Where the Value Is Actually Moving in the AI Economy

Career Landscape Reports: 5 of 5 in a series

“Stay adaptable” has become the default advice for navigating AI’s effect on careers, and it’s not wrong so much as empty. It doesn’t tell you what to actually do on a Tuesday.

The first four pieces in this series built toward something more specific. AI genuinely expands the total work available in a field, the evidence for which held up across the software industry data in the first piece. The traditional path into that expanding work is narrowing for people early in their careers. A role’s actual exposure depends on factors you can check for yourself. And the market delivering all this information keeps contradicting itself within the same year. This final piece answers the question those four were building toward: given all of that, where does the growing work actually sit, what does it look like day to day, and what does building toward it on purpose actually involve.

The Growth Is Real and It’s Concentrated in a Specific Place

The World Economic Forum’s Future of Jobs Report 2025 projects roughly 170 million new roles created globally by 2030 against about 92 million displaced, a net gain of 78 million, alongside an estimate that 39% of existing skill sets will be transformed or become outdated. It’s worth being precise about what those figures actually measure: they’re macrotrend-driven, combining technological change with the green transition, demographic shifts, economic uncertainty, and geoeconomic fragmentation, not AI’s effect in isolation. The report’s own narrower breakout for AI and information processing technology specifically is far smaller, about 11 million jobs created and 9 million displaced by 2030. These are employer expectations, not guarantees, and they don’t by themselves tell you how much of the larger number AI accounts for.

The more useful picture comes from looking at where AI-specific hiring is showing up right now, not projected out to 2030. The share of US job postings requiring AI skills reached 2.5% in 2025, according to Lightcast data cited in Stanford’s AI Index 2026, up 55% in a single year. PwC’s AI Jobs Barometer measures a related but distinct thing, postings specifically for AI specialists, and finds those grew 68.9% in 2025 against 8.6% for the job market overall. The wage premium for having AI skills within an existing occupation, not switching jobs but adding the skill to a role you already hold, reached 62% according to PwC, up from 56% the year before. Demand is running well ahead of the supply of people who can do this work: Bain projects the US will have more than 1.3 million AI-related jobs by 2027 against roughly 645,000 qualified people to fill them.

None of that proves AI itself is expanding total employment in the fields it touches. A wage premium and rising posting counts are also consistent with simple skill scarcity, or with existing work being reclassified under a new label, or with compensation concentrating toward fewer specialists rather than the field as a whole growing. The more defensible claim, and the one this piece rests on, is narrower: AI is creating and expanding some categories of work even as it compresses or eliminates others, and the following section shows what that looks like in the hiring data on specific roles, which is more informative than any aggregate growth number.

What the Fastest-Growing Roles Actually Do — Where the Data Is Cleanest

Putting LinkedIn’s Jobs on the Rise rankings alongside posting-volume data from Lightcast and Indeed, a consistent shape emerges, and it isn’t the shape most people picture when they imagine “AI jobs.” Worth being upfront about why the clearest evidence for this pattern comes from technology hiring data specifically: new AI-adjacent roles in tech get distinct job titles that posting trackers can count cleanly. The same shift in law, medicine, or finance shows up as an existing role’s day-to-day work changing, not a new title appearing, which makes it real but much harder to see in aggregate hiring statistics. The tech data is the clearest illustration of the pattern. It is not the only place the pattern is happening, and the next section looks at where else it shows up.

AI Engineer tops the list by volume, with roughly 1,550 new US postings a week. But the fastest-growing roles by percentage tell a sharper story, with the caveat that fast percentage growth from a small starting base can look more dramatic than its actual scale. Forward Deployed Engineer postings are up more than 1,000% year over year, still a comparatively small category in absolute terms even as it grows quickly. This is not someone building models in a research lab. It’s someone who goes to a specific customer’s business, looks at their actual messy processes, and wires AI capability into that specific context, a job description that’s closer to a skilled consultant than a machine learning researcher. Agentic AI Engineer postings grew 280% in a year, also still a small share of overall postings but growing faster than almost anything else tracked. AI Consultant and AI Strategist ranks second on LinkedIn’s list two years running, and the people filling these roles have a median of eight years of experience, arriving most often from product management rather than from a technical pipeline.

The pattern underneath all of these roles is the same one this series has been building toward since the first piece: the growth isn’t concentrated in people who build the underlying technology. It’s concentrated in people who sit in the gap between what AI can technically do and what a specific business, team, or process actually needs, and who have enough contextual judgment, domain knowledge, and willingness to take accountability for the result to close that gap. More than half of all AI-related job postings in the US, and 59% in Germany, sit entirely outside IT departments, in marketing, sales, customer support, and administrative functions where someone understands both the tool’s real limits and the actual work well enough to make the two fit together.

