QR code
here
here
Future of Work
in an AI-enabled World
QR
code
What I'm going to walk you through today rests on a much broader piece of work from Dragonfly: Future of Work in an AI-Enabled World. It's a structural analysis of how AI is reorganising work across three scales — workers, firms, and countries — over the next five to ten years. The full set of interactive visualisations is behind the QR code. I'd genuinely encourage you to scan it. Today I'm going to focus on the part of that work that bites the investment industry hardest — investment analysis. The visualisations on this slide are the main four you'll find when you go and explore. First — where AI-pressure lands and how fast. The key determinant is whether it is Head work versus Hand or Heart work, and within Head work, how verifiable it is. Second — the worker-shape archetypes; the diamond-complete framing I'll use later sits on top of an empirical four-pointer model. Third — what drives K-shape divergence at worker, organisation, and country level. Fourth — country-cluster pace-layer alignment; most of you in this room sit in what we call the mid-tier-laggard cluster, currently carrying a five-to-ten-year structural deployment-depth drag relative to sovereign-fast economies. That's the foundational thinking. We then ran a Dragonfly robustness workflow on top of it specifically focused on implications for asset-owners — and that workflow's output is what we'll come to in the second half of this session. But first — the three things from this work that I want you to leave the room with.
code
The visual essay
Future of Work visualisations
Full set of visualisations behind the QR.
01 · Pressure
02 · Workers
03 · K-Shape
04 · Countries
Anthea Roberts and Sue Brake, 2026
The reframe
AI is a capability to compound
not a tool to deploy
Three things change.
Shift one
The work
Where AI bites first — and how the team pyramid restructures around what AI can't do.
Shift two
The shape
What shape of worker compounds with AI capability — and what shape erodes against it.
Shift three
The pipeline
How the next generation of senior judgment is made when AI is doing the apprentice's work.
Anthea Roberts and Sue Brake, 2026
The work
The team pyramid restructures around what AI can't do
Verifiability predicts whether AI substitutes or augments
Verifiable — the loss column
Cost-out, not edge
Analyst-weeks collapse to minutes. Already in production.
Analyst-weeks
AI-minutes
Comp builds & screening
Basic credit work
ESG flags
DD questionnaires
Transactional legal & filings
Standard valuation mechanics
Unverifiable — the gain column
Where alpha lives
AI doesn't substitute it. AI compounds it — for the right shape of worker.
Human judgment
baseline
× AI
compounds
Regime calls
Illiquid sourcing
Active stewardship & engagements
Manager selection · trust over time
Counter-cyclical allocation
Mandate design under deep uncertainty
November 2025 — fully-verifiable code inflection · five independent witnesses (Willison · Cherny · Cat Wu · Vo · Schoening)
Anthea Roberts and Sue Brake, 2026
The worker
Diamond-complete is the new T-shape
Focus on hiring or building T-shape with AI fluency
I-shape
Deep specialist, no AI
Generalist
Breadth, no anchor
T-shape
What we used to hire for
Upside-down T
Generalist + AI, no anchor — useful catalysts
Diamond-complete
T-shape + AI fluency. What compounds.
Anthea Roberts and Sue Brake, 2026
The team
The team pyramid restructures into one of four shapes
The Diamond is the design target. The Barbell is the failure mode.
Traditional Pyramid
today's structure — juniors subsidise seniors
The Obelisk
same shape, dramatically narrower
The Barbell
missing generation in the middle
✗ failure mode
The Diamond
orchestrators in the widest band, working alongside AI
★ design target
The T
thin human layer over an automated stack
Roberts & Wilkins (2026), "The Pyramid Transformation" · adapted from the Harvard Law School project
Anthea Roberts and Sue Brake, 2026
The pipeline
Don't let the team become a barbell
Three answers — and one only this room can fund
01
Anchor depth is being redefined, not eliminated.
From “I can do this task better than the model”
to “I know where the model fails in my corner of the world, and I have the trained intuition to catch it before it costs us.”
to “I know where the model fails in my corner of the world, and I have the trained intuition to catch it before it costs us.”
02
The apprenticeship architecture has to be deliberately preserved against AI's optimisation pressure.
Three-way deliberate practice — junior generates, AI critiques, senior evaluates. AI as sparring partner, not deliverable producer.
03
The credentialing standard is undefined — and only this room can fund it collectively.
