travel.
Cognitive Shift: More contextual judgment could be exchanged immediately, while availability, interruption, and weaker written records became part of office work.
Caveat: Faster communication is not a universal firm-level productivity measure, and access remained uneven by region and social group.
Date: 1951-1960s
Phase: Calculation centralizes
Title: Mainframes centralize administrative calculation
Work Change: Payroll, billing, inventory, accounting, and reservation work migrated from ledgers and punched cards into centrally operated electronic data-processing systems.
Cognitive Shift: Administrative rules and institutional memory moved toward programmers, systems analysts, and data-processing departments.
Caveat: Early systems were costly, batch-oriented, and limited in deployment; they did not make every office process immediate or automated.
Date: 1971-1980s
Phase: Revision gets cheaper
Title: Word processing separates text from the physical page
Work Change: Stored text could be corrected, rearranged, reused, and reprinted without retyping an entire document.
Cognitive Shift: Revision became cheaper and authors increasingly composed directly, though some organizations centralized the work in specialist pools.
Caveat: Lower editing cost can also create more drafts, formatting work, electronic filing, and support tasks.
Date: 1979-1990s
Phase: Modeling decentralizes
Title: Spreadsheets move modeling and scenario testing to the desk
Work Change: Interactive spreadsheets let analysts and managers alter an assumption and immediately recalculate dependent cells.
Cognitive Shift: More people could build and test models without submitting each request to a central programming department.
Caveat: End-user computing also introduced formula errors, inconsistent versions, weak documentation, and shadow information systems.
Date: 1971-1995
Phase: Coordination becomes asynchronous
Title: Network email expands who can initiate and receive work
Work Change: Store-and-forward messaging, replies, forwarding, and distribution lists enabled collaboration across distance and time.
Cognitive Shift: Communication could bypass secretarial and hierarchical routing, shifting more filtering, prioritization, and record-keeping work to recipients.
Caveat: Email added information that would otherwise not be exchanged, but also introduced overload, reduced social cues, and fragmented systems.
Date: 1995-1998
Phase: Information becomes on-demand
Title: The searchable web lowers the cost of finding reference information
Work Change: Browsers and ranked search shifted research away from libraries, directories, intermediaries, and personal files toward on-demand retrieval.
Cognitive Shift: The work of locating information became faster in many cases, while assessing source quality and relevance became a larger part of the task.
Caveat: A controlled search experiment studied discrete reference questions after search engines had matured; it is not a measure of long-run workplace output.
Date: 2007-2014
Phase: Responsiveness detaches from place
Title: Smartphones make communication and coordination continuously available
Work Change: Push email and mobile internet made it normal to read, respond, schedule, and coordinate outside a fixed office or desktop.
Cognitive Shift: Workers gained flexibility and reach, but also faced a greater expectation of continual accessibility and interruptions.
Caveat: Self-reported productivity gains coexist with longer reported hours; speed and flexibility do not capture stress, effort, or time pressure.
Date: 2011-2019
Phase: Documents become shared systems
Title: Cloud workspaces make documents, calendars, and conversations jointly editable
Work Change: Shared documents, calendars, chat, conferencing, and version history moved coordination away from local copies and attachments.
Cognitive Shift: Context, handoffs, and coauthoring became persistent team assets rather than exchanges between individual inboxes and desktops.
Caveat: There is limited clean causal evidence that cloud coauthoring itself raises firm productivity; implementation and organizational redesign remain central.
Date: 2020-2022
Phase: Distributed work becomes the operating model
Title: Remote collaboration moves coordination, supervision, and onboarding into digital channels
Work Change: The pandemic forced meetings, informal coordination, document work, and management routines into cloud services at organizational scale.
Cognitive Shift: Place and synchronous presence became less necessary for many tasks, while maintaining cross-team ties and informal knowledge transfer became harder.
Caveat: Remote work effects vary by job design; measured output, employee experience, promotion, mentoring, and long-run innovation are separate outcomes.
Date: 2022-2023
Phase: First drafts become conversational
Title: Generative AI turns drafting, summarizing, and explanation into natural-language requests
Work Change: Workers can ask a general-purpose system for drafts, summaries, code, analyses, and explanations instead of only retrieving or manipulating existing information.
Cognitive Shift: The worker can move earlier into problem framing, verification, integration, and accountable use of the produced artifact.
Caveat: Strong task-level results on professional writing do not establish sustained job performance, fact-checking burden, firm-level productivity, or displacement.
Date: 2024-present
Phase: Work is increasingly delegated
Title: Agentic systems make supervised multi-step delegation a work-design question
Work Change: People can increasingly assign multi-step research, coding, and coordination tasks, then review outputs, exceptions, and actions.
