AI Flashpoints Certain to Arrive in the Next 1-3 Years — Three Design Leads’ Near-Term Flashpoints (2026-2029)

Question: “What are the issues around AI use in your own domain that are almost certain to arise within the next 1-3 years?” Three design leads (business-only = only three perspectives: does it make money / can it be billed to clients / can AI make our own work more efficient; no scholars, doctrines, or theoretical terms) were asked to enumerate, each for their own domain, only the issues that are not future predictions but already showing signs and certain to ignite within 1-3 years. The primary deliverable is not convergence but a sourced table of disagreements.

Industry personas: F = design-firm-lead (IDEO/frog-type full-service) / C = strategy-consulting-design-lead (McKinsey/Accenture Song type) / O = independent-design-office-lead (Pentagram/boutique type). Business evidence ledgers: industry.md / voices.md. Related: designer-role-ai-roundtable (the debate on the role itself) / ai-in-design-industry / ai-in-design-2026 / the roundtable on delivering multi-agent value / eu-ai-act-design-impact (an Article 50 analysis that grounds O’s disclosure dilemma and IP clauses as legal obligations).

Handling of figures: Company-issued figures (McKinsey/Bain fee structures, Design Index +32pp, etc.) are treated as sales-pitch material = position talk, correlation not causation; headcount reductions (McKinsey/Accenture) as real figures corroborated by multiple consistent reports; vendor surveys (Forrester 40%, Upwork 25-60%) as interest-laden. [requires verification] = primary source not yet reached.

Structure of This Note (hub)

This note is the debate log (minutes) of R1-R5. Reusable cross-cutting claims are carved out into atomic notes and connected to related clusters (the compounding principle of the LLM Wiki: units of knowledge are turned into nodes and allowed to emerge as a graph).

• Debate log (chronological): R1 enumeration of near-term flashpoints -> R2 cross-rebuttals (gains and losses invert by position) -> R3 the danger of deliverable-based pricing (R2 retracted) -> R4 survival prescriptions for the good-enough tier -> R5 theoretical examination by academic critics
• Extracted claims (atomic, reusable nodes):
  • democratization-recommodification-paradox — the asymmetry by which democratization turns “anyone can do it” into “no one can bill for it” (deskilling / social closure / symbolic capital) -> to the AI inequality cluster
  • design-pricing-vs-ai-commoditization — pricing models’ resistance to AI commoditization (deliverable pricing is the most fragile; hard-to-replicate embedded operations does not equal high margins; the gatekeeper business collapses) -> to the industry cluster

TL;DR

Although the three were run separately, they independently pointed to the same three tectonic shifts (= the highest-confidence findings).

1. Who captures the efficiency gains: When AI cuts work hours, staying on person-month or hourly billing turns efficiency into your own price cut rather than increased revenue. Clients will invariably demand discounts on the grounds that “AI must have made it cheaper.” Only those who restructure their billing model toward outcome-, value-, or name-based pricing can keep the efficiency gains in their margins.
2. Can taste / authenticity be written on an invoice?: Because output homogenizes, differentiation shifts to the human eye, editing, and authenticity — all three agree on the direction. But “there is no real data on rates where a taste fee was set as a proposal line item and approved and paid.” Value that cannot be translated onto an invoice becomes the first line cut from a proposal.
3. Evaporation of the middle tier: Junior mass-production work — the entry point into the profession — shifts to AI, and in three years the bearers of taste (seniors) run dry (firm / consulting). The middle range of “good-enough work at good-enough rates” collapses against $20-a-month AI, polarizing into top-end name commissions or bottom-end mass production (independent).

And all three rejected the optimism of “AI = productivity gains = everyone wins,” each on profit-and-loss grounds.

Cross-Cutting Cores (Independently Identified by All Three)

Core	F (Firm)	C (Consulting)	O (Independent)
(1) Who captures efficiency gains / discount pressure	At 20h -> 5h under person-month billing, revenue drops 75% = “timesheet arithmetic”	Shifting to outcome-based fees (risk moves to our side); efficiency gains get seized as discount ammunition	84% use AI routinely = efficiency mutates from differentiator into discount material; staying on hourly billing strangles you
(2) No proof of the premium	No rate data where “selling taste/edges” was written as a proposal line and approved	ROI figures like +32pp are correlation, not causation; procurement’s rising literacy will demand provenance	Evidence that authenticity commands premium rates is thin (Scher adopted AI because “there was no budget”)
(3) Middle-tier evaporation and polarization	Junior entry point disappears -> bearers run dry (IDEO from about 725 to under 400, DF-01)	Leverage-pyramid compression (underway in real numbers)	The middle price band collapses to AI subscriptions; polarization into top-end name work or bottom-end mass production

