Executive Q&A: Gleb Tsipursky on Why AI Hiring Growth Needs a Career-Ladder Test

A year ago many leaders were still asking whether to adopt AI. What has changed?

The center of gravity has shifted from whether to use AI to how to redesign work around it. A year ago, many leadership teams treated generative AI as an experimental tool that employees could try at the edges of their jobs. Now AI is entering core workflows, budgets, performance expectations, hiring plans, and organizational design. That makes the hard questions much more human: Which tasks should change? Who owns the new workflow? How do we preserve judgment and accountability? How do we help people build confidence without letting them become dependent on the tool? The technology has improved quickly, but the management challenge has become more important because adoption now affects how people actually work.

CBRE says New York has passed the Bay Area in total tech talent, while finance is hiring heavily for AI. What does that tell you?

It tells me that AI talent is becoming an economy-wide capability rather than a technology-sector specialty. CBRE's 2026 data show New York becoming the largest overall tech-talent market, while the San Francisco Bay Area remains exceptionally strong in AI concentration. Financial services are a major employer of AI-skilled talent in New York. That shift matters because banks, insurers, professional-services firms, health systems, manufacturers, and other employers now compete for the same scarce people. It also means companies should stop assuming they can simply buy all the capability they need from the outside. The labor market will punish that strategy through cost, turnover, and slow hiring.

You have argued for a career-ladder test. What should a CHRO measure next quarter, and what result should worry them?

I would measure whether AI-driven efficiency is reducing the number of developmental assignments that teach junior employees how to exercise judgment. Track entry-level hiring, coaching hours, supervised review work, internal promotion readiness, and the share of junior tasks that were automated without a replacement learning experience. The warning sign is a productivity gain paired with a weaker talent pipeline. If work gets faster while junior employees receive fewer chances to diagnose problems, explain decisions, handle exceptions, and learn from feedback, the organization may be borrowing productivity from its future managers and experts.

Why might internal mobility beat buying specialists, and over what horizon should leaders judge the payoff?

External specialists matter, but hiring alone cannot solve an organization-wide adoption problem. Internal employees already understand the firm's customers, systems, risk tolerances, informal processes, and decision context. When you combine that domain knowledge with AI capability, you create people who can redesign real workflows rather than demonstrate generic tools. I would judge internal mobility over a six-to-twelve-month horizon, with earlier indicators such as faster staffing of AI-enabled projects, stronger retention, more internal fills, and shorter time to productive use. The deeper payoff comes when the organization can grow its own AI-capable managers and subject-matter experts instead of repeatedly bidding for scarce outside talent.

What does an AI apprenticeship look like in 2026? Have you seen an employer build one?

A good AI apprenticeship makes the tool part of supervised practice. The junior employee uses AI for a first pass, then verifies the output, explains the reasoning, identifies uncertainty, and escalates exceptions. A more experienced employee reviews both the result and the thinking process, then gives feedback. Over time, the junior employee handles harder cases with less assistance. The key is that AI should accelerate exposure to useful work without removing the feedback loop that turns experience into judgment. I have seen organizations build pieces of this model, especially structured review and coaching around AI-assisted work. I would hesitate to name one public company as a complete exemplar without verified permission, because most firms are still building the system rather than operating a mature apprenticeship end to end.

Stanford's revised Canaries data show a 19% employment gap for workers ages 22 to 25 in highly AI-exposed occupations, mainly through reduced hiring. Is a hollowing-out already underway? What happens over five years?

The risk is real, but the Stanford researchers correctly describe these as descriptive patterns rather than a causal estimate that AI alone produced the gap. Their August 2026 revision finds employment among 22-to-25-year-olds in highly exposed occupations about 19% below where it would be if it had kept pace with less-exposed peers, with the adjustment showing up primarily through reduced hiring rather than separations. If that pattern persists for several years, companies can create a pipeline problem: fewer people enter, fewer accumulate tacit knowledge, and eventually there are fewer experienced workers ready for senior roles. Five years from now, the shortage could show up as higher costs for experienced talent, thinner management benches, and greater dependence on external hiring. The practical response is to redesign entry-level work before the pipeline erodes further.