Cognitive Shift: The proposed move is from producing every intermediate artifact toward specifying, supervising, integrating, and remaining accountable for delegated work.
Caveat: This remains an emerging and uneven pattern. Capability benchmarks, vendor telemetry, and task-level experiments do not establish reliable autonomous work or economy-wide productivity.
Source 25 Source 26 Source 27 Source 28
Future Scenarios
Title: Delegation and stewardship
Description: Workers increasingly specify goals, evaluate outputs, integrate evidence, document decisions, and remain accountable for delegated work.
Evidence Boundary: Plausible where task-level generation improves but reliability remains uneven. It would weaken if representative longitudinal evidence finds no sustained increase in supervision, verification, or AI-integration work.
Title: Expertise compression
Description: AI may help less-experienced workers access codified expert practices faster and may allow individuals to perform work that formerly required larger teams.
Evidence Boundary: Evidence is strongest in specific firms and tasks. It does not tell us whether workers build durable expertise or lose opportunities to acquire it.
Title: Persistent hybrid work
Description: Generation becomes cheaper while domain judgment, oversight, interpersonal coordination, and responsibility remain central to consequential work.
Evidence Boundary: This follows from the uneven performance observed inside and outside a model's capability frontier; it is not evidence that humans will always retain the same task mix.
Title: Slow and uneven transformation
Description: Workflow integration, cost, skills, governance, trust, and local institutional conditions may keep realized labor effects far below technical exposure estimates.
Evidence Boundary: Exposure is not displacement. Current administrative evidence identifies task reorganization before measurable changes in recorded earnings or hours.
Profession Comparison
Eyebrow: A qualitative occupation comparison
Title: The workday changed differently by profession
Description: The same office technology changed each profession's work differently. Developer workflows are consequential, but only one of several occupational stories; these are plausible concentrations of time or cognitive effort, not measured workday shares.
Evidence Legend
Marker: D
Label: Documented duty or work system
Marker: D/I
Label: Documented evidence with bounded inference
Marker: I
Label: Bounded editorial inference
Marker: S
Label: Future scenario, not a prediction
Professions
Name: Lawyer
Source 32 Source 33 Source 34 Source 26
Assessments
Evidence: I
Focus: Printed authorities, correspondence, and document preparation; interpretation and advocacy.
Evidence: D/I
Focus: Research, writing, advice, and representation; faster paperwork.
Evidence: I
Focus: Electronic research, drafting, revision, and case strategy.
Evidence: D/I
Focus: Online research, filings, and shared documents; precedent analysis.
Evidence: I
Focus: AI-assisted research and drafts; authority, confidentiality, and strategic review.
Evidence: S
Focus: Bounded discovery and drafting workflows; human judgment, negotiation, and liability.
Name: Accountant / auditor
Source 32 Source 35 Source 36 Source 31
Assessments
Evidence: I
Focus: Ledgers and reconciliation; discrepancy detection and assurance.
Evidence: D/I
Focus: Record review, auditing, tax and cost work; client advice.
Evidence: I
Focus: Spreadsheets, controls, exceptions, and financial analysis.
Evidence: D/I
Focus: Digital records, audit evidence, compliance, and management advice.
Evidence: I
Focus: Extraction, research, and draft support; provenance, skepticism, and sign-off.
Evidence: S
Focus: Continuous controls and exceptions; human materiality and fraud judgment.
Name: Marketing / communications
Source 32 Source 37 Source 30 Source 25
Assessments
Evidence: I
Focus: Print copy and artwork, media relationships, and direct mail; audience judgment.
Evidence: D/I
Focus: Research, copy, layout, client coordination, and media buying.
Evidence: I
Focus: Segmentation, production iteration, and database-led analysis.
Evidence: D/I
Focus: Digital content, channel coordination, and audience and performance analysis.
Evidence: I
Focus: Ideation, copy variants, and analysis; positioning, brand, factual, and IP review.
Evidence: S
Focus: Bounded content and testing cycles; people set strategy and consent rules.
Name: Manager / project manager
Source 1 Source 32 Source 38 Source 26
Assessments
Evidence: I
Focus: Supervision, correspondence, work allocation, and exception resolution.
Evidence: D/I
Focus: Planning, budgets, coordination, reports, and meetings.
Evidence: I
Focus: Schedules, budgets, dependencies, and status reporting.
Evidence: D/I
Focus: Stakeholder communication, coordination, documentation, and issue resolution.
Evidence: I
Focus: Summaries, plans, and risk prompts; goals, trade-offs, and accountability.
Evidence: S
Focus: Agents maintain plans and chase updates; humans align work and make decisions.
