Member
November 4, 2026

Knowledge Organizations in Five Domains In Next Five Years

A SRIJAN SANCHAR FORESIGHT REPORT

Knowledge Organizations in Five Domains, 2026–2031

How design, law, consulting, education and healthcare will change what they sell, to whom, and on what terms

4 November 2026

Contents

Part 1. Foresights

Part 2. Introduction

Part 3. Purpose

Part 4. Methodology and process

Part 5. Challenges

Part 6. Caveats

Part 7. Way forward

Annex. Sources

Part 1 is written to be read on its own. Parts 2 to 7 explain why the foresight was made, how it was made, how far to trust it, and what to do next.

Part 1. Foresights

1.1 The foresight in brief

Knowledge organizations sell scarce expertise. AI is making routine expertise abundant, so the scarce thing becomes trust: who stands behind the answer, and who is still being trained to give it. Across design, law, consulting, education and healthcare the pattern is shared, though its pace depends on how strongly regulation and liability protect the human.

By 2031, execution becomes cheap and judgment becomes the premium. Clients and patients stop paying for time. They want fixed prices, proof of impact, verified skills and attention rather than screens. Law and consulting are already moving: McKinsey reportedly ties about a quarter of fees to outcomes, and clients are writing alternative fees into legal budgets. Healthcare adopts ambient notes and AI-read scans first, while diagnosis stays human. Design moves to subscriptions and outcome-linked creative work. Education shifts toward verified skills and stackable micro-credentials.

New business models follow: flat fees, subscriptions, outcome-linked fees, living design systems, AI tutors with employer alliances, and AI-run service lines under accountable professionals. The weakest link is the apprenticeship pipeline, because routine work once trained juniors and AI now does it. Winners will re-price commodity work, productise accountability and rebuild apprenticeships deliberately. Those who wait will compete on price alone.

1.2 The shape of 2031

In every domain the profession splits into three layers. The speed of the split differs, but its shape does not.

Layer	What it contains	How it is sold
Accountable autopilot	Routine, measurable work: contract review, production design, analysis and benchmarking, standard assessment, documentation and coding. AI-run, with a named professional on exceptions and sign-off.	Per outcome, fixed price or subscription
Judgment layer	Strategy, negotiation, brand direction, mentoring, diagnosis and treatment decisions, regulated sign-off, change management.	Premium advisory, retainers, performance partnerships
Squeezed middle	Routine work still sold by the hour, seat or visit, with no automation advantage and no accountability premium.	Hours, seats, visits. Deflating each year

How fast each domain moves

Domain	Pace of change	Why
Law	Fast	Output is easy to price; corporate clients hold power and have their own AI capacity; but licence and liability still protect sign-off.
Consulting	Fast	No licence protects the work; CFOs demand proof of return; outcome-linked fees already exist.
Design	Fast for production, slower for brand	Production output is easy to measure and AI-made; brand and direction rely on taste and trust.
Education	Medium	Employers push skills-based hiring quickly; accreditation and funding move slowly.
Healthcare	Fast for paperwork, slow for care	Documentation and imaging automate readily; payers and regulators fix the unit of sale and the liability.

1.3 The garden of knowledge

Five houses of expertise around a common tree. The pattern at its centre: sell the outcome, guard the judgment, grow the apprentices.

1.4 Domain foresights

For each domain: the evidence now, the outcome by 2031, how stakeholder needs change, the new business models, and first moves. Confidence refers to the direction of change, not its pace.

Design

Question	Answer
Evidence now	About 75% of designers use AI tools, up from 35% in 2023 (AIGA figure via an aggregator). Adobe’s 2025 survey says 90% of creative professionals save time on menial tasks. Figma launched an AI design agent on 20 May 2026. One report says the launch of Anthropic’s Claude Design on 18 April 2026 knocked Figma’s shares by 7% (single source). “Unlimited design” subscription services are reported to serve 100,000+ businesses.
Outcome by 2031	Layouts, variants and prototypes are close to free. Value sits in briefing, taste, research, brand coherence and integration with code and operations. Mid-size generalist agencies are squeezed; one-person studios and domain-deep agencies do well. Design tools turn into design agents that build working product, not pictures of it.
Stakeholder needs	Marketers and product teams want working output inside their own systems and templates, in hours, with clear ownership and provenance of AI-made assets and brand safety. Designers want tools that keep authorship.
New business models	Design-as-a-subscription with unlimited variants; performance fees tied to campaign results; living design systems kept on retainer; vertical specialist studios; a creative-direction premium; studios that also run their own AI tooling.
First moves	Price production work by subscription or outcome; sell the design system rather than the artwork; invest in research, taste and brand governance; train juniors as directors of AI.
Confidence in direction	Fairly high

