AI is making people faster at work. But faster employees do not necessarily produce better business results.
That was one of the central arguments in a presentation by Arvind Karunakaran, Assistant Professor of Management Science and Engineering at Stanford University.
Karunakaran shared research on how generative AI changes job responsibilities, what happens when employees use AI to work outside their areas of expertise, and why some organizations struggle to convert individual productivity gains into measurable value.
His findings are particularly relevant to CEOs and CFOs trying to understand the returns on enterprise AI investments.
Open Future Forum's 2026 research has identified a growing gap between AI adoption and financial accountability. Karunakaran's work helps explain what may be happening inside organizations to produce that gap.
Nearly half of occupation-specific AI use crosses traditional job boundaries
One of the most striking slides drew on OpenAI Economic Research's July 2026 Work at the Frontier study.
The research examined more than 800,000 work-related ChatGPT messages from US users.
It found that 43.5% of occupation-specific messages involved tasks traditionally associated with another occupation. Across all work-related messages, including generic activities, the figure was 16.8%.
This is a significant development. Employees are increasingly using AI to undertake activities that previously might have required help from colleagues in another department.
The pattern differs substantially by occupation.
| Occupation | Share of occupation-specific AI messages involving work outside the occupation |
|---|---|
| Customer experience | 77% |
| Design | 75% |
| Human resources | 69% |
| Legal | 56% |
| Marketing | 53% |
The analysis also found that marketing and engineering tasks frequently appeared in the AI use of people from other professions.
For example, 28% of occupation-specific messages from designers were classified as engineering tasks, while another 28% were classified as marketing tasks.
Among finance workers, 25% were classified as marketing tasks and 22% as engineering tasks.
These figures describe how AI is being used, not whether the work meets professional standards.
That distinction matters.
An employee using AI to perform an unfamiliar task is not necessarily acquiring the expertise traditionally associated with that work.
How this compares with Open Future Forum's research
Our CMO AI Leverage research found that marketing executives increasingly see AI as a way to expand team capabilities.
In the August cohort of 127 responses:
- 50% identified doing the work of more people as an important opportunity.
- 43% identified understanding customers better.
- 43% identified creating content faster.
The findings are consistent with expanding occupational boundaries.
However, our survey measures executive perceptions of opportunity, while OpenAI's research measures patterns of AI use.
Together, they suggest that job responsibilities are expanding before organizations have necessarily redesigned their formal roles, training, or performance systems.
The AI Wall Effect: AI makes people faster, but expertise still matters
Karunakaran also discussed the GenAI Wall Effect, research conducted with colleagues from Stanford and Harvard Business School.
The study involved 78 workers at IG Group, a financial services company.
Participants were divided into three groups: 12 web analysts, 26 marketing specialists, and 40 technology specialists.
Researchers examined whether generative AI could help employees perform work traditionally undertaken by web analysts, including preparing investment-related articles.
AI substantially reduced the time required to complete the work.
| Task | Without AI | With AI | Time reduction |
|---|---|---|---|
| Article conceptualization | 63 minutes | 23 minutes | 63% |
| Article writing | 87 minutes | 22 minutes | 75% |
But the quality of the final work depended on the participants' professional backgrounds.
| Professional group | AI-assisted article quality, out of 5 |
|---|---|
| Web analysts | 3.96 |
| Marketing specialists | 3.92 |
| Technology specialists | 3.42 |
Technology specialists scored approximately 13% below the web analysts despite having access to the same AI assistance.
The difference was much smaller during the conceptualization stage, where all three groups performed at broadly similar levels.
The finding suggests that AI can help employees plan unfamiliar work more readily than it can compensate for missing domain expertise during execution.
Karunakaran described the broader pattern as an AI Wall Effect: technology can raise the performance of less experienced workers, but there are limits to how far it can transfer professional expertise.
For employers, this creates an important distinction between expanding employees' capabilities and replacing specialist knowledge.
Where Open Future Forum agrees
Our research has consistently found that AI adoption and business value should be measured separately.
The GenAI Wall Effect provides a more detailed explanation for that distinction.
A company may achieve substantial reductions in task completion time without achieving equivalent improvements in the quality of its products or services.
The experiment also suggests that the value of professional judgment may increase as AI makes routine execution cheaper.
That is an important question for executives planning workforce redesign.