Worth naming what’s happening to the role most people associate with “AI job” a few years ago. Prompt engineer dropped off the list of roles companies planned to hire for in Microsoft’s Work Trend Index after appearing on it the year before. The underlying skill didn’t disappear, it shows up inside a meaningful share of AI Engineer job postings, but the standalone job title dissolved into broader roles almost as quickly as it appeared. That’s a useful caution against building a career narrowly around any single tool-specific title, since the tools and the titles built around them are moving faster than most people’s career planning horizons.

The Same Shift is Seen In Non-Tech Hiring Data

The reason this matters for readers outside tech is that the underlying pattern, not the specific job titles, is what’s actually durable. Look at four fields with none of the new-title advantage that makes tech easy to track, and the same shape shows up in how existing roles are described.

In law, Thomson Reuters’ 2026 Future of Professionals report found the share of law firm legal teams actively using AI rose from 28% in 2025 to 41% in 2026. What’s changing inside that adoption is specific: the work AI has absorbed is what one industry analysis called “the production layer,” finding cases, drafting first passes, summarizing documents, checking citations, while what it hasn’t touched is “the judgment layer,” deciding which facts matter, which arguments will persuade a specific judge, what risk a client should actually accept. First-year associates, who used to spend a large share of their time producing that first draft themselves, are increasingly spending it verifying and quality-checking an AI-produced draft instead. Axios reported in May 2026 that this shift is measurably straining the traditional path that trains the next generation of senior lawyers, the same entry-level narrowing this series documented in its second piece, showing up again in a completely different profession.

In accounting and finance, the U.S. Bureau of Labor Statistics projects accountants and auditors will grow 5% through 2034 while bookkeeping clerks, whose work is more purely rules-based and repetitive, are projected to decline 6% over the same period, an updated version of the same split this series traced in software development in its first piece. Reporting on the audit profession specifically found that tasks once handled by junior accountants, testing payroll, expenses, and contracts, are increasingly automated, and the role is shifting toward oversight: interpreting what the AI-driven testing actually found and exercising the professional judgment auditing standards require, which under current regulatory rules an AI system cannot substitute for regardless of how capable it is.

In healthcare, radiology remains the clearest illustration, and a decade of watching it play out has only sharpened the pattern this series described in its opening piece. AI-assisted image analysis is now genuinely widespread, and radiologist headcount and pay have continued climbing anyway. That’s consistent with, though not proof of, what this series has argued the adoption actually changed: the radiologist’s job increasingly involves knowing when to trust an AI reading and when something about a specific patient’s history or an unusual presentation means the algorithm’s confident answer is wrong. Growing demand for imaging, clinician shortages, and reimbursement patterns are all plausible contributors to the same headcount and pay trend, and the data here doesn’t rule those out. What it does show clearly is the shape of the job itself changing. Researchers studying this collaboration have identified a real risk worth naming honestly, sometimes called automation complacency, where working alongside a highly reliable AI system for long enough gradually erodes a professional’s vigilance for the rare cases where it’s mistaken. That’s not an argument against the shift. It’s evidence for exactly how much the job now depends on a specific, disciplined kind of judgment that has no equivalent in the pre-AI version of the role.

More broadly, healthcare’s own research on AI adoption doesn’t frame this as human judgment being permanently exempt from automation, and neither should this piece. The American Medical Association’s 2026 Physician Survey on Augmented Intelligence found 81% of physicians now use AI in their practices, more than double the 38% who did in 2023, and deliberately uses the term “augmented intelligence” to signal that the technology is meant to support clinical judgment rather than replace it. The same survey found 88% of physicians worry about skill loss from relying on these tools, concentrated especially among physicians with ten years or less in practice, an echo of the entry-level narrowing this series has traced across several unrelated fields. The honest version of this pattern isn’t that judgment is safe from automation forever. It’s that judgment, verification, and accountability are where the work is currently concentrating, which is a claim about where value sits today, not a permanent exemption.

In marketing, the American Marketing Association’s 2026 State of Marketing Careers Report, based on a survey of 1,412 practitioners conducted between December 2025 and January 2026, put the shift in a single sentence: companies are hiring less execution, not less judgment. The share of marketing job postings mentioning AI nearly doubled over the course of 2025, and total marketing job postings fell more than 8% the same year, with overall marketing employment sitting roughly 27% below its March 2020 level. Read alone, that looks like straightforward contraction, and contraction, cost-cutting, and AI adoption could plausibly be happening at once without any deliberate redesign behind it. What the data more specifically supports is that AI is absorbing the routine production and reporting work that used to fill out a marketing team’s headcount, while the AMA’s own research found that marketing strategy, brand management, and critical thinking are rated among the skills AI is least able to replicate, the ones still driving hiring decisions even as overall postings shrink. That’s consistent with a marketing team built around 2026’s version of the job being a differently shaped one rather than simply a smaller one, weighted more toward the people deciding what the campaign should say and less toward the people who used to produce the first draft of it.

