The Intelligence Transition: AI, Work, and the Future We're Building

The Question Every CEO Must Answer

Every board meeting in 2026 contains the same unspoken question: How fast are we deploying AI, and what happens to our people?

I sit with this question every week. Not as an abstraction — as a CEO who signs off on technology investments and workforce decisions simultaneously. The tension is real and it is mine: AI delivers measurable productivity gains, the competitive pressure to adopt is intense, and the people whose work is being reshaped are colleagues, not line items on a spreadsheet.

The board sees the numbers. McKinsey estimates that 57% of U.S. work hours could theoretically be automated with existing technology — nearly double the 30% estimated in 2023. Goldman Sachs projects a roughly 10-year timeline for firms to adopt AI at scale. The EY AI Pulse Survey reports that 96% of organizations investing in AI see productivity gains. The case for deployment is overwhelming.

But here is what the efficiency slide in the board deck does not show: the 6.1 million U.S. workers that the Brookings Institution identifies as facing both high AI exposure and low adaptive capacity — 86% of them women, concentrated in clerical and administrative roles. It does not show the Goldman Sachs finding that employment among 22-to-25-year-olds in AI-exposed occupations fell 16% between late 2022 and mid-2025. It does not show the 55% of employers who made AI-driven cuts and now regret them.

We are building extraordinary intelligence into our machines. The question is whether we are bringing equivalent intelligence to the decisions we make about the people those machines affect.

This article is my attempt to look at the evidence honestly — all of it — and to reason forward from what we actually know, not from what we hope or fear.

The Evidence: What the Numbers Actually Say

The debate about AI and work has fractured into three camps, each with genuinely strong evidence. The honest answer is that all three are simultaneously correct — and that the transition period, not the endpoint, is where the greatest risk concentrates.

The displacement case is real but concentrated. The St. Louis Fed found a 0.47 correlation between occupational AI exposure and rising unemployment rates from 2022 to 2025. Major companies have explicitly cited AI in workforce reductions: Amazon (14,000 corporate roles), Microsoft (~15,000), Salesforce (4,000 customer support positions), IBM (8,000 HR roles). The IMF estimates roughly 40% of global employment faces AI exposure, with about half of exposed jobs in advanced economies at risk of displacement rather than augmentation. The MIT Iceberg Index found AI can already technically perform tasks corresponding to 11.7% of the U.S. labor market — approximately $1.2 trillion in wages.

Yet aggregate labor data stubbornly refuses to show an apocalypse. The Yale Budget Lab’s comprehensive analysis found “no discernible disruption” 33 months after ChatGPT’s release.

The augmentation case has the strongest empirical support. Erik Brynjolfsson’s landmark study found AI assistance increased customer-support productivity by 15%, with the largest gains among less experienced workers — up to 34% improvement for novices. Across studies, productivity gains range from 10% to 65% for coding, consulting, customer service, and professional writing. PwC’s analysis of approximately one billion job postings found that productivity growth nearly quadrupled in AI-exposed industries since 2022, from 7% to 27%. The AI skills wage premium reached 56% in 2025.

The EY survey is perhaps the most telling: 96% of organizations report productivity gains from AI, yet only 17% reduced headcount. Most reinvested gains into new AI capabilities (47%), cybersecurity (41%), R&D (39%), or reskilling (38%). A CEPR study of 12,000+ European firms found AI adoption increases labor productivity by 4% on average with no evidence of reduced employment in the short run.

The vulnerability of the augmentation argument is temporal: augmentation may be a transitional phase before full automation. As Daron Acemoglu, the 2024 Nobel laureate in economics, argues, experimental productivity gains from “easy-to-learn tasks” cannot be extrapolated to the whole economy — he projects a far more modest 0.53–0.66% total factor productivity gain over a decade.

The reinvention case carries history’s endorsement. David Autor’s landmark study found that 60% of today’s workers hold occupations that did not exist in 1940, and over 85% of employment growth since 1940 was driven by technology creating new job categories. AI-specific roles are already growing at extraordinary rates: AI Engineer postings grew 143% year-over-year, Prompt Engineer 136%, AI Solutions Architect 109%. The WEF projects 170 million new roles created by 2030, yielding a net gain of 78 million jobs after accounting for 92 million displaced.

