Should I focus on mindset or skills to survive AI at work?

Both, in order. Research measuring AI at work is task-based: OpenAI's GDPval benchmark tests 1,320 professional tasks, and Microsoft's study of 200,000 Copilot conversations classifies real usage against O*NET work activities. Start by learning which of your specific tasks are exposed, then apply the adaptive mindset, judgment, supervision, and willingness to change, where the map says it matters.

What did the Goldman Sachs CIO say about AI and careers?

In a June 2026 Harvard Business Review article, Goldman Sachs CIO Marco Argenti argued that workers should stop protecting the small slice of their job AI cannot do and instead rebuild their professional identity around judgment, instincts, and supervising AI. He frames adaptation as a mindset shift rather than a skills checklist.

Is there a part of my job AI will never be able to do?

Probably not a permanently guaranteed one, and hunting for it is the wrong goal. AI capability keeps expanding into new tasks, as benchmarks like GDPval document year over year. The more useful question is which of your tasks are exposed today and which still run on your judgment, so you can decide where to invest next. That answer changes over time, which is why it is worth rechecking.

Is AI already changing employment?

Early evidence says yes, unevenly. A Stanford Digital Economy Lab working paper using ADP payroll data finds employment declines concentrated among early-career workers in occupations where AI automates tasks, while experienced workers in the same occupations have stayed stable. The authors note the evidence is correlational, not proven causal, but the task-level pattern is consistent across studies.

How do I find out which of my tasks are exposed to AI?

Take a task-level assessment rather than relying on job-title predictions. AI Job Risk Check maps your role to U.S. Department of Labor O*NET occupational data, has you rate how your time is actually spent, and scores each task's AI exposure with a transparent scoring methodology. The free score takes about 60 seconds and shows one of your tasks fully analyzed.

Mindset or Skillset? Future-Proof Your Career Against AI

Q: Who actually benefits when AI enters a workplace?

It depends on where you sit. A 2025 study in the Quarterly Journal of Economics tracked 5,172 customer-support agents and found AI assistance raised productivity about 15 percent on average, with less experienced workers improving in speed and quality while the most experienced gained little and saw small quality declines. Same job, opposite outcomes, decided at the task and skill level.

The Short Answer

Argenti's advice to a senior banker was to stop hunting for the 10 percent of the job AI can never touch and instead rebuild toward a new 100 percent built on judgment and supervision. As a mindset, that is sound. As a plan, it is incomplete. Every major research effort measuring AI at work, from OpenAI's GDPval benchmark to Microsoft's analysis of 200,000 real Copilot conversations, measures AI at the level of tasks and work activities, not job titles and not attitudes. If the evidence about AI is task-shaped, your response to it has to be task-shaped too. To future-proof your career against AI, start with a clear picture of which of your tasks are exposed, then apply the mindset shift where it counts.

What did the Goldman Sachs CIO actually argue?

In a June 2026 Harvard Business Review article, Marco Argenti describes a senior banker asking which small slice of his job he should protect because AI will never be able to do it. Argenti's answer was to let that slice go. In his words, the goal is to "resurrect professionally into a new 100%", one built on instincts, judgment, and values rather than on the habits that got you here. He compares it to an experienced horse rider learning to drive: almost none of the riding skills transfer, but the reflexes do.

There is real wisdom here. A permanently safe corner of any job probably does not exist, and clinging to one is a losing strategy. Where the argument thins out is the jump from "no task is permanently safe" to "so stop thinking about your tasks." The evidence he cites points the other way.

Why the evidence behind the mindset argument is built on tasks

To support the claim that AI is advancing quickly, Argenti points to GDPval, OpenAI's benchmark comparing AI agents to human professionals. Look at how GDPval is actually built. It covers 44 occupations from the top nine industries contributing to U.S. GDP, with 1,320 tasks crafted and vetted by professionals averaging more than 14 years of experience, and it selects those occupations using U.S. Department of Labor data: BLS wage figures and O*NET task analysis. Graders compare AI output to human output task by task, in blinded head-to-head judgments.

In other words, the strongest evidence for "AI is coming for professional work" is not a claim about professions. It is a stack of task-by-task comparisons. The benchmark cannot tell you whether "investment banker" survives. It can only tell you how AI performs on specific deliverables that make up parts of that job. The unit of change is the task. That is not our framing imposed on the evidence. It is the shape of the evidence itself.

What does real-world usage data show?