Law

Question	Answer
Evidence now	The billable hour was about 80% of law-firm fee income, per a 2025 Harvard Center on the Legal Profession report as cited by LexisNexis. Legora’s CEO says general counsel are acting on alternative fees for the first time (April 2026). A Cleary Gottlieb senior partner predicts a shift to speed- and outcome-based pricing. About 85% of corporate legal teams reportedly have AI resources and use them to push for alternative fees and pull work in-house. 46 Am Law 200 lawyers moved to AI companies in H1 2026. Commentators stress that change will be incremental.
Outcome by 2031	Routine drafting, review and diligence fall from hours to minutes. Firms keep strategy, negotiation, litigation judgment and regulated sign-off. The pyramid becomes a diamond. Some routine work moves in-house; AI-native legal-service firms take standardised matters.
Stakeholder needs	Corporate clients want fixed prices, speed and predictable risk, tighter panels, confidentiality and audit trails, and explainable AI-assisted advice.
New business models	Flat and capped fees, portfolio subscriptions, success fees; productised services such as contract review; “legal engineer” roles; AI-native law firms; joint ventures with legal AI vendors; licensing of firm knowledge assets.
First moves	Price standard matters as products; point associate time at judgment and client context; run apprenticeship through supervised AI workflows; warrant outputs and insure them.
Confidence in direction	Fairly high

Consulting

Question	Answer
Evidence now	McKinsey reportedly runs about 25,000 AI agents beside about 40,000 people, down from 45,000 in 2023, and earns about a quarter of fees from outcome-based work. Its leadership describes underwriting outcomes by tying fees to client impact. PwC US reportedly planned to cut graduate hiring by a third, and Big Four starting pay has stagnated. Commentators note that no firm has standardised what an outcome is worth.
Outcome by 2031	Analysis, benchmarking and deck production are abundant. Scarce: problem framing, change management, access and accountability for results. Pyramids shrink to diamonds; specialist and AI-native boutiques take share; advisory moves toward performance partnerships.
Stakeholder needs	CFOs demand proof of return, implementation rather than advice, outcome guarantees, speed, and visibility into agent-produced work.
New business models	Gain-share and hybrid fixed-plus-success fees; consulting assets (agent platforms and diagnostics) sold as subscriptions; embedded delivery teams; fees partly in equity or performance; productised diagnostics.
First moves	Write outcome contracts with baselines and an agreed adjudicator; build reusable agent assets; hire people who combine domain and AI-integration skill; protect apprentices through rotation.
Confidence in direction	Fairly high

Education

Question	Answer
Evidence now	Coursera and Rep Data (Feb–Mar 2026, a vendor survey): 98% of employers use skills-based hiring in some form, 86% extensively for entry-level roles, and every surveyed Indian employer uses it. Employers expect 39% of key skills to change by 2030 (a World Economic Forum figure cited by Coursera). UPCEA expects colleges to fold continuing education into core strategy, embed micro-credentials in degrees and treat AI as core infrastructure in 2026.
Outcome by 2031	Content delivery and routine assessment commoditise; a personal AI tutor is expected. Degrees remain but unbundle into stackable credentials. Value moves to mentoring, cohort, supervised practice, validation of skills and employer pathways. Institutions that cannot prove outcomes lose enrolment.
Stakeholder needs	Learners want verified, job-relevant skills, quickly, affordably and flexibly. Employers want proof of applied skill. Working adults want continuous reskilling. Everyone wants trustworthy assessment.
New business models	Credit-bearing stackable micro-credentials; employer-designed pathways; AI tutoring built into tuition; outcome-linked or income-share tuition; lifelong-learning subscriptions; assessment as a service; apprenticeships paired with study.
First moves	Map each course to demonstrable skills and assessments; embed micro-credentials; pilot AI tutors under mentor oversight; contract with employers around outcomes.
Confidence in direction	Fairly high