Why automating one task can damage another
Karunakaran argued that jobs are more than collections of independent tasks.
They also contain what he described as invisible glue: the context, learning, judgment, and accountability that connect different responsibilities.
Some jobs have relatively weak connections between their tasks. These can be easier to divide between people and AI.
Other jobs contain tightly connected activities. Removing one responsibility may undermine an employee's ability to perform another.
Karunakaran discussed research involving radiation oncology professionals, where automating aspects of image interpretation reportedly affected performance on other responsibilities over time.
The proposed explanation was that performing the original task helped develop the contextual knowledge needed for subsequent decisions.
The broader lesson is relevant beyond healthcare.
A junior lawyer may develop judgment through document review. An engineer may learn architecture through writing and debugging code. A finance professional may develop commercial understanding through detailed financial analysis.
Removing these activities could save time while weakening the experience required for more senior responsibilities.
The challenge is therefore not simply identifying tasks AI can perform.
It is determining which tasks can be separated without damaging the rest of the job.
Karunakaran proposed examining four characteristics of work:
Bundles: How strongly are tasks connected by learning, context, and accountability?
Chains: Are AI-capable activities grouped together, or are they separated by steps requiring human involvement?
Bottlenecks: Which activity limits the quality or throughput of the entire workflow?
Vacancy chains: If AI removes an activity from one job, where do the remaining responsibilities move?
This framework is particularly useful for organizations considering restructuring around AI.
What Open Future Forum's research adds
Our enterprise AI research has concentrated on purchasing authority, budgets, and accountability.
Karunakaran's framework examines a different question: how work should be reorganized after the technology has been purchased.
The two approaches are complementary.
A CEO can approve an AI investment and a CFO can establish a financial target, but neither decision establishes whether the underlying workflow has been redesigned effectively.
The operating model matters as much as the purchasing decision.
Individual productivity is not organizational productivity
Another important argument from the presentation was that AI can improve individual performance while reducing the efficiency of a team.
Consider a software engineer who produces more code using AI.
That may look like a productivity improvement.
But if the additional code requires more review, debugging, and integration work, the engineering organization may not ship more features.
The same issue applies to marketing, legal work, financial analysis, and customer service.
More documents, more content, and more completed tasks are not necessarily equivalent to better business outcomes.
Karunakaran argued that organizations should redesign and evaluate entire workflows rather than focusing exclusively on individual tasks.
Open Future Forum's ROI findings
Our September 2026 enterprise AI research found that financial accountability is becoming more important as AI adoption expands.
In the August finance cohort, 65% identified proving ROI as a principal blocker, compared with 53% in the earlier cumulative comparison.
The same research found that CFO or finance involvement in AI purchase approval reached 43%, compared with 33% previously.
This suggests that finance executives are becoming more involved precisely as companies face greater pressure to demonstrate economic returns.
The comparison with Karunakaran's findings is revealing.
His research identifies ways in which productivity improvements can fail to translate into organizational performance.
Our research identifies the executives increasingly responsible for determining whether those improvements justify the investment.
Both point toward the need for better measurement.
CEOs and CFOs are looking at different evidence
Open Future Forum's research found a substantial difference between CEO and CFO expectations for AI payback.
70% of CEO-seat respondents expected measurable AI payback within six months, compared with 42% of finance respondents.
We described this 28-percentage-point difference as the Optimism Gap.
There are several possible explanations.
CEOs may place greater weight on strategic opportunities, competitive positioning, and increased organizational capability.
CFOs may place greater weight on measurable financial effects, implementation costs, and the timing of returns.
Karunakaran's research suggests another possibility: organizations may be observing genuine individual productivity gains without having established how those gains affect the overall business.
His proposed approach separates AI measurement into three levels.
| Measurement level | What matters |
|---|---|
| Tier 1: Adoption and absorption | AI usage, experimentation, and capacity redirected toward named activities |
| Tier 2: Operational outcomes | Workflow throughput, decision speed, bottlenecks, and handoffs |
| Tier 3: Financial outcomes | Revenue, costs, margins, or other measurable P&L effects |
This is a useful discipline for CFOs.
The company must determine whether those hours resulted in lower costs, increased output, better customer outcomes, or some other measurable benefit.
It must also account for implementation, supervision, and downstream costs.
Leadership can change AI adoption by more than 50%
One of the clearest quantitative findings concerns how managers introduce AI to employees.