Name: Analyst / researcher
Source 6 Source 10 Source 39 Source 27
Assessments
Evidence: I
Focus: Collection, hand classification and tabulation, and reports; question framing.
Evidence: D/I
Focus: Data collection, punched-card and statistical processing, analysis, and reports.
Evidence: I
Focus: Data work, coding, model construction, and larger-dataset interpretation.
Evidence: D/I
Focus: Data validation, modeling, literature review, testing, and reporting.
Evidence: I
Focus: Search, code, query, and draft support; question definition and validation.
Evidence: S
Focus: Reproducible research pipelines; humans govern questions, evidence, and interpretation.
Name: Software engineer
Source 6 Source 7 Source 40 Source 41 Source 28
Assessments
Evidence: I
Focus: The modern occupation did not yet exist; related work centered on mechanical and electrical systems, calculation, and documentation.
Evidence: I
Focus: Early electronic computing introduced program planning, coding, debugging, and documentation.
Evidence: I
Focus: Programming, debugging, testing, systems integration, and deployment.
Evidence: D/I
Focus: Collaborative code, distributed systems, web and cloud delivery, and production operations.
Evidence: D/I
Focus: AI-assisted coding and review; specification, testing, integration, security, and ownership.
Evidence: S
Focus: Bounded coding agents implement and test; engineers set architecture, evaluate, secure, integrate, and operate systems.
Questions
Is AI removing tasks, changing how people perform them, or removing whole jobs?
Who verifies the verifier, and who remains accountable when delegated work fails?
Does AI help people build expertise, or remove the experiences through which expertise develops?
Do measured productivity gains become higher output, better quality, shorter hours, lower staffing, or something else?
What evidence would make us revise the delegation-and-stewardship story?
Title: For the worker, the task changes before the job does
Description: As technology makes routine cognitive friction cheaper, the scarce contribution of a high-performing worker moves toward craft, critical thinking, learning, judgment, relationships, and accountable outcomes.
Takeaway: Technology changes where a worker can create the most value: less time handling information, more time framing, creating, learning, judging, and owning the outcome.
Note: Read each row as a qualitative shift in how work is performed. The Future of Work column summarizes research-bounded scenarios, not a prediction. Equal widths do not represent labor share, cognitive load, productivity, adoption, or the value of one activity over another.
Eras
Date: 1850-1910
Label: Paper office
Technology: Type, file, phone
Date: 1910-1960
Label: Mechanized
Technology: Power, telephone
Date: 1960-1995
Label: Computerized
Technology: Mainframes, word, spreadsheet
Date: 1995-2022
Label: Networked
Technology: Search, mobile, cloud
Date: 2022+ GenAI; 2024+ agents
Label: GenAI + emerging agents
Technology: AI assistance; supervised agents
Date: Research-bounded
Label: Possible futures of work
Technology: Possible delegation; human oversight
Is Future: true
Activities
Label: Education and learning
Shift: Workers build and refresh expertise through practice, instruction, feedback, and access to knowledge.
Stages
Apprenticeship
Formal training
Computer learning
Online learning
AI-assisted learning
AI-supported practice
Label: Routine cognitive friction
Shift: Copying, filing, calculation, and scheduling become progressively cheaper to execute.
Stages
Copy, file, calculate
Standardize
Spreadsheet and email
Search and workflow
AI-assisted first pass
Supervised task delegation
Label: Craft and creation
Shift: Workers use each generation of tools to investigate, compose, design, explain, and solve problems in new ways.
Stages
Compose and explain
Specialist tools
Digital creation
Coauthor
AI-assisted creation
Human-AI creation
Label: Critical thinking
Shift: Workers evaluate evidence, challenge assumptions, and test the quality of answers before acting on them.
Stages
Question claims
Compare evidence
Test assumptions
Filter information
Verify AI outputs
Evidence stewardship
Label: Judgment and accountability
Shift: The worker remains responsible for framing the problem, deciding what matters, checking work, and owning outcomes.
Stages
Prioritize and decide
Interpret outputs
Validate models
Integrate context
Frame and verify outputs
Human direction and review
Source references
Source 1: The Railroads: The First Modern Business Enterprises
Cambridge University Press
Published: 2010
Alfred Chandler's account of nineteenth-century railroads and modern business enterprise. It supports the organizational history of formal management and reporting, not a universal model for every firm or profession.
Source 2: Underwood Model 5 typewriter
Smithsonian National Museum of American History
Published: 1874
Museum record documenting a prominent early typewriter model. It supports the material history of standardized office correspondence, not a quantitative productivity claim.
