The weekly paper
for curious people
See the connections.
Understand the consequences.
Swipe to see more sections →
Sick of fake news?
Same evidence. Different perspective.
Viewing through
Start with the shortest supported answer, then follow the evidence and the limits together.
What comes into focusIt keeps the claim, its source and the point where certainty ends in one view.
What this view may missA specialist Lens can make one practical consequence easier to see.
Changing the Lens changes what comes into focus. It never changes the evidence underneath.
Issue 003 Analysis 02 · Work and expertise
AI can help newcomers do specialist work and learn faster. It may also remove some of the work through which people once developed judgement. So who keeps the history, catches the exceptions and takes responsibility?
What people heard
AI will make everyone a generalist and specialists obsolete.
Not established
Is task capability the same as expertise?
Short answer. AI improves performance on some tasks and can help newcomers learn. It does not make everyone an expert or remove the need for feedback, history, responsibility and human judgement.
Choose how to read this
The facts do not change. Each view uses the same published sources and leaves the same questions open.
See through another Lens
Each view notices something useful. None is allowed to stand in for the complete evidence.
Worker
AI can help one person do more without making them an expert in every field.
What this view explainsWhy AI performs well on some tasks and poorly on others.
What it may missOne short-term productivity gain cannot show how expertise develops over years.
Current understanding
In some settings, AI improves performance and helps newer workers most.
Practice, feedback and refinement matter. Time served alone is not enough.
The risk is plausible, but AI can also help people learn. The evidence does not settle it.
Lens proposes clear responsibility for keeping a field's history and connecting work across teams. These are ideas to test, not proven jobs of the future.
How the pieces connect
The diagrams show which is which. Two events appearing close together does not mean one caused the other, and a risk is not the same as a finding.
More people can complete some specialist tasks.
Someone keeps the history, exceptions and open questions.
Someone watches what falls between teams.
People remain responsible for approval and review.
Leaders choose the purpose and accept the consequences.
A proposal, not a forecast. The full design has not been tested.
Tasks were never the whole job
An AI can research a rule, compare documents and draft an analysis. An experienced specialist may also remember the incident that created the rule, the option that failed, the advice that changed and the exception that never reached the manual.
That memory can be valuable, incomplete, outdated or wrong. Time served is not proof of expertise. Research shows that focused practice, feedback and repeated correction matter, but they explain only part of why some people become experts.
Lens’s argument is simple: expertise grows when knowledge meets real results and real responsibility. That is our interpretation, not a scientific definition.
The productivity trap
If AI removes junior work, it may also remove some of the moments when people encounter patterns, make mistakes and learn under supervision. Economic and labour-market papers raise that risk, but they do not prove an economy-wide shortage of experts is coming.
There is strong evidence in the other direction too. Studies in customer support, financial-services training and a small apprenticeship experiment show AI can help newcomers perform and learn.
The risk is real enough to design for. It is not destiny.
Redesigning apprenticeship
Future apprentices could review AI decisions, investigate unusual cases, compare predictions with outcomes, revisit earlier decisions and challenge accepted advice under human supervision.
AI can help preserve context and reveal patterns. It cannot make an old expert opinion correct simply because an experienced person recorded it.
People still need permission to disagree, test the advice and change it when the evidence changes.
Lens proposal
That person would make sure the organisation can answer basic questions: What do we currently believe? What evidence supports it? Who decided? What changed? What went wrong? What remains unknown?
Mentors, editors, knowledge brokers and other existing roles already do parts of this job. Putting all of that responsibility in one place is still an untested idea.
The important question is no longer only who can do this task. It is who makes sure the organisation still understands the field.
The connection role
Regulation, operations, finance, safety and community impact can each be understood well while the organisation still misses a problem where those areas meet.
This person would keep track of the links, disagreements and unanswered questions between teams without overruling the experts in each one.
The future may need people who can do more kinds of work and people who protect deep understanding. Those are not opposing camps.
What we still do not know
Show me the evidence
We separate established facts, reported claims, our own analysis and questions the evidence cannot yet answer.
Quarterly Journal of Economics
Organization Science
IZA discussion paper
Intelligent Decision Technologies
Academic Emergency Medicine and Cochrane review
Psychological Science meta-analysis
IESE and IZA working papers
Research on how organisations share knowledge
Lens proposal
Lens proposal
Lens evidence assessment
The question
Find out what AI helps newcomers do, whether it weakens the path to expertise, and who should keep the knowledge behind the work.