Healthcare

Question	Answer
Evidence now	About two-thirds of Epic hospitals use AI documentation (Q1 2026 round-up). Spending on ambient scribes grew 2.4 times in 2025 in a survey of 700+ executives. The US Veterans Affairs department planned a national rollout across 170 medical centres in 2026. Where scribes are offered to all clinicians, sustained use often settles at 20–50%. In a Stanford survey, 73.6% of 2,202 patients who experienced it wanted it used again. More than 1,000 FDA-cleared AI devices are in radiology. Adoption is concentrated in documentation, imaging and coding, with humans responsible for diagnosis and treatment. MIT Technology Review says it is not yet known whether this helps patients.
Outcome by 2031	Documentation, coding, authorisation and imaging triage are largely machine-run under clinician sign-off. Visits spend more time on the patient. An AI second reader is standard in imaging. Patients arrive pre-informed by chatbots. Whether freed capacity becomes better access depends on pay and liability structures.
Stakeholder needs	Patients want attention, speedy access and clear explanation. Clinicians want less paperwork. Payers want proof of outcomes and lower cost. Regulators want safety, consent and data protection.
New business models	Ambient documentation as standard infrastructure; value-based or outcome-linked payment for AI-enabled care pathways; AI-supported nurse-led and remote care; subscription primary care; AI-read imaging as a service; automated revenue cycle; consent and data stewardship as a trust asset.
First moves	Measure patient outcomes and time-to-care, not only note time; govern consent; turn AI savings into access rather than only cost cuts; keep clinician sign-off accountable.
Confidence in direction	Fairly high

1.5 The road to 2031

Horizon	What to expect across the five domains	Confidence
2027	Pressure year. Alternative fees and outcome pilots spread in law and consulting; graduate intake stays subdued; ambient clinical documentation widens; employers hire more by demonstrated skill; design teams adopt AI design agents.	Fairly high
2028	Re-pricing year. Fixed, subscription and outcome-linked terms become normal for routine work in law, consulting and design; AI-native firms reach visible scale in legal and advisory work; credit-bearing micro-credentials spread in universities.	Medium
2029	Apprenticeship reset. Firms that cut junior intake meet a shortage of experienced judgment; structured, paid apprenticeships return in new form; payers begin to link AI-enabled care pathways to outcomes.	Medium
2030–31	Split professions. Routine work is machine-run under accountable sign-off and sold per outcome; a premium judgment layer remains human-led; education is stackable and employer-verified; healthcare spends clinician time on patients rather than paperwork.	Medium

1.6 How knowledge organizations will respond

Six responses recur across the domains. Each carries a first step for the next 90 days and the main risk to guard against.

#	Response	First 90 days	Main risk and guard
1	Unbundle and re-price. Split services into a routine layer (fixed, subscription or outcome terms) and a judgment layer (premium advisory).	List services; mark which are routine and measurable; set a price per unit; offer a dual quote on the next two proposals.	Fixed-price overrun. Guard: scoped definitions, volume bands, gain-share.
2	Build the AI-run line. Turn playbooks into agent workflows with an exception queue and a named accountable professional.	Pick one service; measure auto-resolution, quality and cost per unit weekly.	Quality drift. Guard: sampling audits, escalation rules.
3	Rebuild apprenticeship. Paid, supervised practice where juniors review and correct AI output, plus simulation.	Map vanishing junior tasks; design a 12-month programme; set an intake floor.	Short-term cost cuts starve later judgment. Guard: minimum intake, senior mentoring time.
4	Productise accountability. Warranties, attestations, audit trails and consent records.	Define what the organization will stand behind in writing; agree insurance.	Liability exposure. Guard: scoped warranties, cover.
5	Own the client’s context. Matter databases, design systems, learner records, clinical pathways kept current.	Choose one client or cohort; build the living record.	Lock-in backlash. Guard: portability and clear data terms.
6	Choose a posture toward AI suppliers. Co-builder or accountable integrator, not a reseller.	Map supplier moves into your clients; negotiate terms; design for more than one supplier.	Dependence. Guard: multi-supplier architecture.

