AI has expanded what technology can do. But adoption is not transformation, productivity is not automatically value, and a better model does not automatically create better work.
Adoption tells us whether people used the technology. Transformation tells us whether the work, decisions, and outcomes actually changed.
The real progression is not a login, a license, or a prompt count. It is the movement from activity to value:
Usage
Behavior change
Workflow change
Operational outcome
Value and impact
AI changes what is technically possible. It does not eliminate the need to understand the problem, redesign the work, assign accountability, measure outcomes, and learn from reality.
PRACTICAL FRAMEWORKS
Use the questions to redesign the work.
WORK-AI FIT TEST
Where does AI belong in the workflow?
AI performance depends on the task, the worker, the context, the workflow, the system, and the consequences of being wrong. Use these questions before calling a task AI-ready.
01Is the output objectively verifiable?
02What happens when it is wrong?
03How much hidden context does the human already possess?
04Is this task inside or outside the model's demonstrated capability frontier?
05Does AI improve the work, or merely add another interaction?
The capability frontier moves as models, retrieval, context, tooling, and workflow design improve. Retest assumptions over time.
MONDAY MORNING AI TEST
Is the initiative changing how we work to create measurable incremental value?
Use this check in product reviews, transformation meetings, roadmap discussions, and leadership reviews when an AI initiative is consuming meaningful time or attention.
01What measurable outcome should change?
02What work or behavior has to change for that outcome to move?
03What should become unnecessary, automated, improved, or newly possible?
04Where does consequence require human judgment?
05How will the system know whether it worked?
Did how you work change? Or just the tools?
RESEARCH & CITATIONS
The task, population, and context matter.
There is no universal “AI productivity” number.
The studies below examine different populations, tasks, models, time periods, and work environments. They should not be treated as directly comparable estimates of one universal effect.
The contrast is the point: AI impact depends on the work and the context.
McKinsey & Company: The State of AI in 2026
The survey behind the slide's approximately 9 in 10 adoption figure, 8 in 10 productivity figure, and approximately 4 in 10 enterprise financial contribution figure.
Dell'Acqua et al.: The Jagged Technological Frontier
The research behind the task-dependent frontier described in the deck, including faster inside-frontier work and weaker performance on the selected outside-frontier task.
Impact increases, reversibility decreases, uncertainty increases: governance should get stronger.
This is Ashish Sharma's practitioner shorthand, informed by risk-proportionate approaches in established AI governance frameworks. It is not presented as a formula from NIST or the EU AI Act.
Low consequence
Baseline controls
Access control
Ownership
Basic logging
Normal product review
Material consequence
Active monitoring
Evaluation thresholds
Quality monitoring
Exception queues
Escalation rules
High consequence
Hard boundaries
Pre-action approval
Programmatic fail-safes
Independent testing
Audit and rollback
QUESTIONS THAT LAST
AI changed the tech. Not the questions.
The tech
•Context windows
•Token optimization
•Vector embeddings
•Model drift
•System prompts
•MCP / A2A
•Agent registries
•Model benchmarks
The work
•What problem are we solving? For whom?
•What behavior or workflow must change?
The accountability
•What outcome should move?
•What must not get worse?
•How will we know it worked?
QUICK ANSWERS
Questions from the talk.
What does Beyond AI Adoption mean?
AI adoption measures whether people are using a technology. Moving beyond adoption means determining whether the technology changes how work is performed, how decisions are made, where human judgment is applied, and whether measurable operational, customer, or business outcomes improve.
Why is AI adoption not the same as AI value?
Adoption measures use. Value requires evidence that behavior or workflows changed and that those changes produced measurable operational, customer, or financial outcomes.
What is the jagged technological frontier?
The jagged technological frontier describes how AI can perform extremely well on some tasks while performing poorly on seemingly adjacent tasks. The boundary is uneven and changes as models, tools, context, retrieval, and workflows improve.
What is the Work-AI Fit Test?
The Work-AI Fit Test is a five-question practitioner framework for evaluating whether a specific task or workflow is a good fit for AI and where human judgment should remain.
What is the Monday Morning AI Test?
The Monday Morning AI Test is a five-question framework for evaluating whether an AI initiative is producing real transformation rather than merely adoption or activity.
How should organizations govern AI risk?
Governance should scale with consequence. As potential impact increases, reversibility decreases, or uncertainty rises, controls should become stronger.
ABOUT ASHISH SHARMA
The work after the talk.
Ashish Sharma is a technology and product leader focused on artificial intelligence, decision intelligence, product strategy, and the organizational realities of putting AI into production.
He writes about the practical challenges of implementing AI, from workflow redesign and adoption to governance, economics, accountability, and the human consequences of technological change.
The Work-AI Fit Test, Monday Morning AI Test, and practitioner governance shorthand are frameworks developed for this presentation by Ashish Sharma. External research findings are attributed and linked to their original sources.
Connect on LinkedIn