> ## Content Index > Fetch the complete content index at: https://www.groktop.us/llms.txt > Use this file to discover other available public pages before exploring further. # The Best AI Companies Are Hiring for More Judgment - URL: https://www.groktop.us/hiring-for-judgment/ - Published: 2026-09-09T15:44:52.000Z - Updated: 2026-09-09T15:44:52.000Z - Description: The strongest AI adopters are not simply buying tools. They are changing what human judgment is worth. - Author: Magnus Hedemark The loudest AI jobs story says the machine is coming for everyone. The early firm-level evidence points somewhere more interesting: companies making the heaviest AI investments grew their workforces, while low-intensity adopters saw no statistically significant change. A [Ramp and Revelio Labs analysis of more than 21,000 US firms](https://ramp.com/data/heavy-ai-adopters-hire-more?ref=groktop.us) found 10.2% headcount growth among high-intensity adopters in the two years after adoption. That doesn't prove AI created those jobs. Heavy adopters were already larger, more engineering-intensive, more likely to be venture-backed, and growing faster. Some of the growth may belong to the companies, not the tools. Still, the result cuts against the simple replacement story. In companies capable of turning AI into new capacity and new value, the first visible labor-market signal is expansion. This is chapter four of a five-part argument about what happens to human work when AI gets serious. Jobs broaden first. Organizational boundaries start to move. Workers need agency over the time and decisions that come back to them. Then the labor market responds. It starts rewarding people who can use AI without handing their judgment to it. The claim needs a boundary, but the pattern is useful. The data doesn't support the simple story that AI takes jobs everywhere. It shows something more specific: [high-intensity adopters are expanding while low-intensity adopters show no statistically significant employment gain](https://ramp.com/data/ai-jobs-impact?ref=groktop.us). A separate evidence chain shows employers paying more for AI fluency and asking for judgment, creativity, leadership, and strategic work. Taken together, the evidence points toward a distinction worth testing. Companies that turn AI into new capacity may expand. Companies that use it mainly to squeeze existing work may not create the same demand for labor. The studies don't prove that mechanism. They give us a reason to look for it. ## The hiring signal is real, but bounded The Ramp and Revelio result punctures the neatest version of the automation story. Entry-level headcount among high-intensity adopters climbed 12% over 24 months. Entry-level share also rose 1.15 percentage points relative to the comparison group, according to the [working-paper summary](https://ramp.com/data/ai-jobs-impact?ref=groktop.us). Most functions looked broadly similar or net-neutral. Junior hiring was the exception worth noticing. No, we can't infer that every new job demanded more judgment. But this sample doesn't look like a substitution machine where software arrives and junior work vanishes on cue. It adds weight to the case that [headcount reduction was always the wrong AI objective](https://www.groktop.us/the-ai-hiring-reversal-why-headcount-reduction-was-always-the-wrong-goal/). The more useful question is what the company does with the capacity AI creates. A serious AI program can sit beside growth when leaders use that capacity to do more, not only to ask fewer people to carry the same operation. The global data tells a similar story. The [2026 PwC Global AI Jobs Barometer](https://www.pwc.com/gx/en/news-room/press-releases/2026/pwc-2026-ai-jobs-barometer.html?ref=groktop.us) puts headcount at the most AI-exposed companies 52% above its 2018 baseline in 2025\. The least exposed companies reached 36% above baseline. Productivity grew faster in the more exposed group too, with the widest gaps among top-performing firms. Exposure isn't adoption. PwC can't tell us how deeply a company changed its workflows, whether its workers used the tools well, or what the same company would have done without AI. Sector, geography, capital, and firm quality all muddy the comparison. What we have is two observational datasets in which serious AI exposure or investment travels with faster expansion. Neither one proves AI created the extra jobs. ## The premium is moving toward fluency and judgment The more interesting signal is what employers appear to value. PwC reports greater emphasis on [judgement, creativity, and leadership](https://www.pwc.com/gx/en/1/services/ai/ai-jobs-barometer.html?ref=groktop.us). The average wage premium for AI skills reached 62%, up from the 56% premium in PwC's [2025 jobs analysis](https://www.pwc.com/gx/en/issues/artificial-intelligence/job-barometer/2025/report.pdf?ref=groktop.us). Postings that required AI skills grew 69%, while the broader posting market grew 9%. ![AI fluency matters most when paired with interpretation, leadership, and accountable judgment.](https://storage.ghost.io/c/f1/0e/f10e80f4-9285-43fc-acd4-35910a12c5f0/content/images/2026/09/01-fluency-judgment-1.png) AI fluency matters most when paired with interpretation, leadership, and accountable judgment. Those numbers point to two different needs. Companies want people who can put the technology to work. They also want people who can decide where it belongs, what deserves trust, and what happens when confidence runs out. AI fluency without judgment produces fast mistakes; judgment without fluency leaves useful capacity sitting on the table. The scarce profile holds both. Judgment isn't a quality you sprinkle over “human work.” It shows up when someone chooses the objective, asks whether the evidence fits the case, catches an exception, challenges a plausible answer, escalates low confidence, and owns the consequence. A model can help at every point. It can't accept the accountability. That is why [AI governance remains human work](https://www.groktop.us/ai-governance-human-work/). Better models move the control problem; they don't erase it. Someone still sets thresholds, interprets ambiguity, resolves competing goals, and decides when automation has gone far enough. ## A login doesn't redesign work The difference is not whether a company bought an AI tool. In a survey of more than 10,600 white-collar workers across 11 countries and regions, [Boston Consulting Group found regular generative AI use](https://www.bcg.com/publications/2025/ai-at-work-momentum-builds-but-gaps-remain?ref=groktop.us) above