Each Lead’s Flashpoints (business-only)

F — Full-Service Firm (IDEO/frog type)

1. The self-destruction of person-month billing: With hours falling, staying on person-month billing means efficiency = cutting your own price. Only organizations that have moved onto value-based pricing keep their margins. The 2026 contract-renewal cycle will force this on every firm.
2. The unbillability of taste: The direction — differentiation moving to taste and editorial judgment — is right, but neither our firm nor the industry has real data on rates set as proposal line items and approved. Value that cannot be written down is the first line cut from a proposal.
3. Disappearance of the junior entry point and the collapse of training: Mass production, tracing, and first drafts — the only entry point that raises the next seniors — shift to AI. Short term, per-project margins rise; medium term, hiring costs eat the gains. The shrinkage already shows in hiring-plan numbers.
4. The self-contradiction of the “drop production, sell capability” pivot: Building in-house capability = making the client self-sufficient = that project does not come back next year. You are cutting your own repeat business. There is no supporting data on post-pivot repeat rates or fees.
5. Bundle unbundling -> leakage to in-house AI: If production comes out cheap, clients will run it in-house with AI. Only the thin-margin production gets taken, and the package gets picked apart (in-house 71%, IS-03).
6. Unable to measure AI’s effect internally, defenseless in discount negotiations: Without real figures on “which process was cut by what percentage and where the freed time went,” you cannot push back. Even with 92% of UX practitioners using AI (OR-02, possibly overstated), the effect cannot be shown in numbers.

C — Strategy Consulting (McKinsey/Accenture Song type)

1. Outcome-based fees vs. the collapse of effort-based billing: “We don’t pay for decks; we pay tied to KPIs.” Roughly 25% of McKinsey’s fees are outcome-linked, Bain roughly 30% (company-issued = treated as sales-pitch material). Project P&L built on day-rate x utilization breaks down.
2. Commoditization of generated output vs. the craft premium: Functional UI now comes out of in-house non-designers in days (the fear behind Figma’s -7% stock drop; 60% of Figma files are created by non-designers). Revenue disappears for teams that billed UI production hours.
3. Leverage-pyramid compression vs. junior development: High-volume junior work is exactly what AI is good at. McKinsey roughly 3,000-4,000 and Accenture roughly 11,000 cut, plus 22,000 deemed unretrainable and let go ($865M). Underway in real numbers. Short-term margin up, medium-term senior drought.
4. Is “product builder” an evolution of the craft or a pretext for headcount cuts?: “PRDs disappear, 30-50% of code is machine-generated, one person orchestrates AI” ([requires verification], commentator statement). But every firm is cutting people at the same time — suspect a glorified cost cut that reduces billable headcount.
5. PoC mass production vs. the 16% wall of enterprise-wide rollout: PoCs are thin-margin; the money only comes with enterprise-wide rollout. Only 16% of companies have fully redesigned (own survey = sales-pitch material).
6. Trust inflation in ROI figures vs. demands for verification: +32pp/+56pp (Design Index) is correlation, not causation. As outcome-based fees spread, correlation-based sales pitches stop working and you get pinned to verifiable KPIs.
7. The tug-of-war over efficiency gains: Even after compressing in-house work by 40% (Forrester = vendor survey), clients come to seize it as discount ammunition, asking “why isn’t this reflected in the price?” Efficiency does not automatically mean profit.

O — Independent Studio (Pentagram/boutique type)