Evidence checked
We checked workplace studies, research reviews, small experiments and economic papers available through 20 August 2026. They do not prove that experts will disappear or that apprenticeships will collapse across the economy.
Best-supported answer
AI can help newcomers do some specialist tasks and learn faster. That does not make everyone an expert, and it does not prove specialists will disappear. Organisations still need people who remember the history, check the evidence, connect problems across teams and take responsibility for decisions. Lens suggests making those responsibilities explicit. The idea still needs to be tested.
Why we told it this way
The evidence lets us compare what AI can do now, how people learn and what may be lost when junior work disappears. It does not support a certain prediction about the future of work.
The evidence supports both points but cannot tell us whether enough experts will exist in the long term.
Similar work already exists in some organisations, but this complete model has not been tested.
Checked 28 August 2026 Sources, disagreements and unanswered questions remain visible.
Possible effects · We cannot say how likely
AI-assisted productivity could widen access to specialist tasks while increasing the value of explicit responsibility for history, evidence, cross-team connections and judgement. It does not show that experts will disappear or that the proposed responsibilities will work.
Studies found gains for some novice workers, uneven performance across tasks, and continuing importance for practice, feedback and knowledge-sharing roles.
If routine junior work changes, organisations may need to protect the feedback and responsibility through which future experts learn and difficult cases are judged.
Where the connection stopsThe evidence supports task-level gains and risks, not an economy-wide forecast or proof that Lens's proposed roles improve outcomes.
How people may respond
How AI-assisted task performance is interpreted could change hiring, training and responsibility before long-term expertise effects are known.
What the evidence does not showTask-level gains and a jagged frontier are documented; economy-wide skill formation and the proposed roles remain untested.
Productivity gains are presented beside feedback, practice and task limits.
Organisations may use AI to accelerate work while preserving review and progressively difficult learning.
Attention remains on transfer, judgement and performance without assistance.
Employers, educators and professionals: Design work that combines assistance with feedback and accountable review.
AI-supported capability and future expertise may develop together.
What we know has changedWe have not established that this possible change has happened.
What this does not showShort-term performance does not establish durable learning.
This is the first time Lens has mapped this path. We have no later evidence showing whether it is happening more, less or about the same.
Visible output gains dominate less visible apprenticeship and exception handling.
Organisations may remove junior work or expert oversight before replacing its learning function.
Attention moves to current throughput while future capability receives less weight.
Employers and labour-market institutions: Restructure roles, training and staffing around assisted output.
A short-term design choice could commit the organisation to a weaker expert pipeline.
What we know has changedWe have not established that this possible change has happened.
What this does not showThe risk is plausible, not an economy-wide forecast.
This is the first time Lens has mapped this path. We have no later evidence showing whether it is happening more, less or about the same.
If AI-supported work retains feedback, varied cases and accountable expert review
Then newcomers could become productive sooner while still developing judgement and recognising model limits.
Would weaken this: Performance collapses outside the assisted task. Feedback and progressively difficult work disappear.
Scope: Specific work settings, not the whole labour market. Horizon: Across training and career progression.
If organisations automate formative junior work without replacing its learning function
Then short-term output could improve while the later supply of people able to handle exceptions and accept responsibility weakens.
Would weaken this: New learning pathways produce equivalent or better expertise. Expert supply remains stable after sustained adoption.
Scope: The apprenticeship mechanism, not a forecast of mass expert disappearance. Horizon: Several training and promotion cycles.
A large customer-support field study found productivity gains from generative AI, with larger gains for novice and lower-skilled workers in that setting.
Open evidence ↗In a field experiment, consultants using AI performed much better on some tested tasks but worse on a task the AI handled poorly.
Open evidence ↗Expertise research and a large audit-and-feedback review support focused practice, feedback, evaluation and repeated refinement as contributors to professional performance.
Open evidence ↗Economic and labour-market papers warn that AI may change how experienced workers teach newcomers and may raise the bar for entry-level jobs.
Open evidence ↗Research describes mentors, editors and other people whose job includes sharing knowledge and connecting work across teams.
Open evidence ↗What could change this assessment?