Benchmarks measure what AI can do. Usage data shows what workers actually do with it, and the picture is just as task-shaped.

Microsoft Research analyzed 200,000 anonymized conversations with Bing Copilot and classified each one against the work activities in the O*NET occupational database, the same U.S. Department of Labor data our task-by-task AI exposure report is built on. The most common activities people brought to AI were gathering information and writing. The study also draws a distinction that matters for every worker: what the user was trying to accomplish and what the AI actually performed were often different activities. AI applicability, the researchers found, cuts across sectors because most occupations contain some information work, but it lands unevenly inside each job.

Anthropic's Economic Index, which maps millions of real AI conversations to O*NET tasks, tells the same story from another platform. In its latest reports, about 49 percent of jobs in the sample had seen AI used for at least a quarter of their tasks, up from 36 percent in the first report a year earlier. Adoption is not arriving as whole jobs switching over. It is arriving task by task, a quarter of a job at a time. (In the interest of transparency: our product runs on Anthropic models, and we cite their research here because it is one of the largest open datasets of real AI usage, not because of that relationship.)

Bar graph showing the share of jobs where AI was used on at least a quarter of their tasks rising from 36 percent in early 2025 to 49 percent in early 2026, per the Anthropic Economic Index — AI adoption is arriving task by task. Data: Anthropic Economic Index.

Who actually benefits when AI enters a workplace?

This is where the research most directly complicates a pure mindset prescription. The strongest peer-reviewed evidence to date is Brynjolfsson, Li, and Raymond's study published in the Quarterly Journal of Economics in 2025. They tracked the staggered rollout of a generative AI assistant across 5,172 customer-support agents. Productivity rose about 15 percent on average, but the average hides the real finding. Less experienced, lower-skilled workers improved in both speed and quality. The most experienced, highest-skilled workers gained a little speed and lost a little quality.

Notice what happened to the novices. AI did not ask them to let go of their skills and rebuild from scratch. It transferred the tacit task knowledge of top performers to them and moved them down the experience curve faster. For the veterans, the calculus was different: the tool added less and interfered more. Same job title, same company, same AI, opposite experiences. The right move depended entirely on where each worker sat in the task and experience map.

A working paper from the Stanford Digital Economy Lab points the same direction at economy scale. Using ADP payroll records covering millions of workers, Brynjolfsson, Chandar, and Chen find that recent employment declines concentrate among early-career workers in occupations where AI tends to automate tasks, while experienced workers in the same occupations have stayed stable, and roles where AI augments work have held up. The authors are careful to call this evidence correlational rather than causal, and we hold it to the same standard. But the pattern is consistent: what AI does to a career runs through the composition of its tasks, not through the job title on the badge.

So is it mindset or skillset? It is a map, then a mindset.

Here is the honest synthesis of Argenti's argument and the research record.

Question	Mindset alone answers it?	A task map answers it?
Should I stop defending my old habits?	Yes	No
Which of my tasks is AI already doing well?	No	Yes
Where does my judgment still carry the value?	No	Yes
What should I learn next, in what order?	No	Yes
Will I keep adapting as AI improves?	Yes	No

Argenti is right that courage, curiosity, and a willingness to let habits die are the traits that carry people through technological change. But courage needs a direction. A medical transcriptionist, a bookkeeper, and a financial analyst who all adopt the same brave mindset still face three completely different sets of exposed tasks, and three different answers to "what do I do Monday morning." Telling all three to embrace a new 100 percent without showing them which parts of their current 100 percent are moving is optimism, not strategy.

That is the gap a task-level diagnostic fills. Our assessment uses time-weighted task analysis built on O*NET occupational data to show you which of your actual tasks carry AI exposure and how much of your week sits in them, using a transparent scoring methodology you can read in full. It takes about 60 seconds to get your free score on the homepage. Your score is not a prediction of job loss. It is a starting point for deciding which skills, tasks, and career options deserve your attention next. The mindset Argenti describes is what you bring to that starting point.

Get your free AI Job Risk Score. Tell us your job title and how you actually spend your time, and we will show you which of your tasks are exposed today. Free. 60 seconds. No sign-up required.

Get My Free Score

Sources

AI Job Risk Check uses task data from O*NET, provided by the U.S. Department of Labor, Employment and Training Administration (USDOL/ETA), used under the CC BY 4.0 license and modified by Phronesis Labs LLC. USDOL/ETA does not endorse this product.