A three-phase path

Phase	Period	Moves	Test of progress
1. Wedge	Nov 2026 – Dec 2027	Choose two or three routine, measurable services; offer fixed, subscription or outcome terms alongside hourly quotes; capture workflows as AI-run lines; redesign junior roles.	At least two contracts on non-hourly terms at gross margin at or above the hourly baseline.
2. Scale	2028 – 2029	Move renewals to new terms; run structured apprenticeships; partner or acquire AI-native capability; publish outcome results.	Most routine work offered on non-hourly terms; junior pipeline stable.
3. Crown	2030 – 2031	Productise accountable sign-off and the judgment premium; keep client context as the asset.	Margin per professional and trust scores above the 2026 baseline after deflation.

1.7 What to watch

What to look for	Review
Share of law-firm fees that are non-hourly, and the share of corporate matters on alternative fees	Jan 2027, then annually
Share of consulting fees tied to outcomes at McKinsey and the Big Four	Nov 2027, Nov 2028
Graduate and entry-level intake at Big Four firms, large law firms and consultancies	Jan 2027, then annually
Employer and university acceptance of credit-bearing micro-credentials and skills-based admission	Jul 2027, Jul 2028
Sustained clinician use of ambient scribes and measured effect on patient access and outcomes	Nov 2027, Nov 2029
Share of design production bought as subscription or outcome-linked work, and designer entry-level hiring	Nov 2027, Nov 2028
Regulator rulings on AI supervision and liability in law, medicine, audit and assessment	Annually from Nov 2027

The conditions that would prove each signpost wrong are set out in Part 4.

Part 2. Introduction

Knowledge organizations are bodies whose product is applied expertise: design studios and agencies, law firms, consultancies, universities and training providers, and hospitals and clinics. They share an economic model that is about to be tested. They hire people with scarce skills, bill for their time, seats or visits, and train juniors by giving them routine work.

Generative AI attacks each part of that model at once. It makes routine expertise cheap, it lets buyers see the saving, and it removes the routine work that trained the next generation. Evidence of the change is already visible in 2026: consultancies running fleets of AI agents, clients writing alternative fees into legal budgets, hospitals adopting ambient clinical notes, employers hiring by demonstrated skill, and design tools that generate working interfaces.

Scope

• Domains: design, law, consulting, education and healthcare. Software services were examined in an earlier Srijan Sanchar foresight and are used here as the reference case.

• Horizon: November 2026 to November 2031.

• Questions asked: What outcomes will each domain reach? How will the needs of clients, learners and patients change? How will knowledge organizations respond with new business models?

Who should read this

Leaders of knowledge organizations deciding how to price, staff and structure their work; professional bodies and regulators; funders and employers who buy expertise; and educators preparing the next generation of professionals.

How to read the report

Part 1 stands alone and contains the foresight itself. The remaining parts are written for readers who want to judge it: Part 3 states the purpose, Part 4 describes how the foresight was produced, Part 5 lists the difficulties met, Part 6 states how far the findings can be trusted, and Part 7 proposes next steps.

Part 3. Purpose

The purpose of the report is to help leaders of knowledge organizations decide, before 2028, what to sell, on what terms and with what talent model, by showing how outcomes, stakeholder needs and business models are likely to change to 2031.

Decisions it is meant to support

• Whether to keep selling time, seats and visits, or to move routine work to fixed, subscription or outcome-linked terms.

• Which services to automate first, and who will be accountable for the result.

• How many juniors to hire, and how to train them when routine work disappears.

• Whether to partner with AI suppliers, compete with them, or both.

What it is not

• It is not a prediction. It names the most plausible shapes of the future and the signs that would show which is arriving.

• It is not professional advice for any single firm, school or hospital. Each organization must test the findings against its own context.

• It is not a forecast of company or market performance, and carries no investment guidance.

Assumed actor

The report assumes a reader at the level of a board or head of a knowledge organization. This assumption shaped what counts as valuable (durable trust and margin per professional) and is open to change by the steward.

Part 4. Methodology and Process

The foresight was produced with Integrated Epistemic Foresight (IEF) version 4.0, a Srijan Sanchar method, run in its Lite, AI-assisted mode. The method moves from history to constraints, from constraints to a map of possible futures, from the map to a weighed forecast, and from the forecast to actions and signposts. It separates what is likely from what is valuable.

