The Transition Is The Crisis: A DEEP Dive on AI, Jobs, & The Future Of Work Over the Next 5 Years
Why the economy can look fine while careers quietly collapse—and what to do in the 5-year window that matters most.
This is the paid edition of my Weekly Briefing. The free post this week focused on leadership reinvention in the AI era, and we looked into how delegation, talent flow, and adaptive capacity are being redesigned in real time. This week’s briefing takes that analysis one step further. The new signals suggest that while leaders chase productivity gains, they may be undermining the long-term human infrastructure their organizations require to survive.
If you are not a paid subscriber, you can upgrade to premium here.
Want a trusted peer group to pressure-test what you’re doing in HR, EX, and the future of work? Future of Work Leaders is a private, pitch-free CHRO community for candid discussions, Q&As, and practical insights. Members include LEGO, Novartis, PwC, Saks Global, and more. Request an invite at Future of Work Leaders or email Jacob[at]thefutureorganization[dot]com.
In September 2025, Salesforce CEO Marc Benioff went on a podcast and said something that should have made every knowledge worker sit up straight. His company had just used AI agents to cut its customer support team from 9,000 people to roughly 5,000 and the AI was achieving the same customer satisfaction scores as the humans it replaced. “I need less heads,” he said.
That same month, a Stanford study showed that software developers aged 22 to 25 had seen their employment drop nearly 20% from its peak while developers over 26 were doing fine. A Vanguard analysis found that the approximately 100 occupations most exposed to AI were actually outperforming the rest of the labor market in both job growth and wages. And a rigorous NBER study of 25,000 Danish workers found zero measurable effect of AI on earnings or hours.
All four of these things are true simultaneously. And that contradiction is the entire story.
The most common framing of the AI-and-jobs debate forces you to pick a side: either AI will cause mass unemployment, or technology always creates more jobs than it destroys. Both sides can cite evidence. Both are correct at some level of abstraction. And both are functionally useless for anyone trying to navigate the next five years. I spent a few weeks combing through over 35 sources including academic studies, government data, company case studies, and labor market surveys.
This post makes a different argument. The long-term optimists are probably right about the destination. History suggests AI will create enormous new categories of work, just as the PC, the internet, and the steam engine did before it. But the optimists are dangerously wrong about the journey.
The transition period I’m talking about is roughly 2025 to 2030, this is where real people lose real jobs, real careers get derailed, and real communities face disruption. Unlike previous technology waves, this one is arriving faster, targeting a broader range of cognitive tasks, and hitting the knowledge-worker class that was supposed to be safe. The programmer who spent four years earning a computer science degree in 2021 graduated into a job market that no longer needs what they were trained to do. No amount of historical optimism helps that individual.
The most dangerous lie in the AI jobs debate is that you have to pick a side. The optimists are right about the destination. The doomers are right about the journey. And the journey is where people’s lives actually happen.
Instead of forcing a single prediction, the evidence in this briefing separates into three buckets:
What we know (from the data now): Macro indicators can remain stable while harm concentrates in specific groups — especially entry-level knowledge workers, freelancers, and routinized cognitive roles.
What we don’t know (yet): Whether organizational restructuring at scale will follow individual AI adoption — and if so, how quickly. The “Productivity J-Curve” (Section 4) is one plausible model for this lag, but it is an assumption drawn from historical technology patterns, not an established fact about AI specifically. It is also possible that AI produces modest individual convenience without ever triggering mass restructuring. Scenario C in Section 14 describes this outcome.
What we’re watching (leading indicators): Agentic AI deployments, entry-level hiring ratios, freelance platform revenue, and labor share of GDP (detailed in Section 15).
Let’s jump into the very deep dive around AI, automation, jobs, and the future of work.
Before examining the evidence, we need a framework for interpreting it. The most influential academic model for understanding automation and employment comes from MIT economists Daron Acemoglu and Pascual Restrepo (NBER Working Paper 24196, 2018; Econometrica 2022; Journal of Economic Perspectives 2019). Their task-based model makes a critical distinction that most public discussion misses: AI automates tasks, not jobs.
As we all know, a job is a bundle of tasks. A software developer writes code, designs systems, communicates with stakeholders, reviews pull requests, mentors juniors, and navigates organizational politics. When AI automates one of those tasks (writing code), the job doesn’t disappear but it certainly changes. The balance of what’s valuable in the role shifts.
Three competing effects determine what happens next:
1. The Displacement Effect. AI replaces human labor in tasks it can now perform. This directly reduces demand for workers and puts downward pressure on wages. This is the effect that dominates headlines when companies announce layoffs “due to AI.” In the Acemoglu-Restrepo framework, the displacement effect always reduces the labor share of income in directly affected industries, meaning more economic output flows to capital (technology owners) and less to workers.
2. The Productivity Effect. The cost savings from automation increase total economic output, which raises demand for labor in the tasks that remain human-performed. If AI makes a company 30% more productive, it can expand output, enter new markets, and serve more customers, creating demand for the human tasks that AI can’t do. This is the mechanism behind the Vanguard finding that AI-exposed occupations are growing: AI makes the remaining human tasks in those occupations more valuable. This is also, as we’ll see later, exactly what happened with ATMs and bank tellers.
3. New Task Creation. Technology creates entirely new categories of work that didn’t previously exist. MIT’s David Autor (NBER WP 32140, 2024) has shown that 60% of today’s U.S. employment is in job categories that didn’t exist in 1940. Social media managers, cloud architects, UX designers, AI prompt engineers — none of these roles existed a generation ago. This effect is the strongest argument for long-term optimism, but it operates on a timescale of years to decades, while displacement can happen in months.
What this chart shows: The balance between displacement, productivity, and new task creation varies dramatically by task type. For routine cognitive tasks (data entry, basic bookkeeping, template-based writing), displacement overwhelms the other effects — the net result is clearly negative for employment. For creative and strategic tasks (system design, leadership, R&D), productivity gains and new task creation dominate — the net result is job growth. The critical battleground is non-routine cognitive work (analysis, coding, professional writing) where all three effects are strong and nearly balanced. This is why we see contradictory signals: jobs in these categories are simultaneously being created AND destroyed, depending on experience level, specific sub-tasks, and industry context.