Four fields, four different vocabularies, the same underlying shape: the work being absorbed is the codified, repeatable part, and the work expanding is deciding when the codified part is wrong for this specific case. None of these professionals needs a new job title to be standing in the part of their field that’s growing. The shift is happening inside the role they already hold.

Why So Much of This Work Fails Anyway

The clearest evidence for where the actual value sits isn’t the growth numbers. It’s the failure pattern showing up across every field this piece has covered. A hospital deploys diagnostic AI without building in the review discipline described above, and clinicians either distrust it or defer to it too readily. A law firm rolls out AI drafting without redesigning how junior associates verify it, and the verification work either doesn’t happen or falls on people never trained to do it. A finance team automates reconciliation without deciding who owns the judgment calls that survive automation, and errors slip through in exactly the gap nobody assigned. A marketing team measures how much AI it’s adopted instead of how well the judgment layer around it is working, and the AMA’s own research found that’s precisely the wrong thing to be counting. In every case, the failure isn’t that the AI didn’t work. It’s that the surrounding judgment work, the part this entire piece has argued is where the growth actually is, didn’t get designed around it.

Technology happens to be the one sector where this pattern has been measured with real precision, which is worth citing directly. Gartner predicts that more than 40% of agentic AI projects currently underway will be canceled by the end of 2027, citing escalating costs, unclear business value, and inadequate governance as the leading causes. Gartner’s own analyst on the research, Anushree Verma, described the underlying problem in terms that apply well beyond software: most current agentic AI projects are early-stage experiments driven by hype, which blinds organizations to the real cost and complexity of getting AI to work reliably at scale. If the technical capability alone were sufficient, failure at that rate wouldn’t happen, since the underlying models keep getting more capable every year. Connecting a capable model correctly to a specific organization’s data, processes, and actual decision points, in a way that produces something reliable enough to trust, is the hard part in every field this piece has covered. Technology is simply the field with a research firm counting the casualties precisely enough to put a number on it.

This connects directly to the entry-level narrowing this series covered in its second piece. Stanford’s Digital Economy Lab, updating its research on payroll data in August 2026, found that employment among workers aged 22 to 25 in highly AI-exposed occupations now sits about 19% below where it would be had it kept pace with employment among similarly aged workers in less-exposed occupations, up from a 15% gap in the prior year’s data, with experienced workers showing no comparable divergence. The researchers themselves are careful to frame this as no evidence of widespread, economy-wide displacement, driven primarily by reduced hiring of young workers rather than layoffs. It’s the same tilt toward seniority and judgment this entire series has documented from multiple independent angles and multiple unrelated fields, showing up again in the freshest available data.

Building Toward It on Purpose

None of this points toward a single job title to chase, since the titles themselves are visibly unstable, and it doesn’t point toward switching fields either. It points toward a kind of positioning within whatever field you’re already in, one that holds up regardless of which specific job title or tool is fashionable this year.

Resist building your professional identity around any single tool or narrow technical skill, the way prompt engineering briefly looked like a durable specialty and turned out not to be. The skill underneath a fading title usually survives and gets absorbed into a broader role. Build toward the broader role directly rather than betting on the narrow one, and recognize that in most fields outside tech, that broader role doesn’t need a new name to exist. It’s just more of the judgment and oversight portion of the job you already have, and less of the execution portion.

Look specifically for work that sits in the gap between AI’s technical capability and a real, specific process in your own field, rather than work that’s purely about using a tool in the abstract. Whether that’s a lawyer who understands exactly which AI-drafted clauses need a second look, an accountant who knows which automated reconciliations are trustworthy and which aren’t, a clinician who’s built real discipline around when to override a confident-sounding AI read, a marketer who can tell a technically on-brief AI draft from one that will actually land with a specific audience, or the tech-sector consultant roles described above, the pattern is the same: domain expertise plus a clear-eyed, current understanding of where the tool actually fails. That’s not a certification. It’s available to build from wherever you’re already standing professionally.

Treat the high failure rate of AI projects, in your field or any other, as the opportunity it actually is rather than as evidence against this direction. Given that a large share of these projects fail specifically from unclear business value and inadequate governance, the professional who can look at a proposed AI deployment and tell the difference between the version that will actually work and the version that will get quietly canceled or quietly cause harm is doing exactly the judgment work this entire series has pointed toward, in an area where organizations are demonstrating both growing need and a shortage of people who combine domain expertise with practical AI judgment.