4.1 How the work was done

Step	What was done
Frame	The phenomenon, purpose, assumed actor, horizon, lenses, pressures and provisional axes were set (Parts 2 and 3).
Gather evidence	Five targeted web searches, one per domain, for 2026 evidence on pricing, hiring, adoption and client demand. Results were read as search extracts, not full pages, and are listed in the Annex. Findings from the earlier software services foresight informed the reference case.
Diverge	Ten theses on where knowledge organizations go were set out, including counter-theses and contrarian views, and each was scored.
Ground	Historical convergence points were mapped, seven constraints were identified with a political filter, and the deepest tension was stated.
Map	Two independent axes were confirmed and a four-cell futures matrix built, with one cell ruled out as a steady state. Each domain was placed on the matrix.
Weigh	Every thesis was tested against the constraints and a forecast state was named. A forward forecast was drafted by year.
Direct	The likely path was separated from the valuable path; gaps, a phased trajectory and six responses were derived.
Integrate	Signposts with falsifiable tests and review dates were set; checks, reflexivity and limits were recorded.

4.2 Thesis inventory

P is plausibility on a 1–5 scale (not a probability). E is the strength of current evidence on a 1–5 scale. Both are AI-drafted for steward confirmation.

ID	Thesis	Origin	P	E	Evidence note
T1	The unit of sale moves from time, seats and visits to outcomes, fastest where output is easy to measure and slowest where payers and regulators fix the unit.	Structural	4	3	McKinsey reportedly earns about a quarter of fees from outcome-based work; a Cleary Gottlieb senior partner predicts a shift to outcome-based pricing. The billable hour is not disappearing in 2026.
T2	Execution commoditises; judgment, taste and accountability earn the premium.	Structural	4	3	Adobe’s 2025 survey: 90% of creative professionals say generative AI saves time on menial tasks.
T3	The apprenticeship pipeline collapses: entry-level work disappears before replacement training exists.	Structural	4	3	PwC US reportedly planned to cut graduate hiring by a third; McKinsey headcount fell from 45,000 to about 40,000; law commentary describes a junior “training gap”.
T4	Clients bring routine work in-house and use AI capacity as leverage on price.	Evidence-led	4	3	About 85% of corporate legal teams reportedly have dedicated AI resources.
T5	AI vendors and AI-native firms become competitors and talent magnets.	Structural	3	3	46 Am Law 200 lawyers moved to AI companies in H1 2026; Harvey is reported in 60+ Am Law 100 firms.
T6	The accountability moat is strongest where licences and liability bind and weakest in design and general consulting.	Structural	4	3	Clinicians stay responsible for diagnosis and treatment in current deployments.
T7	Credentials unbundle: verified skills become the currency of education and hiring.	Structural	4	3	Coursera and Rep Data (a vendor survey): 98% of employers use skills-based hiring in some form.
T8	Healthcare automates the paperwork first and the diagnosis last; patient benefit is unproven.	Evidence-led	4	3	About two-thirds of Epic hospitals use AI documentation; sustained use often settles at 20–50%.
T9	Jevons expansion: cheaper expertise raises total demand.	Contrarian	3	2	Thin direct evidence.
T10	Stall: outcome pricing and AI adoption move slower than expected.	Contrarian	3	3	Hourly-rate RFPs persist; clinician uptake is uneven.

4.3 Historical convergence points

Earlier points are background knowledge, not from opened sources; dates are approximate.

Period	Convergence point	What hardened into a constraint
Medieval to 19th century	Guilds, bar associations, universities and medical colleges form to certify expertise.	Licensing and apprenticeship become the gates to the professions.
1950–1970	US law firms move to the billable hour as the main fee method.	Time becomes the unit of sale; efficiency lowers revenue.
1960s–2000s	Consulting and the Big Four perfect the leverage pyramid.	Junior-heavy staffing funds partner profit.
1990s–2010s	Digital tools wrap around the professional: design software, online courses, electronic health records.	Tools help the human; they do not replace the sale of human time.
2022–2026	Generative AI reaches production: AI-native firms, agent fleets, AI design agents, ambient clinical scribes.	Clients, learners and patients start to ask for the saving.

4.4 Constraint field and political filter

The political filter asks who gains if a constraint persists, because that group will defend it.