1. Evaporation of the middle price band: The middle range — logos, light rebrands — loses to $20-a-month AI. Corporate freelance spending fell from 0.66% to 0.14% while AI model spending rose from 0 to 2.85%. The only move is to swing fully to one pole: top-end name commissions or bottom-end mass production.
2. Standardization of AI-disclosure and IP-ownership clauses in contracts: Clients are inserting clauses like “rights to deliverables, prompts, and inputs belong to us” and “no AI use or training without prior approval” (premised on the distrust that 58% used AI without disclosing it). Once bound, the hours you had eliminated come back and margins erode. In 2026, voluntary frameworks give way to mandatory compliance. The GSA also drafted AI procurement clauses in March 2026.
3. Copyright-infringement liability fixed on the studio: Nearly every AI tool says “the output is yours; you indemnify against infringement claims.” “The AI made it” is no defense. A small studio has neither the balance sheet nor the legal staff for damages, and one case can sink it (the $1.5B Anthropic settlement made “user = indemnifying party” real).
4. The disclosure dilemma (tell or hide, you lose either way): Paula Scher/Pentagram used Midjourney on performance.gov because “there was no budget to commission illustration” and was slammed as “theft” (D-01, already occurred). Disclose and you lose goodwill; hide and you are cornered on contracts and liability.
5. The unproven authenticity premium: “Human-to-human resonates more” (Sagmeister-esque, [requires verification]) is a sentiment; evidence that anyone pays extra for it is thin. Bet on authenticity without proof while carrying fixed costs, and the price collapse swallows you.
6. Who captures efficiency gains (on hourly billing it turns into a price cut): Route mass production and prep work to AI and redirect the time to upstream name commissions, and margins rise. But: “if AI made it faster, make it cheaper.” With 84% using AI routinely (up from 41% in 2023), efficiency is negotiating material, not a differentiator.

Each Lead’s “Single Most Unavoidable Point”

Position	Most unavoidable flashpoint	Why it is certain, not a prediction
F	With hours falling, staying on person-month billing turns efficiency into a price cut	Timesheet-and-invoice arithmetic. The 2026 contract-renewal cycle forces it on every firm
C	Leverage-pyramid compression (junior cuts = short-term margin up, medium-term senior drought)	Underway in real numbers (McKinsey roughly 3,000-4,000; Accenture roughly 11,000 plus 22,000 let go). Not “might happen” — already happened
O	Contractual AI disclosure + IP ownership + liability fixed on the studio	Already arriving as contract clauses. Whether or not they use AI, every studio will be asked “can you explain provenance and price liability into your fees?”

Unresolved Disagreements (Not Smoothed Over — the Primary Deliverable)

Issue	Position A	Position B	What would settle it
Is efficiency more revenue or a price cut?	Depends on the billing model (stay on person-months and efficiency = lower revenue, a self-imposed price cut)	Restructure to value/outcome/name-based pricing and it converts to revenue growth (the billing model is a choice variable)	Whether any organization that moved to value-based pricing kept the efficiency gains (neither F nor C has real figures)
Who owns the efficiency gains?	The studio/firm can turn them into margin	The client seizes them as discount ammunition (“it must be cheaper now”)	Real examples of contract structures that converted reduction rates into margin (the contract-model tug-of-war is unsettled)
Can taste/authenticity be billed?	Differentiation is moving there (the direction all three share)	No line-item rates that got approved; no evidence authenticity commands premium rates	Project rates where a “taste fee” was line-itemed, approved, and paid / data converting non-famous designers’ authorship into income
How to judge junior cuts	Short-term per-project margin up (the C/F economics)	In three years the stock of seniors (bearers of taste) runs dry	Real medium-term figures on senior supply and hiring costs after the cuts
The nature of the product builder	An evolution of the craft (one person orchestrating AI)	A pretext for cuts that shrink billable headcount (firms are cutting people at the same time)	Research into product builders’ discretion, actual work content, and billing arrangements ([requires verification])
Should AI use be disclosed?	Disclosure = clean provenance, contract compliance	Disclosure = the basis of the authorship premium gets questioned and goodwill is lost	Both options lose under the current structure; no optimum exists (the disclosure dilemma is unresolved)

Conclusions Not to Adopt (Unanimously Rejected by All Three)

• “AI = productivity gains = everyone wins”: Clients will come to seize the efficiency gains as discount ammunition / unless you restructure your billing model, you are only cutting your own price. Any productivity argument that does not say who converts the reduction rate into margin is ignoring the reality of the invoice (O).
• “The role moves up to strategy, taste, and curation, so handing the mass-production entry point to AI is fine” (F): Untenable on three counts: (1) no approved taste-fee rate data, so it does not make money; (2) it cannot be written as a billing line and gets cut from proposals; (3) junior seats disappear and the bearers run dry in three years.
• “Go small and hyper-efficient with product builders and it scales into profit” (rejected by C): If you just cut people while staying on person-month billing, it is lower revenue plus junior layoffs. Earn the “evolution” label only after showing, in numbers, whether billable headcount rises or falls.
• “We have authorship, so AI does not concern us; guard the craft and we can eat” (rejected by O): No numbers behind it; it collapses immediately against middle-tier evaporation and the unproven authenticity premium. A studio that stays in the middle carrying fixed costs in the name of craft will see its margins break within 1-3 years.