None of this has to start as a career move. It can start as an hour spent on one piece of your current work. Pick one recurring process you’re already responsible for. Separate what you do into two piles: the production tasks, the parts with a defined right answer that mostly involve executing, and the judgment tasks, the parts where you’re deciding, verifying, taking accountability, or handling something that doesn’t fit the standard case. Identify where AI could plausibly help with the first pile, and where getting it wrong would actually matter. Then run one small, reversible experiment, using the two-way-door thinking from the fourth piece in this series, and pay attention to what improved, what failed, and what still needed you specifically. That last part is the evidence worth keeping: a revised procedure, a case study, a concrete example you can point to the next time someone asks what you actually bring to the work.

A sharper version of the same exercise: what part of your field gets harder, not easier, once AI is introduced into it? The person who works out how to solve that specific problem is usually standing exactly where this piece has been pointing.

Closing the Series

Put all five pieces together and the picture holds together as one argument rather than five separate ones. AI genuinely expands the total work available in a field, which the software industry data in the first piece bore out directly, and not wishful thinking.

The traditional path into that expanding work, learning judgment through routine execution as a junior person, is narrowing for a specific and identifiable group. That’s the uncomfortable complication to the first point, not a contradiction of it.

Whether any individual role benefits from the expansion depends on checkable factors: physical presence, cost of error, labor market tightness, and the depth of a specific person’s accumulated judgment. Not a single risk score.

The market delivering signals about all of this reverses itself often enough that the right response is favoring decisions you can walk back over decisions you can’t, rather than waiting for noise that may never resolve into clarity.

And the growth, whether it’s visible in a new job title or buried inside an old one, sits specifically in the judgment-and-integration work this entire series has been describing from the start. Not in narrow technical execution, not in building the underlying models themselves, and not in any one field more than another.

None of that adds up to a guarantee, and it isn’t meant to. It adds up to something more useful than “stay adaptable”: a specific description of where the work is moving, and specific, checkable reasons to believe it, that you can act on starting this week rather than waiting for the picture to fully settle.

---

Sources

This piece drew on the following research and reporting (several entries are secondary reporting on the underlying studies, not the original reports themselves):

• World Economic Forum, “Future of Jobs Report 2025,” January 2025. Source for the 170 million/92 million/78 million macrotrend-driven job projections and the narrower AI-specific breakout of roughly 11 million created and 9 million displaced.
• Maksim Ilin, “The AI Job Market in 2026: Who Gets Hired, What They Earn, and Which Roles Are Fading,” September 6, 2026. Synthesis of LinkedIn Jobs on the Rise, WEF, Stanford AI Index, PwC AI Jobs Barometer, Lightcast, Indeed, and Bain data on specific role growth, wage premiums, and the Forward Deployed Engineer and Agentic AI Engineer trends.
• PwC, “2026 AI Jobs Barometer,” 2026. Source for AI specialist posting growth and the AI skills wage premium.
• Thomson Reuters, “2026 Future of Professionals Report,” 2026. Source for law firm AI adoption figures.
• Claude for Lawyers, “Will AI Replace Lawyers? What the Data Actually Says (2026),” 2026. Reporting on the shift in junior associate work toward AI-output verification, and the Axios May 2026 reporting on entry-level pipeline strain in BigLaw.
• Controllers Council, “Why Audit Firms Are Rewriting the Role of the Modern Accountant,” May 8, 2026.
• AIIFI, “Will AI Replace Accountants? The 2026 Reality, Role by Role,” 2026. Source for the BLS growth projections for accountants and auditors versus bookkeeping clerks.
• Knowable Magazine, “AI Won’t Replace Radiologists, But Will Dramatically Change Their Jobs,” by Lola Butcher, July 22, 2026. Source for the automation complacency research in radiology.
• American Marketing Association, “2026 State of Marketing Careers Report,” 2026. Survey of 1,412 marketing practitioners on AI’s effect on marketing job postings, skills, and team structure.
• Gartner, “Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027,” press release, June 25, 2025, featuring commentary from Senior Director Analyst Anushree Verma.
• Stanford Digital Economy Lab, “Canaries in the Coal Mine? Six Facts About the Recent Employment Effects of Artificial Intelligence,” by Erik Brynjolfsson, Bharat Chandar, and Ruyu Chen, revised August 12, 2026. Source for the 22-to-25 employment gap in AI-exposed occupations.
• American Medical Association, “AI Usage Among Doctors Doubles as Confidence in Technology Grows,” 2026 Physician Survey on Augmented Intelligence, March 12, 2026. Survey of nearly 1,700 physicians on AI adoption and skill-loss concerns.

I use AI tools to help research sources and draft these posts. I review, edit, and take responsibility for everything published here, including any errors.

This is Report 5 of 5 in the Career Landscape series.