But the reinvention argument has a critical limitation: 77% of new AI jobs require master’s degrees. Displaced retail clerks and administrative assistants cannot easily become AI engineers. And Autor’s own data reveals that since 1980, new job creation has polarized toward either high-skill/high-pay or low-skill/low-pay roles, hollowing out the middle class.

These three dynamics are not mutually exclusive — they operate simultaneously. A single job can be augmented in some tasks, see others displaced, and have new responsibilities created, all at once. The question is not which argument is right. The question is: during the transition period, who bears the cost?

Measuring the Machine: The Capability Indexes

A new generation of measurement tools has emerged to track AI’s economic capabilities. Their collective finding is striking: the gap between what AI can theoretically do and what it actually does in practice remains enormous — but it is closing fast.

The Office Work Capability Index (OWCI), published by Bionic Advertising Systems, translates fragmented benchmark data into a single 0–100 scale where 100 equals human expert-level office work capability. Built on a weighted composite of knowledge work, workflow completion, reasoning, and implementation benchmarks, the OWCI places current frontier models in the 70s-to-80s range — the boundary between “junior analyst” and “competent professional.” The index projects frontier systems may reach human-expert parity (OWCI 100) in late 2026, with the broader frontier cluster reaching that level around 2027.

OpenAI’s GDPval benchmark evaluates AI performance on 1,320 real-world tasks across 44 occupations contributing roughly $3 trillion annually in U.S. wages. Tasks are constructed from actual work products created by experts averaging 14+ years of experience. GPT-5.2 achieved a 70.9% win-or-tie rate against human industry experts while completing tasks 11 times faster at less than 1% of cost. Performance more than tripled from GPT-4o (spring 2024) to GPT-5 (summer 2025). Critics rightly note that GDPval tasks are “precisely specified” one-shot exercises — nothing like the ambiguous, multi-stakeholder workflows that characterize real employment.

The OECD AI Capability Indicators compare AI to human abilities across nine domains on a 1–5 scale. Current AI rates at Level 3 in language, knowledge, and creativity; Level 2 in problem solving, metacognition, and social interaction; and lower levels in physical manipulation and robotic intelligence. The framework reveals a critical wedge: AI is rapidly mastering “brain work” while still struggling with “heart and hand” work.

Anthropic’s Observed Exposure metric maps approximately one million real conversations to occupational tasks and reveals a massive gap between theoretical and actual use. Computer and math occupations show 94.3% theoretical capability but only 35.8% observed exposure — a realization rate of just 38%. Legal occupations: 89% theoretical, 20.4% observed. The top individual occupations by observed exposure are computer programmers (74.5%), customer service representatives (70.1%), and data entry keyers (67.1%). Meanwhile, 30% of workers have zero AI coverage. A crucial finding: workers in the most exposed professions are older, more educated, and earn 47% more than those in low-exposure roles — the demographic opposite of previous automation waves.

The ARC Prize benchmark measures fluid intelligence — the ability to solve novel problems not present in training data. Humans score 100%. Until recently, AI models scored below 4%. By early 2026, Gemini 3 Deep Think reached 84.6% and GPT-5.4 Pro reached 83.3% — a dramatic leap driven by reasoning systems and test-time compute.

Scale AI’s Remote Labor Index provides the most grounding reality check. Testing AI agents on 240 real freelance projects averaging 28.9 hours of human work, the best-performing agent achieved a mere 2.5% automation rate, earning just $1,720 out of $143,991 possible. Common failure modes included incompleteness (36%), sub-professional quality (46%), and file errors (18%).

And Stanford HAI documents the acceleration: SWE-Bench scores leapt from 4.4% to 71.7% in a single year. The cost of GPT-3.5-level inference collapsed from $20 per million tokens to $0.07 — a 280-fold reduction. Global private AI investment hit $252.3 billion in 2024.

The picture is clear: AI is impressively capable on isolated tasks, approaching expert-level on structured work, but still far from replacing the messy, multi-stakeholder reality of most jobs. The gap is closing — the question is how fast, and whether through augmentation or automation.

Twelve Sectors, Twelve Realities

Finance and banking lead the augmentation wave. JPMorgan dedicated $18 billion in technology spending for 2025, with AI projected to deliver $1.5–2.0 billion in annual value. Over 200,000 JPMorgan employees have access to generative AI tools, and Morgan Stanley’s AI Debrief achieved 98% adoption among financial advisor teams. Citigroup reports that 54% of financial jobs have high automation potential. Bloomberg Intelligence projects 200,000 Wall Street jobs cut over three to five years — yet banks’ pretax profits are projected to rise 12–17% by 2027 from AI.