To Verify (Practice Data Needed Next)

• Real examples of organizations that moved to value-based pricing and kept the efficiency gains (settling efficiency = revenue growth vs. price cut).
• Project rates where a “taste fee” was set as a proposal line item, approved, and paid / post-AI production-hour reduction rates, utilization, and proposal win rates (F; untallied at both firm and industry level).
• Medium-term senior supply and hiring costs after junior cuts (C) / PoC-to-enterprise-rollout yield (real figures behind the 16% wall).
• Data converting non-famous designers’ authorship into income and empirical proof of payment for the authenticity premium (O).
• Product builders’ discretion, actual work content, and billing arrangements (to distinguish evolution from a pretext for cuts, [requires verification]).
• Corpus caveats: DF-01 is one full-service firm’s example / OR-02’s 92% may be overstated / 30-50% machine-generated code is a commentator statement with causation unverified / company-issued figures (McKinsey roughly 25%, Bain roughly 30%, 16%, +32pp/+56pp) are sales-pitch material, correlation not causation.

R2 — Cross-Rebuttals (Each Flashpoint Tested by Leads from Other Positions)

Method: Each flashpoint from R1 was worked over by the other two leads in business-only profit-and-loss terms — steelman -> rebuttal -> “works in reverse in my domain / does not apply” (F -> C and O / C -> F and O / O -> F and C). The primary deliverable is not convergence but making visible how the same flashpoint’s gains and losses invert by position.

The Core of R2 (Revealed by the Three-Way Cross)

1. “Hourly / person-month / effort-based billing x efficiency = self-destruction” (F1 = C7 = O6) is unanimous. But the means of neutralizing it diverge, and only F has little escape. O: hide the hours and charge for the work itself, and efficiency flips from a price cut into margin improvement / C: never sold person-months or hours in the first place / F: person-month dependent, little escape. F was hit by both C and O with “that is your pricing disease, not an AI problem.”
2. R2’s biggest discovery: on junior-training collapse (F3 = C3), the small studio is in fact the most fragile. O self-corrected: “A big shop has a thick pyramid base and keeps running if one person leaves; but our commissions are tied to a specific author, so if that author departs, the revenue evaporates instantly along with the name.” F’s “training collapse” and C’s “pyramid compression” bite deepest at O as “no successor to the author.” An overlooked issue sitting before O5 (unproven authenticity): “the supply side of name commissions runs dry.”
3. On O-originated O2+O3 (contractual AI disclosure + IP ownership + liability fixation), F and C both conceded “our losses are larger.” F: “Our project sizes are larger so damages are larger, and since we have subcontractors use AI, supply-chain risk is also heavy.” C: “Heavy on public-sector and regulated work including the GSA. But double-edged — it also becomes a new billing line, ‘AI governance audit.’” What began as O’s solo flashpoint turned out to spill over to all three.

Flashpoints That Invert or Get Rejected by Position (Asymmetry Map)

Flashpoint (origin)	F’s gains/losses	C’s gains/losses	O’s gains/losses
F1 Person-month billing self-destruction	The core disease (person-month dependent, little escape)	Does not apply (does not sell person-months)	Margin improvement in reverse (under work-based pricing, efficiency = gain)
C1 Outcome-based fees	Not transplantable (design outcomes cannot be isolated within business metrics)	Home turf	Unfavorable (a business of charging for unmeasurable outcomes) -> not adopted
C4 Product builder	Rejected (a pretext for cuts; total orders do not grow)	Own flashpoint (though C itself retains the headcount-cut suspicion)	Rejected (not one cent of relevance to work-based pricing)
C5 The 16% wall	An opportunity in reverse (adoption support is an escape billable by person-months)	Own flashpoint	A tailwind in reverse (AI transformation stalling = human work remains)
F2 Unbillability of taste	The core disease	Partial agreement (translates it into billing, but track record unconfirmed)	Rejected (itemizing it is already losing; dissolve it into the package via the name)
O4 Disclosure dilemma / O5 Authenticity	Partially applies (but selling process gives higher backlash resistance than O)	A tailwind in reverse (AI use is a selling point, not something to hide)	The core trap (specific to the authorship brand)
O2+O3 Liability fixation	Own losses larger	Heavier for itself (double-edged: also turns into a new billing line)	Core, most unavoidable

