Toptal Tells Lets Data Science Why Data Science Demand Jumped 28% While Pay Stayed Flat

Asked what the divergence says about how experience has been repriced, Stettler was precise about the distinction.

"The divergence suggests that employers are not simply paying more for tenure; they are placing a greater value on the judgment, organizational context and ability to assume responsibility that experience often represents," he told Lets Data Science.

The mechanism, in his account, is that AI has absorbed a growing share of routine execution. That squeezes roles built on reproducible tasks while increasing the leverage of people who decide what should be done, where AI belongs in the workflow, and what the output means for the wider organization.

"Managing an AI-enabled workflow requires more than knowing how to use a particular tool," he said. "It entails decomposing problems, coordinating technology with product and business objectives, recognizing failure modes and deciding where human involvement remains essential."

He was careful not to let the argument collapse into a defense of seniority. "Years of experience are not themselves the fundamental requirement," he said. "Some younger professionals develop exceptional judgment and systems-level thinking quite quickly." Experience correlates with those capabilities on average, and signals them to employers, which is a different claim from saying it creates them.

He does not expect the two curves to diverge forever. Convergence, though, depends on something employers control rather than something juniors can fix alone: rebuilding the pathways through which people acquire judgment and responsibility in the first place. The gap "could persist longer if companies continue reducing the junior work through which those capabilities were traditionally developed."

If postings for experienced data scientists are up 28%, why has none of it reached salaries?

"The immediate explanation is that demand for professionals has improved more quickly than their bargaining power," Stettler said. Layoffs and cautious hiring across the technology sector have left employers with a deep pool of qualified applicants, so companies can add experienced roles "while generally remaining within existing compensation bands."

He also cautioned against reading the part-time figure too literally. A rise in postings does not mean employers are chasing the same specializations, seniority levels or geographies as a year ago, and part-time rates move sharply with the mix of projects in the data. "I would therefore interpret the 48% decline as evidence of substantial pressure in that market, but not necessarily as a uniform 48% reduction in the rates paid to a comparable individual."

What would actually move pay is the market becoming candidate-constrained rather than merely busier. Stettler named the signals to watch: longer vacancy durations, higher offer-rejection and counteroffer rates, a thinning supply of experienced candidates from the pool of laid-off professionals, and rising competition for people who have shipped AI and data systems into production.

"Posting growth is the first stage of a recovery," he said. "Compensation usually responds when employers begin having difficulty filling those postings at prevailing salaries."

The most pointed part of Stettler's answers concerns people entering the field now, and it cuts against the assumption that AI is straightforwardly good for beginners.

His route into the experienced tier is not what most job seekers optimize for. "The best way to cross into the experienced segment is therefore to not simply accumulate years on the job or multitudes of AI certifications. It is to produce evidence that you can take an ambiguous problem from definition to reliable result." In practice that means understanding the business objective, choosing the methods, using AI to accelerate the work, testing its output, communicating uncertainty, and knowing where human judgment still has to sit. A self-directed project on a public dataset, documented properly and published, can carry that evidence as well as a job title.

Then comes the warning. "There is also a danger in being spared too much of the foundational work," he said. AI can strip the drudgery out of data cleaning and preparation, but close contact with messy data is precisely what taught analysts to recognize outliers, reporting errors, selection effects and biases before they contaminated a model.

"Early-career professionals should use automation, but they should inspect and question the data as rigorously as if they had prepared every observation themselves," he told Lets Data Science. "Never underestimate what careful descriptive analysis and visualization can reveal before running sophisticated models."

Domain expertise compounds the effect. Someone who understands both the tooling and the field it is applied to, whether that is economics, operations, healthcare or scientific discovery, contributes judgment that neither a context-free specialist nor an unspecialized generalist can.

"Everyone increasingly has access to similar tools," Stettler said. "What will distinguish a career is the ability to apply those tools critically, contextually and responsibly."

Software engineering was hit unusually hard, Stettler noted, because it was both among the first occupations affected by the hiring slowdown and among the first to get genuinely capable AI tooling. Companies paused to work out how much of the job AI could do. Postings have been rebounding since mid-2025, and the report treats AI as a complement to engineers rather than a substitute.

"The central distinction is whether a professional owns the problem and the result, or merely performs a reproducible step between them," Stettler said. He added a caution that applies to anyone treating tool fluency as a career strategy: "Functional familiarity with AI tools is already a baseline expectation rather than a durable differentiator."

The value, in his framing, accrues to people who know which problems are worth solving and can use AI "to scale good judgment without also scaling undetected mistakes."

Toptal's model projects a departure from a multi-year downtrend in Q3 2026, with remote and hybrid hiring as the leading indicator. Stettler explained why distributed hiring moves first: it is the part of the market where companies can respond most flexibly, reaching specialized expertise across geographies and staffing exploratory work before committing to a larger headcount expansion.

He was unusually direct about what would falsify his own forecast. He would conclude it was wrong if remote and hybrid demand declined for several consecutive months, if the improvement stayed confined to contingent or experimental projects without spreading into full-time roles, or if later data revisions erased the turning point entirely. A renewed decline in software, data and professional services hiring, despite easier year-over-year comparisons, would count as evidence against it.

The external risks he named are a sudden jump in AI capability that sends employers back into reassessment, and a macroeconomic or geopolitical shock. He singled out energy through two distinct channels: a sustained price shock feeding inflation and production costs, and separately the electricity generation, grid capacity and data-center investment that AI expansion depends on, where constraints or sharp cost increases could slow deployment.

"The forecast therefore remains positive, but it is a conditional forecast," he said. "The leading indicator must spread into broader hiring rather than merely produce a temporary rebound in flexible demand."

The full dataset is in Toptal's Q2 2026 High-Skilled Job Report.