Karunakaran and colleagues studied AI adoption in organizations including a law firm, an advertising agency, and an IT services firm.
The research included interviews with 184 employees.
At the law firm, researchers compared two divisions using the same AI tool.
One division was encouraged to use AI primarily to improve productivity and complete existing tasks faster.
The other was encouraged to explore how AI could make work more interesting and enable employees to undertake more valuable responsibilities.
The second division recorded:
58% higher AI usage and 70% more experimentation.
These differences are substantial.
However, the divisions also differed in the support available to employees. The job-enrichment group received additional training, knowledge-sharing opportunities, and time to experiment.
The evidence therefore supports the importance of a credible management approach, not merely changing the language used to describe AI.
Employees need to see that new responsibilities and development opportunities are real.
A further connection with Open Future Forum
Our research shows that enterprise AI decisions are increasingly concentrated among senior executives.
In the September synthesis, 47% of finance-lane operators identified the CEO as the AI purchase signer.
But purchasing authority is only one part of implementation.
Karunakaran's findings suggest that managers also need to create conditions in which employees are willing to experiment, develop skills, and take responsibility for new work.
This adds an important operational dimension to our findings on executive accountability.
Where the research agrees, and where questions remain
There is considerable alignment between the two research programs.
Both suggest that AI adoption should not be treated as a sufficient measure of success.
Both point toward the importance of management decisions, performance measurement, and organizational accountability.
But the research also has different limitations.
Karunakaran's experimental findings examine specific tasks, organizations, and professional groups. They should not be generalized automatically to every industry or AI deployment.
Open Future Forum's surveys capture executive experience and expectations across multiple functions, but they do not directly measure the productivity of employees performing specific tasks.
Our Investor AI Report, for example, found that 86% of surveyed investors identified at least one measurable AI effect across their portfolios.
That is encouraging evidence of perceived business impact.
It is not the same as demonstrating that AI has increased profitability across 86% of portfolio companies.
Karunakaran's findings reinforce the importance of that distinction.
A measurable improvement may be real without being financial. A financial improvement may be real without being attributable entirely to AI. And a productivity improvement may be offset by additional work elsewhere.
These are not reasons to slow AI adoption indiscriminately. They are reasons to measure it more carefully.
What enterprise leaders should take away
The practical implication is that AI strategy needs to move beyond tool deployment.
Organizations should identify commercially important workflows, understand the dependencies between tasks, and determine where AI can improve performance without weakening professional judgment.
They should update employee responsibilities and performance measures together.
They should give employees credible opportunities to develop new skills rather than relying exclusively on productivity targets.
And they should connect operational improvements to financial outcomes with clear executive ownership.
Open Future Forum's research has shown that authority over AI purchasing is increasingly concentrated among CEOs and CFOs, while budgets and ROI measurement remain unresolved in many organizations.
Karunakaran's work provides a valuable explanation for why those challenges persist.
The technology can make individual tasks faster without making the organization more effective.
Closing that gap requires changing how work is organized, how people are evaluated, and how financial value is measured.
For CEOs and CFOs, the next challenge is not simply deploying more AI. It is proving that the organization performs better because of it.
Research sources
- Arvind Karunakaran, Stanford University, presentation on AI deployment and organizational redesign, October 2026.
- OpenAI Economic Research, How AI Is Expanding What People Do at Work, July 27, 2026.
- Vendraminelli, DiSorbo, Hildebrandt, McFowland, Karunakaran and Bojinov, The GenAI Wall Effect, Harvard Business School Working Paper 26-011.
- Stanford HAI, Want Employees to Embrace AI? Stop Selling It as a Productivity Tool, October 7, 2026.
- Open Future Forum, Enterprise AI in 2026: Why Ownership Is Consolidating Faster Than Accountability, September 2026, together with its CMO AI Leverage and Investor AI research.
Editorial note: The radiation oncology example is attributed to the presentation because the underlying study has not yet been independently checked. Open Future Forum survey findings reflect their respective cohorts and are not population-wide estimates.
Last updated: October 31, 2026
Frequently Asked Questions
See the Data Behind the Optimism Gap
Open Future Forum's AI Leverage reports track who signs AI purchases, how budgets are set and when executives expect payback, across CEOs, CFOs, CMOs, CISOs and investors.