Shifts in Own Verdicts (the Single Most Unavoidable Point)

• F: Keeps F1 in first place while upgrading O2+O3 a rank from before. Reason: F1 is “a gradual loss where rates creep down”; O2+O3 is “a tail risk where a single contract blows away the margin — and our loss amount is larger than O’s.”
• C: C3 = F3 tied for first (the industry’s total billable volume shrinks — it does not go away even if you change the fee structure); O2 second as double-edged; F1/O6 third as avoidable through pricing. Biggest lesson: F6 — “selling other people’s ROI while not measuring our own AI efficiency is the greatest hypocrisy and the greatest vulnerability.”
• O: Keeps O2+O3 as most unavoidable. But concedes C3 = F3 is “the one threat you cannot escape by changing the billing model” and discovers that O itself (small headcount) is the most fragile. Places O3 (next month’s contract risk) and C3 (succession collapse five years out) side by side as separate axes on different time scales (refuses to rank them).

Non-Convergence Remaining After R2

• The single most important point remains split by position: F = F1 (gradual self-destruction) / C = C3 = F3 (shrinking total billable volume) / O = O2+O3 (immediate tail risk of liability fixation). The kinds of pain differ, so there is no convergence.
• No one has demonstrated an alternative fee structure: For the escape routes from “hourly/effort billing x efficiency = self-destruction” (fixed / outcome-based / licensing), none of F, C, or O has a real example where the efficiency gains stayed in-house.
• Everyone’s intangible-value rate record is [requires verification]: F’s taste translation, C’s taste/ROI translation, O’s authenticity premium — none has documentation of an approved rate (C6 boomerangs onto C’s own C1/C2/C3 as well).
• Hiding hours and AI-disclosure demands cannot coexist: O and F say “if we do not show the hours, we can neutralize the efficiency discount pressure,” but this collides head-on with the client’s “if you used AI, make it cheaper” (O4). Unresolved across all three.
• The few points of three-way agreement: (1) hourly/effort billing x efficiency = self-destruction; (2) the disappearing junior entry point = bearers and successors running dry; (3) ROI and intangible-value figures should be discounted by everyone as correlation, not causation.

R3 — Response to the Client Counterargument “Isn’t Deliverable-Based Pricing More Dangerous?”

Counterargument (from the client side): The pressure of “AI made production more efficient, so lower the price.” More clients of commissioned production are satisfied with the quality of output AI produces in a few hours, and the market comes to want “good-enough quality, cheap.” R2’s claim that “with deliverable (work) pricing, efficiency becomes margin improvement rather than a price cut” fails if AI can produce an equivalent deliverable in hours — the market price of the deliverable itself should fall, so isn’t deliverable-based pricing in fact the more dangerous one? And as a result, wouldn’t the entire creative domain shrink? Conclusion: The counterargument hit home. All three conceded that “deliverable-based pricing is more dangerous” and retracted or revised their R2 claims.

Pricing Models’ Resistance to AI Commoditization (Verdicts by All Three)

Extracted claim: design-pricing-vs-ai-commoditization (the distilled, reusable node for the pricing-model findings from this section onward) | | Deliverable (object) pricing | Hourly / person-month billing | Outcome-based fees | |---|---|---|---| | F | Weakest (an object = reproducible = shelved next to AI; the floor drops out faster than person-months) | Middle | Strongest (but few projects where it can be made to work) | | C | Middle to weak | Weakest (fewer hours = immediate price cut) | Strongest (but admits pure outcome-linking barely exists, depends on fixed fees, and that it had oversold the safety) | | O | Without a name, defenseless / with a name, a different thing entirely | — | — |

• The crux: “The moment you define the deliverable as an object, its price slides toward AI’s marginal cost (near zero)” (F). F and C split only on “which is more dangerous, person-months or deliverables” (non-convergent).

O’s Core Self-Correction (the Most Valuable Turn in This Note)

“What protected the price was not ‘hiding the hours’ but ‘the name (who made it).’ My prescription generalized something that holds only for the named few, as if it were the achievement of the deliverable-pricing format. This is the most dangerous form of position talk: disguising one’s own privilege as a ‘tactic anyone can use.’ I retract it.” -> “Deliverable-based pricing” splits into work pricing (who made it = defensible) and deliverable pricing (what was delivered = defenseless). Most commissioned production is the latter and, having no name, cannot be defended.