Source 3: Woodruff File Cabinet
Smithsonian National Museum of American History
Published: 1900
Museum record for office filing technology. It supports the historical shift toward retrievable organizational records, not a complete account of information-management practice.
Source 4: Bell experimental telephone and 1876 patent
Smithsonian National Museum of American History
Published: 1876
Museum record for Bell's early telephone. It establishes the technology's historical availability, not a universal productivity or communication-quality effect.
Source 5: America Calling: A Social History of the Telephone to 1940
University of California Press
Published: 1992
Claude Fischer's social history cautions against technological determinism in accounts of telephone adoption and use.
Source 6: UNIVAC
Computer History Museum
Published: 1951
Museum history of UNIVAC and early electronic data processing. It documents early capability and limited deployment, not universal office automation.
Source 7: System/360
IBM
Published: 1964
IBM's historical record for System/360. It supports a product-history claim about a compatible computer family, not a general measure of organizational productivity.
Source 8: Evelyn Berezin and the Data Secretary
National Inventors Hall of Fame
Published: 1971
Biographical record of the Data Secretary word-processing system. It supports the historical introduction of stored-text editing, not a universal labor-market effect.
Source 9: Automation of America's Offices
U.S. Office of Technology Assessment
Published: 1985
U.S. Office of Technology Assessment report examining office automation. It emphasizes that work outcomes depend on organizational and managerial choices as well as equipment.
Source 10: VisiCalc
Computer History Museum
Published: 1979
Museum history of VisiCalc. It supports the historical shift to interactive spreadsheet modeling, not a measured economy-wide productivity change.
Source 11: The Productivity Paradox of Information Technology
Communications of the ACM
Published: 1993
Erik Brynjolfsson's analysis of the information-technology productivity paradox. It identifies measurement and organizational lags as competing explanations, not proof that IT lacked value.
Source 12: A Brief History of the Internet
Internet Society
Published: 1999
Internet Society history of networked communication and internet protocols. It supports broad chronology, not a measured workplace impact.
Source 13: Reducing Social Context Cues: Electronic Mail in Organizational Communication
Management Science
Published: 1986
Field study of electronic mail in a Fortune 500 organization. It found email changed the amount and social pattern of communication; it does not establish universal organizational effects.
Source 14: Our story
Published: 1998
Google's company history documents its origins and incorporation. It establishes chronology, not the causal impact of web search on knowledge work.
Source 15: A day without a search engine: an experimental study of online and offline searches
Experimental Economics
Published: 2014
Controlled comparison of online and offline searches on discrete reference questions. It supports task-level differences in search speed and answers, not a long-run workplace productivity estimate.
Source 16: Apple Reinvents the Phone with iPhone
Apple Newsroom
Published: 2007-01-09
Apple's product announcement establishes the iPhone's availability and intended use, not a causal effect on work practices.
Source 17: Technology's Impact on Workers
Pew Research Center
Published: 2014-12-30
U.S. survey of employed internet users on technology at work. It reports perceived productivity and flexibility alongside longer reported hours; the measures are self-reported.
Source 18: The Autonomy Paradox: The Implications of Mobile Email Devices for Knowledge Professionals
Organization Science
Published: 2013
Study of mobile-email-using professionals describing how flexibility and control can coexist with expectations of continual accessibility. It does not establish the effect for every profession or device user.
Source 19: Microsoft Launches Office 365 Globally
Microsoft News
Published: 2011-06-28
Microsoft's launch announcement documents intended cloud collaboration features. It is evidence of product availability, not independent evidence of adoption or productivity.
Source 20: Information, Technology, and Information Worker Productivity
Information Systems Research
Published: 2012
Study of information-worker networks and project performance in an executive-search firm. It supports context-dependent effects of IT-enabled networks, not a general cloud-collaboration estimate.
Source 21: Does Working from Home Work? Evidence from a Chinese Experiment
Quarterly Journal of Economics
Published: 2015
Randomized experiment among Ctrip call-center workers. Its measured performance and attrition effects apply to a standardized setting, not all knowledge work.
Source 22: The effects of remote work on collaboration among information workers
Nature Human Behaviour
Published: 2022
Difference-in-differences analysis of more than 61,000 Microsoft information workers during pandemic remote work. It documents network changes at one company, not universal effects on output or innovation.
Source 23: Introducing ChatGPT
OpenAI
Published: 2022-11-30
OpenAI's announcement of ChatGPT as a dialogue model made available for public feedback. It is a product announcement, not an independent measure of capability or adoption.
Source 24: Experimental evidence on the productivity effects of generative artificial intelligence
Science
Published: 2023
Preregistered experiment with 453 college-educated professionals on occupation-specific writing tasks. It measures short task completion and rated quality, not sustained job performance or firm outcomes.
