Almost everyone has adopted AI. Almost no one has redesigned around it.
We read twelve of the most-cited agentic-AI reports of 2025–26 — from a boutique AI-native consultancy to McKinsey, Bain, BCG, Deloitte, PwC, Accenture and IBM, to the AI vendors themselves. Despite different vocabularies, they say one thing in unison: the constraint on value is no longer the technology or its cost. It is the willingness to redesign how work is organized. This is that synthesis — and the uncomfortable twist is that the firms selling the redesign are the clearest example of not having done it yet.
01 The headline number
88% adopted. ~6% actually profit.
McKinsey's State of AI in 2025 (1,993 respondents, 105 nations) found near-universal use — but a vanishing few translate it into the P&L. The drop-off is the whole story.
The enterprise AI drop-off (% of organizations)
| Label | Value |
|---|---|
| Use AI in ≥1 function | 88% |
| Any enterprise EBIT impact | 39% |
| High performers (>5% EBIT) | 6% |
- 6% ≠ 'redesigned'. That figure measures financial impact, not redesign per se — but redesign is what produces it.
- 55% vs ~20%. High performers are ~2.8× more likely to have fundamentally reworked workflows. That gap, not tooling, is the differentiator.
- 5% of pilots. MIT's Project NANDA found only 5% of integrated AI pilots extract millions in value — despite an estimated $30–40bn invested.
Twelve reports, one verdict
Different names for the same idea — the 'agentic organization', the 'frontier firm', the 'agentic enterprise', the 'combinatorial workforce'. Every one lands on redesign over tooling.
ENDGAME
Agentic Target Operating Model
Rewire delivery, not tooling: agents build a fact base, prove on bottlenecks, hand back on 'proven rails'. Anti-pyramid by design.
McKinsey
The Agentic Organization
Five pillars; org charts become 'work charts'; cross-functional agent teams; humans 'above the loop'; governance by design.
Deloitte
State of AI 2026 · Untapped Edge
25% report transformative impact, but only 30% redesign key processes and just 21% have mature agent governance.
Bain
Foundation for Agentic AI
Systems, data, governance — 'the boring stuff that wins'. Technology is a third of the challenge; data and change are two-thirds.
Accenture
Agentic Platform Strategy
Agents are the new 'users' spanning systems. Align AI, platform and business strategy → ~2× the revenue growth of peers.
BCG
Machines That Manage Themselves
Agentic AI is 'software and colleague'. 60%+ cost cuts — but only if processes are redesigned end-to-end.
IBM
Agentic AI Operating Model
Four components with a 'Sovereign Core'. 79% of execs expect major value by 2030; only 24% feel ready — the execution gap.
McKinsey
Foundations at Scale
Infrastructure before intelligence: <10% have scaled agents; eight in ten cite data as the roadblock. 'The bottleneck is the foundation.'
Deloitte
Agentic Enterprise 2028
A five-rung autonomy ladder, operator → orchestrator. Full autonomy <1% today, projected 5–10% by 2028.
PwC
Agents Take Center Stage 2026
The 80/20 rule: technology is ~20% of value, work redesign the other 80%. Pyramid → 'diamond'/'hourglass'.
Microsoft
The Frontier Firm
'Human-led, agent-operated.' 82% of leaders plan digital labor within 12–18 months. Metric: 'return on intelligence'.
Accenture × Wharton
Humans, Agents & Robots
A new value chain of people + agents + machines. ~55% of biopharma work hours are impacted; only 11% are equipped to co-learn.
03 The 80/20 truth
The value is in the redesign, not the tool
Four firms, four different measurements — the same conclusion. Bolting agents onto legacy workflows buys marginal gains; reworking the work end-to-end is where the returns live.
Share of an AI initiative's value that comes from work redesign, not technology (%)
| Label | Value |
|---|---|
| PwC (80/20 rule) | 80% |
| Bain (⅔ / ⅓) | 67% |
- BCG: 60%+ cost cuts — but only when the process is redesigned end-to-end, not automated task-by-task.
- Deloitte: only 30% of organizations are redesigning key processes; 37% use AI only at a surface level.
- McKinsey: horizontal vs vertical. Copilots build fluency but sit loosely on top of work; value comes from 'AI inside' high-value domains.
04 The blueprint
What an agentic operating model actually is
Across all twelve reports, six design principles recur almost verbatim. Together they describe a company organized around outcomes and human+agent teams — not functions and individuals.
Structure
Flatter, more fluid, outcome-oriented. Org charts give way to 'work charts' and agentic networks of cross-functional squads.
Unit of value
The human+agent team, not the individual contributor. Value is produced by the pairing, and measured that way.
Human role
Move 'above the loop' — judgment, oversight, architecture and empathy — rather than executing inside every step.
Governance
Real-time and embedded, with human accountability by design. Critic, guardrail and compliance agents live inside the workflow.
Foundation
Modular, API-accessible, real-time data architecture — a 'mesh'. Data quality becomes a strategic differentiator.
Talent
Reskilling at scale and new roles — agent engineer, escalation specialist, trainer/tuner — the 'managers of agents'.
Autonomy is a ladder, not a switch
Every framework describes a phased climb, with the human role rising up the value stack as autonomy increases — never a big-bang replacement. Deloitte's version is the most explicit.
Assist / advise
Human = operator
Systems help and recommend; people execute the work.
Coordinate
Human = supervisor
Agents run defined tasks end-to-end; people oversee and correct.
Orchestrate
Human = orchestrator
People lead teams of agents, setting goals and adjudicating exceptions.
Self-evolve
Human = architect
A self-governing 'agent mesh' improves itself; people set direction and guardrails.
<1%
of enterprises run fully autonomous systems today
Deloitte's own estimate — the top rung is largely aspirational.