Only What Is Not Turned into an Object Withstands Discount Pressure

• Cannot withstand (= the domain of “good-enough cheap objects”): one-off UI, logos, landing pages, banners, mass production, and “making things” itself. F: “You cannot justify a price to a client who is satisfied. Build the business on the assumption of losing this tier.” “Educate them on the quality difference and win them back” was rejected as a wish that never becomes an invoice.
• Can withstand: acting as proxy for decision-making, assuming responsibility/liability, integration, regulatory compliance, the name. Only what cannot be carved out as an object. But “no one has real data showing the same amount billed in non-object units” (both F and O: [requires verification]). Integration too will be eroded by AI within three years (F).

Will the Creative Domain Shrink? — Opposite Answers for “Headcount” and “Money”

• The number of projects and initiatives grows (production gets cheaper, the base widens) / the total amount paid for “making things” shrinks (rates get pulled toward AI’s marginal cost) / the headcount of makers falls, without question.
• O’s honest summation: “Saying ‘it is redistribution, so the whole does not shrink’ papers over the disappearance of mid-tier makers — a framing convenient for my own shop. Headcount falls. There is no dressing up the fact that this is a human shrinkage of the domain.”
• What disappears is the middle (the tier selling good-enough quality by person-months/hours) and the junior entry point; what remains are the two poles of the top (names, responsibility) and the bottom (mass production at zero marginal cost through AI automation) — a “hollowing out of the middle.”

Non-Convergence Remaining After R3

• F vs. C: which is more dangerous, person-month billing or deliverable-based pricing.
• The safety of outcome-based fees: C itself admits “pure outcome-linking barely exists and depends on fixed fees; design outcomes cannot be isolated from price revisions, sales, and seasonal factors” = not a cure-all.
• Redistribution vs. headcount shrinkage: viewed in “money,” redistribution; viewed in “maker headcount,” shrinkage. Unsettled by data.

R4 — Prescription: “What Should an Office Carrying Many Non-Famous ‘Good-Enough’ Designers Do?”

Question: How does a commissioned-work office with little name recognition, authorship, or premium — carrying many good-enough designers (= heavy fixed costs) — survive the next 1-3 years? The light-footed independent’s prescription of “eat off the name” is unavailable. Including what to do with headcount (redeployment / retraining / reduction). Three-way convergence: The three leads, from different positions, converged on nearly the same diagnosis, the same prescriptions, and the same trap (the most convergent round in this note).

Diagnosis: The Cause of Death Is Not “No Name” but “Heavy Fixed Costs”

• F: “The real disadvantage is not aesthetics or talent but the balance sheet. The independent’s retreat is cheap; this tier’s retreat is expensive. The cash runs out before the transition completes.”
• C: “Within 1-3 years headcount becomes almost entirely a liability. A thin apex over a fat base is not a leverage structure, just a lump of payroll. It is an accounting fact that the moment utilization no longer covers it, the loss is locked in.”
• O: “More than the lack of a name, heavy fixed costs are the direct cause of death. What favors independents is not talent but being light.” -> Not a problem of talent or authorship but a finance problem (fixed costs, utilization, cash flow). Every move comes down to “how fast and how correctly you shrink and change shape.”

Two Survival Routes + One Trap Not to Step On

	Substance	Billing	Assessment
A Industry-specialized delivery (most promising)	Stake out 1-2 heavily regulated, taboo-laden industries (healthcare, finance, public sector) and decompose into AI first-pass generation + human QA and liability assurance	One-offs -> annual contracts of N pieces per year at fixed rates, with a 10-20% markup as “peace-of-mind money against redos”	Headcount is re-capitalized as “review of AI output / liability assurance.” O: “An individual name cannot be retrofitted, but a collective name (industry specialization) can be built”
B Operations embedding / residency (helps cash flow)	Stop one-off delivery and embed in operations (updates, A/B testing, brand gatekeeping)	Monthly retainer / seat-based billing = selling “operations capacity”	Predictable revenue makes utilization troughs shallower. But it is not work that continues the craft = a change of occupation. C: “Exit risk of being cut once in-house capability is complete”
Trap C Generic “cheap, fast, mass-produced” BPO	Keep current headcount and use AI to mass-produce cheap and fast, earning on volume	Price cuts x volume	Unanimously judged self-destructive. “You set the floor price yourself and accelerate rates toward AI’s marginal cost”; “the only competition you can win is against offshore-plus-AI, and latecomers melt”

• The core escape not to take (unanimous): “Keep current headcount and earn on volume by producing cheap and fast.” C: “Distrust the painless prescription. The least painful option (cut no one, extend the present) is the most dangerous.” O: “Overreaching with ‘we will compete on authorship too’ is the most expensive way to lose.”