ID	Constraint	Who gains (filter)	Effect on futures
C1	Licence and liability. A named, licensed human must sign off in law, medicine and audit.	Professional bodies, incumbent firms	Protects a judgment layer; slows full automation.
C2	Buyer procurement forms. Hourly rates in RFPs, credit-hour accounting, fee-for-service payment.	Procurement teams, payers, accreditors	Slows outcome pricing; rewards dual quotes.
C3	Trust and confidentiality. Client data, patient consent, provenance of AI-made work.	Incumbents with trusted relationships	Favours firms that can show audit trails and consent.
C4	Apprenticeship. Skill is built by doing the routine work that AI now does.	Short-term margin seekers (they gain now, lose later)	A reflexive brake: today’s saving thins tomorrow’s judgment supply.
C5	Outcome measurability. Defining and adjudicating an outcome is hard; no standard value of an outcome exists.	Large buyers, who benefit from vagueness	Limits outcome pricing to measurable work.
C6	Payer and funder structures. Insurers, governments and students pay in healthcare and education.	Payers and regulators	Makes those domains slower than law and consulting.
C7	Supplier concentration. A few AI suppliers set capability and cost, and some sell services directly.	AI suppliers	Knowledge firms depend on suppliers who may compete with them.

The deepest tension

Knowledge organizations sell scarce expertise, and AI makes routine expertise abundant. What stays scarce is trust: who stands behind the answer, and who is still being trained to give it. Every foresight in Part 1 is a different answer to who keeps the saving and who carries the risk.

4.5 Axes and the futures matrix

Axis	Ends	Shaped by	Observable
1. What the buyer pays for	Time, seats and visits to outcomes and verified results	C2, C5, C6	Share of fees or tuition tied to defined outcomes
2. Who does the work	Humans with AI help to AI executing under accountable human judgment	C1, C3, C4	Share of routine work units resolved without human touch

Independence test: AI can do much of the work while buyers still pay by time (today’s state in most domains), and outcome contracts can be signed with little automation behind them. The axes vary separately.

Pays for time, seats, visits

Pays for outcomes and verified results

Humans with AI help

Who does the work

A. Productivity pass-through (NOW, all five)

AI helps staff; buyers still pay by time and ask for the saving back.

Status: feasible, unstable.

C. Fixed-price hedge

The firm takes outcome risk before automation protects its margin.

Status: contested. Viable on narrow, well-measured scopes.

AI executes under accountable humans

Who does the work

B. Phantom hours

AI does the work while the buyer is billed for time.

Status: forbidden as a steady state. Buyers will not pay for time no human spent.

D. Accountable autopilot

AI-run delivery with an accountable professional on exceptions and sign-off, priced per outcome or subscription.

Status: feasible for measurable, routine work; excluded for judgment-heavy work by C1.

Domain placement

Domain	Today	Pull toward	Output easy to measure	Licence or liability grip	Buyer or payer power
Law	A, with alternative fees growing	C then D for routine matters	Medium	High	High (corporate clients)
Consulting	A to C (about a quarter outcome-based, reported)	C, then D for assets and diagnostics	Medium	Low	High (CFOs)
Design	A, with subscription services already present	C and D for production work	High	Low	Medium
Education	A (credit hours), with micro-credentials growing	C (outcome-linked, stackable)	Medium	Medium (accreditation)	High (employers), slow (regulators)
Healthcare	A (visits, fee-for-service), with D pockets in documentation	D for paperwork, C for pathways	Medium	Very high	Very high (payers)

Ratings are AI-drafted judgments from the evidence in the Annex.

4.6 Constraint test and forecast state

Thesis	Constraints tested	Result
T1 Unit moves to outcomes	C2, C5, C6	Narrowed. Holds for measurable routine work; slower where payers or regulators fix the unit.
T2 Execution commoditises	C1, C3	Survives. Strongest in design and law.
T3 Apprenticeship collapse	C4	Survives. A reflexive constraint shared by all five domains.
T4 Clients insource	C1, C3	Narrowed. Holds for routine work; regulated sign-off stays outside.
T5 AI vendors compete	C7	Survives. Partner-or-compete choice becomes strategic.
T6 Accountability moat	C1	Survives. Explains the different pace by domain.
T7 Credential unbundling	C2, C6	Narrowed. Accreditation and funding slow recognition.
T8 Healthcare paperwork first	C1, C3, C6	Survives. Benefit to patients is the open question.
T9 Jevons expansion	C2, C6	Retained as a watch-thesis; evidence thin.
T10 Stall	C2, C3	Narrowed to a delay risk, not a reversal.

Forecast state: Bifurcation inside every domain, at different speeds. Two cells stay feasible at once, populated by different kinds of work: a routine layer moves to D or C, a judgment layer stays human-led, and the hours-based middle is squeezed. The label is AI-drafted and should be checked against the v4.0 state rule.