No university, no professional body, no vendor produces it. Three or four large funds in this room — partnered with the CFA Institute or its equivalent — could move the needle in a way no other actor in the system could.
Anthea Roberts and Sue Brake, 2026
Recap
Three messages. Three responses.
01
The question
Where does AI bite first?
The response
Verifiability — and it restructures the team pyramid.
Verifiable work collapses to AI-minutes; unverifiable work is where alpha lives.
02
The question
What shape compounds rather than erodes?
The response
Diamond-complete at the worker level; the Diamond at the team level.
Don't become a Barbell. Pair every upside-down-T hire with a senior specialist as a catalyst.
03
The question
How do we develop the next generation when AI is doing the apprentice's work?
The response
Preserve the apprenticeship architecture. A new credentialing standard would benefit the whole industry.
Diamond-complete is currently undefined — no university, professional body, or vendor produces it.
Anthea Roberts and Sue Brake, 2026
Robustness Workflow
Dragonfly & Analyst
CTQ Framing
Four Scenarios (A)
Finds drivers
Prioritises drivers
Chooses & builds most impactful scenarios
✓
boundary
+ extract
+ extract
Systems Mapping (B)
Determines network connections
Finds loops & tipping points
Finds synergies & trade-offs
Synthesises emergent insights
✓
boundary
+ extract
+ extract
Robustness (C)
Stress-tests strategy across scenarios
Designs signposts with triggers
Determines scenario probabilities & final synthesis
✓
boundary
On request
Committee Paper
draft · committee-ready
analyst
Results Explorer
Q&A on analysis & methodology
Reports on demand
analyst
Interactive Dashboard
the analyst's window
{ }
JSON Manifest
source of truth · machine-readable · agent analysis preserved
Scenarios.json
Systems.json
Robustness.json
Audit Log
·
·
·
boundary checks · clean-run record
✓
Analyst
sign-off
sign-off
Governance Committee
Governance
Decision
signed · defensible
Re-frame? Rerun the workflow.
Anthea Roberts and Sue Brake, 2026
What we tested against
Four future-of-work scenarios
Workers atomised
Workers organised
Q2
Negotiated Displacement
Substitution still proceeds, but bargaining captures a share of the surplus and shapes deployment terms — guild, EU sectoral, and co-determination variants.
~12%
Q4
Co-Pilot Compact
Augmentation under co-determination and bargaining — AI funds shorter hours and care-sector wages. Only works if the metacognition gap is also addressed.
~14%
Q1
Hollowed Middle
Substitution at scale. Mid-skill knowledge work collapses, gains concentrate in capital and elite professions, populist backlash without binding response.
~8%
Q3
Productivity Without Power
AI augments rather than substitutes. Productivity rises, wages stagnate, gains accrue to capital. The 1990s decoupling pattern intensifies.
~16%
← AI substitutes
AI augments →
As of May 2026 — both axes near midpoint. ~50% probability sits in the central zone. The corners are individually unlikely but collectively define the design space.
Anthea Roberts and Sue Brake, 2026
From the workflow
Three insights I haven't had time to land
01
Comprehension governance debt is accumulating silently.
The transformation paradox gap is its leading indicator. 2–4 year window before failure event. Measurable metric installable on Monday.
02
You hold both sides of the AI trade.
Capital owner of the disruption. Fiduciary for the disrupted. The first three peer funds to publish an AI-deployment stewardship policy will set the floor.
03
Q2 Negotiated Displacement is not the most likely scenario, but it's the right design case.
The only future where all four operating-model trade-offs stretch at once. Designing for it covers the others as a functional superset.
→Ask Dragonfly-AI about any of them in the next ten minutes — it has loaded all of the robustness analysis as well as Dragonfly's broader Future of Work in an AI-Enabled World context.
Microsoft 2026 WTI · transformation paradox triplet: 65% fear falling behind · 13% rewarded for AI reinvention · 26% see leadership alignment.
Anthea Roberts and Sue Brake, 2026
Dashboard snapshot
The dashboard
Insert Dashboard Snapshot here
Anthea Roberts and Sue Brake, 2026
What we tested against
Scenario pathways — surplus, alternative view
As of May 2026 — both axes near midpoint. ~50% probability sits in the central zone. The four corners are individually unlikely but collectively define the design space.
Anthea Roberts and Sue Brake, 2026