Source 25: The Unequal Adoption of ChatGPT Exacerbates Existing Inequalities Among Workers
Proceedings of the National Academy of Sciences
Published: 2025
Preregistered survey and experiment of 18,000 Danish workers in 11 ChatGPT-exposed occupations, linked to administrative data. It finds early work use alongside employer-policy and training barriers; it does not distinguish ChatGPT plans or measure enterprise adoption, retention, or ROI.
Source 26: The Impact of Generative AI on Critical Thinking: Self-Reported Reductions in Cognitive Effort and Confidence Effects From a Survey of Knowledge Workers
ACM CHI
Published: 2025
Survey of 319 knowledge workers on generative-AI use and critical-thinking effort. It reports perceived shifts toward verification, response integration, and task stewardship; self-reported associations do not establish a universal causal effect.
Source 27: Navigating the Jagged Technological Frontier
Social Science Research Network
Published: 2023
Experiment with consultants showing benefits on tasks within a model's capability frontier and worse performance on a task outside it. It supports conditional performance, not a universal productivity result.
Source 28: Update on our early 2026 AI productivity study
METR
Published: 2026-02-24
METR's continuation of its developer-productivity research. Early-2026 results suggest developers may be faster with newer AI tools, but METR reports severe selection effects and a confidence interval too wide for a reliable point estimate.
Source 29: Generative AI at Work
Quarterly Journal of Economics
Published: 2025
Field evidence from customer-support work showing heterogeneous performance and learning effects after AI adoption. It does not establish results for all knowledge workers or autonomous agents.
Source 30: The Cybernetic Teammate: A Field Experiment on Generative AI Reshaping Teamwork and Expertise
Organization Science
Published: 2026
Field experiment with Procter & Gamble professionals on AI-assisted product-development work. It supports a bounded finding about task performance and team configuration, not universal worker substitution.
Source 31: Generative AI and Jobs: A Refined Global Index of Occupational Exposure
International Labour Organization
Published: 2025
ILO and NASK analysis of occupational exposure to generative AI. Exposure is an upper-bound task estimate under implementation assumptions, not a forecast of job displacement.
Source 32: Occupational Outlook Handbook, 1961 Edition
U.S. Bureau of Labor Statistics
Published: 1961
Occupational descriptions for lawyers, accountants, advertising workers, and business administrators. It documents mid-century duties and working conditions, not comparable workday time shares or a universal occupational experience.
Source 33: Lawyers
National Center for O*NET Development
Published: n.d.
Current U.S. occupational profile for lawyers. It describes reported tasks and work context, not historical practice, universal job design, or a time-use allocation.
Source 34: ABA Issues First Ethics Guidance on a Lawyer's Use of AI Tools
American Bar Association
Published: 2024-07-29
ABA reporting on Formal Opinion 512 and lawyers' duties of competence, confidentiality, communication, supervision, and fees when using generative AI. It establishes professional guidance, not adoption levels or job-wide task change.
Source 35: Accountants and Auditors
National Center for O*NET Development
Published: n.d.
Current U.S. occupational profile for accountants and auditors. It describes reported tasks and work context, not historical practice, universal job design, or a time-use allocation.
Source 36: Staff Update on Outreach Activities Related to the Integration of Generative Artificial Intelligence in Audits and Financial Reporting
Public Company Accounting Oversight Board
Published: 2024-07
PCAOB staff update on generative-AI use in audit and financial reporting. It describes developing, limited use and related considerations; it does not establish autonomous audits or profession-wide adoption.
Source 37: Search Marketing Strategists
National Center for O*NET Development
Published: n.d.
Current U.S. occupational profile for a marketing specialty. It describes reported tasks and work context, not the full marketing and communications profession, historical practice, or a time-use allocation.
Source 38: Project Management Specialists
National Center for O*NET Development
Published: n.d.
Current U.S. occupational profile for project management specialists. It describes reported tasks and work context, not all management work, historical practice, or a time-use allocation.
Source 39: Operations Research Analysts
National Center for O*NET Development
Published: n.d.
Current U.S. occupational profile for operations research analysts. It is a proxy for one analyst occupation, not the full history or time use of research work.
Source 40: Software Developers
National Center for O*NET Development
Published: n.d.
Current U.S. occupational profile for software developers. It describes reported tasks and work context, not the historical evolution of the occupation, universal job design, or a time-use allocation.
Source 41: How Claude Code is used in practice
Anthropic
Published: 2026-06-16
Anthropic research based on approximately 400,000 Claude Code sessions, focused on agentic coding and human-AI collaboration.