4.7 Likely versus valuable, and the gap

Likely: a split inside every domain, with the hours-based middle squeezed and the apprenticeship pipeline thinning. Valuable (assumed lens: durable trust and gross margin per professional by 2031): reach cell D or C early for routine, measurable work, keep the accountable judgment layer as the premium, and rebuild apprenticeship on purpose.

Dimension	Typical firm now	Target, 2031	Gap to close
What is sold	Hours, seats, visits, drawings, decks	Defined outcomes, subscriptions, verified skills, living systems	Product definition, pricing models, outcome adjudication
Delivery	AI assists individual staff	AI-run routine lines with exception queues	Workflow capture, evaluation, unit-cost tracking
Talent	Junior-heavy pyramid	Diamond with structured apprenticeship	Paid supervised practice, simulation, senior retention
Assets	People-hours	Playbooks, client context, data, audit trails	Capture and productise professional knowledge
Trust	Reputation and licence	Warranties, attestations, consent, provenance	Governance, insurance, audit capability
Partnerships	Tool purchaser	Co-builder with AI suppliers and clients	Partner posture and multi-supplier design

4.8 Signposts and falsifiable tests

Thresholds are drafts for the steward to set.

ID	Observable	Tests	Falsified if	Review
S1	Share of law-firm fees that are non-hourly (about 20% in 2025 per the cited report) and share of corporate matters on alternative fees.	T1, T4	Hourly share still about 80% in 2028.	Jan 2027, then annually
S2	Share of consulting fees tied to outcomes at McKinsey (about a quarter, reported) and at Big Four firms.	T1	No rise by Nov 2028.	Nov 2027, Nov 2028
S3	Graduate and entry-level intake at Big Four, Am Law 200 firms and consultancies.	T3	Intake returns to 2023 levels while revenue stays flat.	Jan 2027, annually
S4	Employer and university acceptance of credit-bearing micro-credentials and skills-based admission.	T7	Employers revert to degree-required entry roles.	Jul 2027, Jul 2028
S5	Sustained clinician use of ambient scribes (20–50% now) and measured effect on patient access and outcomes.	T8, T10	Active use stays below 30% after three years, or no outcome evidence appears.	Nov 2027, Nov 2029
S6	Share of design production work bought as subscription or outcome-linked, and designer entry-level hiring.	T2	Agencies still bill most production by project hours in 2028.	Nov 2027, Nov 2028
S7	Regulator rulings on AI supervision and liability in law, medicine, audit and education assessment.	T6	A major regulator permits AI-only sign-off in a licensed practice.	Annually from Nov 2027

4.9 Checks performed

• Constraint test run on every thesis: five survive, four narrowed, one kept as a watch-thesis.

• Axis independence tested with a counter-example for each off-diagonal cell.

• Forbidden cell (B) justified by buyer behaviour, not assumed.

• Sensitivity: if regulators loosen C1, cell D reaches judgment work in law and healthcare sooner. If apprenticeship collapses faster than firms rebuild it (C4), the judgment supply tightens by about 2029 and premiums rise. If AI suppliers sell services directly at scale (C7), the commodity layer moves outside traditional firms.

Part 5. Challenges

The following difficulties arose while producing the foresight. They explain several of the choices in Part 4 and the limits in Part 6.

Challenge	Why it matters	How it was handled
Uneven evidence quality	Some figures come from vendor surveys, firms describing themselves, or aggregator sites rather than audited or primary data.	Sources were marked as indicative where relevant; claims rest on more than one source where possible.
Five different professions	Law, medicine, design, consulting and education differ in licensing, payers and measurability, so one simple story would mislead.	A shared frame was applied, and domains were rated on measurability, licence grip and buyer power to explain different paces.
Defining an outcome	Outcome pricing needs agreed baselines and an adjudicator. No firm has standardised what an outcome is worth.	Outcome pricing was limited to measurable routine work; the difficulty is built in as constraint C5.
Direction versus pace	The direction of change is clearer than its speed, which depends on regulation, procurement and trust.	Confidence ratings apply to direction; pace is described by domain and tested by signposts.
The apprenticeship effect	The effect of automation on training is slow, and little hard data exists on the later shortage of experienced judgment.	Treated as a reflexive constraint and a signpost to watch, not as a measured fact.
Interested sources	AI suppliers, tool vendors and firms with something to sell are among the main sources of evidence.	Interest is stated in Part 6; counter-theses and a stall scenario were included.
Geographic skew	Evidence leans toward the United States, with some from India.	Stated as a limit; regional runs are proposed in Part 7.
Fast-moving facts	Adoption figures, product launches and fee structures change within months.	Dates are given; review dates are set in the signposts.
Method access	The full v4.0 specification was not open during this run, so stage templates were reconstructed from earlier runs.	Flagged in Part 6; steward review is the first step in Part 7.
Paywalled source for the earlier thesis	The source article behind the software foresight was readable only up to its paywall.	That foresight used a secondary interview and other coverage; it is a reference case here, not a base.