Execution Order (F’s Cash-Flow Design — the Heart of “What to Do”)

1. Stop the bleeding (top priority, 6-12 months): Convert headcount to variable cost (outsourcing, carve-outs, attrition via hiring freeze) + compress hours 30-50% with AI to run the same orders with fewer people. The biggest bleed is one-time costs (several million yen per person cut; cutting 20-30% of 100 people runs to hundreds of millions of yen [requires verification]) — whether you can raise this money is life or death. Meanwhile, live off operations and maintenance for existing repeat clients and stay out of new-business discount wars.
2. Move operations to monthly billing (6-18 months): Convert 2-3 existing clients to monthly contracts. Live off existing production revenue in the meantime.
3. Industry specialization (12-24 months, slow-acting): Premised on surviving via (1) and (2). Not everyone can convert.

What to Do with Headcount (the Three Leads’ Numbers Nearly Align; All [requires verification])

All three flatly call “rescue everyone through upskilling” a lie. Aggregated: 30-40% are redefined and kept for direction of AI output / QA / liability assurance / client negotiation; 20-30% are redeployed to operations, project PM, and client-facing roles; about half (40-60%) have no seat within 1-3 years. C: “We do not cut people because their ability is low. The seat disappears because the volume of billable work for that ability shrinks. Macro demand sets the number of seats, and upskilling carries no guarantee the individual is rewarded.”

Survival Rate in Three Years

One office in two to four survives as a “small, sharp office that has cut headcount to half or below and narrowed to industry specialization or operations residency.” Nothing survives in its current form (large payroll, generic commissioned work). The conditions for survival are less about ability than (a) cash to cover the bleeding through the transition, and (b) whether headcount reduction was done early and done right.

The One Number That Decides Life or Death, Which No One Has

The crux all three admitted they hold no real data for: (volume x remaining margin) = “even if AI raises volume, does enough margin remain to offset falling rates with volume and make a living?” O: “Triple the volume at a third the rate and you are flat. Whether you can eat here is the single number that decides this tier’s life or death, and I do not have the answer.” C: “The dividing line is whether the hour savings become your own margin or get siphoned off as discounts. Do not trust it until real data appears.” -> The direction of the prescription (specialization + operations + early headcount reduction) is unanimous, but no one holds final proof that you can live on it.

Interest Disclosure (Grounds for Discounting)

• C self-exposed: “This tier is my ideal acquisition and absorption target, and prescriptions A/B are two sides of the same coin as my own sales pitch for AI transformation. I deliberately left out the McKinsey ROI figures this time.”
• O: “This tier’s survival lies not in aiming upward (authorship) but in descending deep into the unglamorous industry-operations work I would never do, and carving out a separate habitat there. The direction of the shrink is not ‘up toward authorship’ but ‘into depth of industry context and operations.’”

References

The industry personas argue business-only (numbers, engagements, and client reactions alone) and do not ground their claims in scholars, theories, or papers. The following are external sources newly obtained by the three leads. First-party announcements, vendor surveys, and pundit statements are treated as position talk / correlation ≠ causation (see the notes in the body). The business-evidence ledgers are industry.md / voices.md. For the debate on the role itself, see designer-role-ai-roundtable.

Pricing Models (Common to F, C, and O)

• digitalapplied. “AI-Era Agency Pricing Models: A 2026 Decision Guide.” — https://www.digitalapplied.com/blog/ai-agency-pricing-models-2026-decision-guide (the arithmetic of 20h -> 5h yielding a 75% revenue drop)
• Anders CPA. “How AI Is Impacting the Most Common Agency Pricing Models.” — https://anderscpa.com/learn/blog/agency-pricing-models/
• TheStreet. “AI is forcing McKinsey, BCG, Bain to rethink consulting fees.” — https://www.thestreet.com/markets/ai-is-forcing-mckinsey-bcg-bain-to-rethink-consulting-fees
• Consultancy.uk. “Is AI about to kill the billable hour consulting model?” — https://www.consultancy.uk/news/42718/is-ai-about-to-kill-the-billable-hour-consulting-model