three-quarters among leaders and managers but at only 51% among frontline workers. Just one in three respondents said they had been properly trained. Five or more hours of training, paired with in-person coaching, was associated with more regular use and higher confidence. That still isn't a causal estimate. Better-managed companies may bundle training with useful workflows, clear expectations, and supportive leaders. The practical point is easier to see. People need more than a login. BCG also draws a line between companies that deploy tools and those that reshape work around them. Employees in its more advanced “Reshape” organizations reported more time saved, sharper decisions, and more time for strategic work. They also felt less secure about their jobs: 46% expressed concern, compared with 34% in less transformed organizations. These are self-reports, but the tension deserves more than a footnote. Work can move up the decision stack while the people doing it become less sure of their place. Workflow redesign can't be a polite name for squeezing more output from the same people. Name which decisions stay human, which evidence workers can inspect, when they can reject an AI recommendation, how errors get escalated, and where recovered time goes. As Groktopus has argued, [the org chart can become AI transformation's rate limiter](https://www.groktop.us/org-chart-rate-limiter/). The software may be ready long before decision rights, incentives, and management practice catch up. ## Don't automate away the apprenticeship The entry-level growth in the Ramp and Revelio sample matters because it cuts against a familiar fear. It doesn't close the deeper issue. A junior job isn't merely cheap production capacity. It is where someone sees routine cases repeat, makes bounded mistakes, and accumulates the context they will later call judgment. ![An AI-forward organization still needs a credible path from early work to higher-order judgment.](https://storage.ghost.io/c/f1/0e/f10e80f4-9285-43fc-acd4-35910a12c5f0/content/images/2026/09/02-apprenticeship-pathway-1.png) An AI-forward organization still needs a credible path from early work to higher-order judgment. When AI removes routine work, somebody has to rebuild the learning path. A company can't delete first drafts, basic analysis, and standard customer cases, then expect new hires to arrive carrying ten years of tacit knowledge. Give them supervised exception review, scenario practice, model-output critique, customer exposure, and authority that expands as their judgment does. Skip that work and today's efficiency becomes tomorrow's judgment shortage. Recruiting practice is beginning to reflect the shift. LinkedIn's [Future of Recruiting 2025](https://business.linkedin.com/talent-solutions/resources/future-of-recruiting?ref=groktop.us), based on platform data and a survey of more than 1,000 talent professionals, puts quality of hire and skills-based hiring at the center of its AI-era priorities. Read that as a change in recruiting emphasis, not proof that skills-first hiring improves business results or creates more jobs. Company examples help, as long as we don't turn them into experiments. LinkedIn's [2026 Top Companies list](https://news.linkedin.com/2026/LinkedIn-Top-Companies-2026?ref=groktop.us) highlights JPMorgan Chase and Microsoft as employers embedding AI in daily work while training their workforces. It also describes the spread of skills-first hiring. That tells us these practices coexist among career-growth leaders. It can't tell us what caused their headcount to change. Mercor makes the pattern easier to see at AI-native scale. LinkedIn's [2025 Top Startups list](https://www.linkedin.com/pulse/linkedin-top-startups-2025-50-us-companies-rise-linkedin-news-hox6f?ref=groktop.us) reported 160 full-time employees at the AI hiring platform. Common roles included software engineer, intelligence analyst, and machine-learning engineer; its largest functions included engineering, education, and research. That is a company employing both builders and interpreters. It is a snapshot, not proof of current openings or a template for the typical employer. ## Hire for the decision system The practical move isn't “hire more people because AI creates jobs.” The evidence can't support that slogan. Hire for the decisions your new operating model creates. ![Headcount association, skill demand, and judgment are separate evidence claims and must not be collapsed.](https://storage.ghost.io/c/f1/0e/f10e80f4-9285-43fc-acd4-35910a12c5f0/content/images/2026/09/03-evidence-bounds-claim-1.png) Headcount association, skill demand, and judgment are separate evidence claims and must not be collapsed. - **Test judgment in context.** Put an ambiguous case, an AI-generated recommendation, and incomplete evidence in front of a candidate. Give them permission to disagree. Watch what they question, what they verify, and when they escalate. - **Pair domain depth with AI fluency.** Prompt technique can't substitute for understanding the customer, regulation, product, or operational consequence. - **Keep the apprenticeship alive.** Find the automated tasks that used to train junior employees, then replace that experience with supervised practice and graduated authority. - **Name the accountable human.** Make clear who owns the outcome, who can override the system, and what evidence a consequential decision requires. - **Measure the hidden work.** Track review time, exception volume, correction cost, decision quality, and human load alongside model usage and cycle time. Ignore the demo for a moment. Look at the operating model. The early evidence doesn't show that AI creates jobs everywhere. It shows that the companies making the heaviest investments are growing while low-intensity adopters show no comparable employment gain. Other research points in the same direction: a recent [National Bureau of Economic Research working paper](https://www.nber.org/papers/w33509?ref=groktop.us) finds that AI can raise demand for labor through productivity and new tasks, while automation can reduce demand for workers whose tasks disappear. The balance depends on what the company does next. Hiring is the market-facing edge of the changes in the first three chapters. It shows which capabilities companies will pay to bring in. The fifth and harder question is what happens to everyone already inside: whether employers build a path into that higher-judgment work, or reserve it for the next person they recruit.