Which tasks should managers deliberately keep human because they build judgment, and how do you defend that choice to a CFO?

Keep humans deeply involved in problem framing, exception handling, ambiguous tradeoffs, stakeholder conversations, final accountability, postmortems, and any task where the employee must explain why a recommendation makes sense. AI can assist with research, drafts, comparisons, and pattern detection, but people need repeated practice making and defending consequential decisions. I would defend that investment to a CFO as capability maintenance. A company would not stop maintaining critical equipment because preventive maintenance lowers this quarter's output. Judgment is organizational capital. If automation removes every low-risk opportunity to practice it, the future cost appears later through errors, weak succession, expensive external hiring, and managers who never developed the skills their roles require.

Your book is subtitled From Resistance to Results. What does resistance look like beyond fear of job loss?

Job-loss fear is only one form. Resistance also appears as status threat, competence anxiety, distrust of leadership, concern about fairness, fear of looking foolish, overload from yet another change initiative, and identity conflict when people feel the tool devalues expertise they spent years building. Some employees avoid the tool. Others use it privately while publicly appearing skeptical. Still others comply superficially but keep the old workflow intact. Leaders often misread these behaviors as stubbornness or lack of technical skill. In practice, people usually need a credible answer to three questions: What does this mean for me, can I succeed in the new way of working, and will the organization use the change fairly?

MIT NANDA reported that roughly 95% of enterprise generative-AI pilots in its 2025 research showed no measurable P&L impact. What are companies doing wrong, and what works better?

I would treat the 95% figure as a warning signal rather than a universal audited failure rate. The MIT NANDA report was preliminary research based on more than 300 public initiatives, 52 organizational interviews, and 153 leader survey responses during the first half of 2025. The broader lesson fits what I see: companies buy a tool, run a generic pilot, and expect value to emerge without redesigning the workflow. They often lack a baseline, a business owner, a clear success metric, integration with the systems where work happens, and a feedback process for improving use. Better programs start with a specific costly or frustrating workflow, establish the before-state, involve the people who do the work, train them on both tool use and judgment, build review and error-learning loops, and scale only after the organization can show measurable improvement.

AI firms often work in person more frequently than other employers. Does AI change the hybrid-work debate?

AI changes part of the debate because the scarce resource is increasingly learning, coordination, and judgment transfer rather than simple access to software. CBRE has reported that many AI companies work in person five or six days a week. That does not prove every employer needs a five-day office mandate. A mature organization should ask which activities benefit from co-presence. Early-career coaching, pairing, difficult problem solving, relationship building, and live review of ambiguous work can benefit from deliberate overlap. Routine individual production often does not. The better hybrid design schedules people together for the work that gains from proximity instead of treating attendance itself as the outcome.

Will employers fix the talent-pipeline problem themselves, or do government and universities need to act? Is New York a useful test case?

Employers have to lead because they control job design, hiring, promotion, and the day-to-day learning environment. Universities and government can make that easier. Universities should give students repeated practice using AI while still requiring them to demonstrate independent reasoning, domain knowledge, and verification. Government can improve labor-market data, support work-based learning and apprenticeships, and reduce barriers to portable training. New York is a useful test case because it combines a huge tech workforce with finance, professional services, media, health care, and other industries that are rapidly absorbing AI talent. If employers there can build strong internal pipelines instead of relying almost entirely on experienced hires, that model can travel.

What should an early-career worker or someone moving from an adjacent field do now? What is the worst advice you hear?

Become AI-augmented and domain-grounded at the same time. Learn to use the tools well, but also learn the business process, the customer, the data, the regulatory constraints, and the reasons experienced people reject seemingly plausible answers. Keep examples of work that show your reasoning, verification, and improvement over time. Seek managers and roles that provide feedback rather than simply rewarding speed. The worst advice is either extreme: 'AI will do everything, so fundamentals no longer matter,' or 'avoid AI so you preserve your craft.' Both create fragility. The durable advantage comes from combining tool fluency with the judgment to know when the output is wrong, incomplete, risky, or irrelevant.