Part 6. Caveats

Read the foresight with the following in mind.

• This is a Lite, AI-assisted run. Plausibility and evidence scores, domain ratings, the assumed actor, the action threshold and the forecast-state label are AI-drafted for steward confirmation.

• Plausibility is not probability. A score of 4 means a future is strongly consistent with the constraints and evidence, not that it has an 80% chance.

• Several numbers are indicative. Coursera’s findings come from a vendor survey; McKinsey’s agent counts and share of outcome-based fees are reported, not audited; the AIGA design figure comes through an aggregator; the Claude Design effect on Figma’s shares comes from a single source.

• Claude’s position. The foresight was drafted by Claude, made by Anthropic. Anthropic sells AI into these domains, hired eight Am Law 200 lawyers in the first half of 2026, and is reported to have launched a design product. Weigh thesis T5 and response 6 with that in mind.

• Self-fulfilling risk. If buyers read this report, they will press harder for pass-through, strengthening the squeeze it describes.

• Coverage. Five domains are treated at overview depth. Evidence is weighted to the United States and India. Each domain deserves its own run.

• No second reader. No second reader or human-node session tested the work, and the assumed actor and value lens have not been confirmed.

• No novelty claim. Two framings may be new: the apprenticeship pipeline as a shared reflexive constraint across knowledge professions, and the comparison of domains by how tightly licence, payer and buyer power fix the unit of sale. Neither has been traced to prior art, so neither should be presented as original until that check is done.

• Time-sensitivity. Facts are as found in searches up to early November 2026.

• Not advice. The report is not legal, medical, financial or educational advice.

Part 7. Way Forward

The following steps would turn this working foresight into a tested, maintained one. Timings are suggestions for the steward.

#	Step	Purpose	Suggested timing
1	Steward review of the draft judgments. Confirm or change the scores, domain ratings, assumed actor, action threshold and forecast-state label against the v4.0 specification.	Remove the main AI-drafted uncertainty.	Within 2 weeks
2	Second reader and human-node sessions. Test the constraint field and the forbidden cell with practitioners from each domain.	Add lived knowledge the web evidence cannot supply.	Within 2 months
3	Strengthen the evidence. Replace indicative figures with primary sources: law-firm fee surveys, consultancy disclosures, labour data, education and health system reports.	Raise evidence scores; cut reliance on interested sources.	Within 3 months
4	Deepen the domains. Run a full foresight for each domain, starting with law and consulting where change is fastest.	Move from overview to decision-grade detail.	Q1–Q2 2027
5	Add regional lenses. Run India-specific and other regional versions, including the role of public universities and health systems.	Reduce the geographic skew.	Q2 2027
6	Check prior art. Trace the two candidate framings to earlier work before any claim of thought leadership.	Protect credibility.	Within 2 months
7	Monitor the signposts. Keep a calendar of the seven review dates and record what is found.	Know early which branch of the fork is arriving.	From Jan 2027
8	Test responses with pilots. Support two or three willing organizations in trying responses 1 to 3 on one routine service.	Turn the foresight into evidence of what works.	2027
9	Extend the set. Add accounting and audit, architecture, journalism and public administration, and link to the software services foresight.	Build a family of foresights on knowledge work.	2027–2028
10	Refresh. Re-run the foresight at least annually, or sooner if a signpost is crossed.	Keep it current.	Annually, first Nov 2027

The foresight’s central message is practical. Organizations that sell the outcome, guard the judgment and grow the apprentices will set the terms of knowledge work in 2031. Those that keep selling routine hours will find that the saving belongs to someone else.