5–10%
projected to reach full autonomy by 2028
Contested by critics as overstating current maturity.
11%
of organizations are equipped for continuous human-AI 'co-learning'
Accenture × Wharton — the loop most models assume.
Special focus — the consultants’ own mirror
The firms selling this redesign face an acute version of their own advice. Their pyramid economics depend on billable junior hours — precisely the work AI compresses. Genuinely transforming delivery means less revenue per engagement under today’s model. So most firms have moved fast on internal tools and vendor alliances, and slowly on the two things that actually matter: pricing and structure.
Many of the fundamentals of the professional-services model are coming under challenge.
06 Pricing
From billable hours to outcomes
The billable hour is a proxy for value, and AI erodes it. The shift toward outcome- and software-based pricing is real — but still early, and largely narrated by executives at investor events rather than in audited filings.
AI-tied / outcome-based share of firm revenue (%)
| Label | Value |
|---|---|
| BCG 2024 | 20% |
| McKinsey (outcome-based, now) | 25% |
| Bain (AI-enabled, now) | 30% |
| BCG 2026 (expected) | 40% |
| Bain (target) | 50% |
- ~75% of McKinsey fees remain effort-based. The move to outcomes is directional, not yet dominant.
- 'Service-as-a-software.' EY and KPMG frame their platforms around paying for results, not hours.
- The client's question: 'If you advise us on AI transformation, show us how you transformed yourselves.'
07 Internal adoption
Inside, the firms already run on AI
On adopting AI for their own research and analysis, the big firms are genuinely ahead — the scale numbers are striking. It's the economic-model redesign that lags, not the tooling.
500,000+
prompts every month on McKinsey's 'Lilli' assistant
72% of the firm active; up to 30% time savings; searches 100,000+ internal documents.
~92M
prompts recorded on EY's 'EYQ'
Part of a $1.4bn investment; 150 tax agents supporting 80,000 tax professionals.
Deloitte — PairD & Zora AI
PairD rolled out to ~75,000 in EMEA; agentic 'Zora AI' targets a 25% cost reduction and 40% productivity gain.
BCG — Deckster
In-house presentation formatting and drafting, embedded in daily delivery work.
KPMG — Workbench & KymChat
~50 AI assistants (nearly 1,000 more in development); first org to achieve ISO 42001 AI-management certification.
The junior base is compressing
The classic pyramid — a broad junior base, a thin partner top — is bending toward 'diamond' and 'hourglass' shapes. The hiring data is already visible.
-44%
YoY drop in Big Four UK graduate job postings, 2025
Two senior execs estimate UK graduate recruitment could fall ~half next year.
-⅓
planned cut to PwC US graduate hiring over three years
An internal deck cites 'the impact of AI'; the firm will miss its 100,000-hire target.
11,000+
Accenture layoffs in an $865M restructuring
'Exiting' staff who can't be reskilled on AI, having reskilled ~550,000 in gen-AI fundamentals.
3 years
of frozen starting salaries at McKinsey, BCG and Bain
The Big Four haven't raised starting pay since 2022.
- 'The shape, not the size, is what's changing' — KPMG's stated approach is to have juniors 'become managers of agents' rather than vanish.
- ~470,000 employees. Deloitte's Oct 2025 Anthropic deal deploys Claude firm-wide — Anthropic's largest enterprise deployment to date.
- The apprenticeship risk. Cut the junior base too hard and you starve tomorrow's senior ranks. Redesign the pipeline; don't just shrink it.
09 The counterattack
The vendors are coming for the middle
Every major firm has struck an AI-vendor alliance — even as those same vendors launch rival deployment firms aimed squarely at the integration revenue consultants historically captured.
Fresh capital behind vendor-run 'deployment' firms & alliances ($bn)
| Label | Value |
|---|---|
| OpenAI · Deployment Co (2026) | 4$bn |
| Anthropic · deployment (2026) | 1.5$bn |
| OpenAI · Frontier Alliances | 0.15$bn |
- OpenAI 'Frontier Alliances' (Feb 2026): BCG, McKinsey, Accenture and Capgemini build certified practices; OpenAI targets 300,000 certified consultants by end-2026.
- The threat model: vendors captured ~$5–10 of services revenue per $1 of model licensing — and now want it directly.
- Partner and rival at once: the firms are certifying on the platforms of the companies building tools to disintermediate them.
How to be in the 6%, not the 88%
The reports agree on the moves — and on the thresholds that should stop you. Proceed use-case by use-case, and treat a few numbers as tripwires.
Separate the two conversations
'AI productivity' (tools, copilots) belongs with ops/IT. 'Operating model' (pricing, structure, leverage) belongs at the board. Adopting ≠ transforming.
Redesign 1–3 workflows end-to-end
Not task-by-task automation. Use PwC's 80/20 lens: budget 80% of effort for work redesign, change management and governance.
Foundation before scale
80% cite data as the blocker; <10% have scaled. Build modular, real-time, API-accessible data and embedded governance first.
Design the human layer on purpose
Move people 'above the loop'; define own / collaborate / escalate; protect the apprenticeship pipeline as you reshape it.
11 The bottom line
Redesign is the differentiator — and the unfinished work
The evidence is unusually unanimous, from a boutique practitioner to the strategy houses to the AI vendors: adoption is easy and everywhere; redesign is hard and rare, and it is what pays.
88% → 6%
from near-universal adoption to real profit
The whole opportunity lives in that collapse.
55% vs 20%
workflow redesign among winners vs the rest
The measurable difference between the two groups.
1
verdict shared by all 12 reports
Agentic AI is an operating-model shift, not a tool rollout.
Redesign the work, not just the tooling — build your agentic operating model with A4BEE.