Pricing-Model Transition (R3; published by agencies and pricing advisors — discounted as position talk)

• Simon-Kucher. “How professional services can successfully manage the AI price-model transformation.” — https://www.simon-kucher.com/en/insights/how-professional-services-can-successfully-manage-ai-price-model-transformation
• Net-Craft. “Fixed or Flexible? How AI Web Design Agencies Are Changing the Way Clients Pay.” — https://www.net-craft.com/blog/2026/05/30/ai-web-design-pricing/
• Digital Agency Network. “AI Agency Pricing Guide 2026.” — https://digitalagencynetwork.com/ai-agency-pricing/

Headcount Reduction and Leverage (C)

• Computing. “McKinsey considers thousands of job cuts as AI reshapes consultancy work.” — https://www.computing.co.uk/news/2025/ai/mckinsey-considers-thousands-of-job-cuts-as-ai-reshapes-consultancy-work
• CNBC. “Accenture plans on ‘exiting’ staff who can’t be reskilled on AI.” — https://www.cnbc.com/2025/09/26/accenture-plans-on-exiting-staff-who-cant-be-reskilled-on-ai.html
• futureofconsulting.ai. “2026 Consulting’s AI Revolution Update.” — https://futureofconsulting.ai/ai-leadership/2026-consultings-ai-revolution-update/

Commoditization of Deliverables (C)

• LogRocket. “Figma AI in 2026: Everything it can do.” — https://blog.logrocket.com/ux-design/figma-ai-2026-quick-overview/
• Sesame Disk. “Figma in 2026: The AI Disruption Arrives.” — https://sesamedisk.com/figma-ai-disruption-2026/

Collapse of the Mid-Market and Freelancing (O)

• We and the Color. “The Collapse of the Mid-Level Freelance Market Due to AI.” — https://weandthecolor.com/freelance-designers-cant-compete-with-a-20-month-ai-subscription-heres-what-actually-works-now/209620
• Clutch.co. “The State of the Graphic Design Industry — What 2026 Data Reveals.” — https://clutch.co/resources/graphic-design-industry-2026

Contracts, IP, Liability, and Disclosure (O)

• Taft Law. “The Expanding Prevalence of AI Clauses in Contracts.” — https://www.taftlaw.com/news-events/law-bulletins/the-expanding-prevalence-of-ai-clauses-in-contracts/
• Tascon. “AI Clauses In Contracts: The Practical Guide For 2025.” — https://tasconlegal.com/ai-clauses-in-contracts-the-practical-guide-for-2025/
• OpenArt. “The Creator’s Guide to AI Disclosure in 2026.” — https://openart.ai/blog/creators-guide-ai-disclosure-2026/
• Martensen IP. “A Practical Guide to Generative AI Copyright Risk.” — https://www.martensenip.com/blog/2025/november/a-practical-guide-to-generative-ai-copyright-ris/
• Copyright Alliance. “AI Copyright Lawsuit Developments in 2025: A Year in Review.” — https://copyrightalliance.org/ai-copyright-lawsuit-developments-2025/

Pentagram/Scher’s AI Adoption and the Backlash (O, D-01)

• Creative Bloq. “Paula Scher on the backlash to Pentagram’s AI government website design.” — https://www.creativebloq.com/design/ive-never-seen-anything-like-this-before-paula-scher-on-the-backlash-to-pentagrams-ai-government-website-design
• Fast Company. “Pentagram partner Paula Scher unapologetically defends using generative AI.” — https://www.fastcompany.com/91240809/pentagram-partner-paula-scher-unapologetically-defends-using-generative-ai
• Godspeed. “Illustrating Pentagram’s AI (Mid)Journey.” — https://godspeed.ghost.io/pentagram-midjourney-ai/

Update Policy

This note is a living page. When the to-verify practice data arrives (real examples where value-based pricing kept the efficiency gains / approved taste-fee rates / senior supply after junior cuts / empirical proof of payment for the authenticity premium), append it and update updated. As flashpoints settle or dissolve over the next 1-3 years, write the resolutions into the disagreement table. If theoretical connection by academic critics (theory/education/industry) becomes necessary, supplement it in a separate note as the next round of designer-role-ai-roundtable. The industry-persona apparatus is .claude/skills/designer-role-roundtable (convening) + each perspective skill (business-only = .claude/practice-roundtable-protocol.md).