Healthcare tells AI’s most optimistic story. Over 1,247 AI-enabled medical devices have received FDA approval, with 75% in radiology. Medical transcription is already 99% automated. But direct patient care faces no projected negative impact through 2033. The global healthcare worker shortage of 11 million by 2030 means AI primarily fills gaps rather than displaces workers. AI doubles malignant nodule detection when paired with human readers. Nurse practitioner roles are projected to grow 52%.

Legal services show rapid adoption — 47.8% of attorneys at large firms used AI in 2024, and Clifford Chance reports over 90% daily adoption by end of 2025. The Am Law 100 saw 13.3% revenue growth with attorney headcount actually growing 7.7%. A lawyer saving 240 hours annually with AI takes on 15–20% more client work, expanding capacity rather than contracting headcount.

Education is firmly augmentation. The market grew 38.4% to $7.57 billion in 2025. A Harvard study found students using an AI tutor showed approximately double the learning gains of traditional classrooms. Teachers using AI save approximately 5.9 hours per week — the equivalent of six extra weeks per school year.

Manufacturing faces physical-digital convergence. Assembly line employment is projected to decline from 2.1 million (2024) to 1.0 million by 2030. Predictive maintenance AI lowers costs by 25% and reduces unexpected downtime by 30%. Yet 500,000 manufacturing jobs remain unfilled due to skills gaps — AI is solving a labor shortage as much as creating one.

Retail sees gradual automation: cashier employment declining 11% through 2033, AI chatbots handling 70–80% of standard inquiries. But physical stores are expected to still account for 72% of retail revenues by decade’s end, and self-checkout has faced pushback. Routine tasks automate; customer-facing roles persist.

Transportation is approaching an inflection. Waymo completed 14 million+ trips in 2025 across 10 cities with annualized revenue of $350 million. But fully automated driving without safety drivers remains years away. The potential displacement is enormous — roughly 10 million U.S. workers — but the timeline is measured in decades.

Creative industries face the deepest identity crisis. Los Angeles County lost 41,000 film and TV jobs in three years. Writer gig availability fell 42%. Goldman Sachs estimates AI can automate 26% of creative work tasks. Yet creative directors are hired at higher rates even as individual creators are cut — the industry restructures around AI-augmented creative leadership.

Agriculture is primarily addressing chronic labor shortages. Autonomous equipment and precision farming increase yields by up to 25% while reducing water use by 25–30% and pesticide application by 30%.

Government adoption slightly exceeds the private sector, with 43% of employees using AI at least a few times annually and over 1,100 active federal AI use cases — a ninefold increase in generative AI use in one year. Focus areas: document processing, fraud detection (reducing case backlogs by over 40%), and citizen services.

IT faces AI’s deepest irony: U.S. programmer employment fell 27.5% between 2023 and 2025. GitHub Copilot writes 46% of average developer code. Entry-level tech job postings dropped 60% between 2022 and 2024. Yet the sector simultaneously creates the roles that build and maintain AI systems — perfect illustration of displacement, augmentation, and reinvention operating simultaneously.

Outsourcing (India/Philippines) confronts its reckoning. India accounts for 52% of the global outsourcing market with 4.5 million IT workers. Indian BPM net headcount has dropped to fewer than 17,000 new hires annually, down from 130,000 in 2022–2023. The Philippines’ BPO industry paradoxically added 135,000 jobs in 2024, suggesting augmentation may precede displacement even in highly exposed sectors.

The Development Divide

AI’s impact cleaves sharply along development lines. In advanced economies, roughly 60% of jobs face meaningful AI exposure, concentrated among high-skill cognitive workers who paradoxically have the greatest capacity to adapt. In emerging markets, exposure drops to approximately 40%, and in low-income countries to just 26%.

Lower exposure provides cold comfort when it means being bypassed by AI’s benefits rather than shielded from its disruptions.

India’s outsourcing sector faces its reckoning. India’s IT sector contributes 7.5% of GDP and employs 4.5 million people. Jefferies predicted a 50% revenue hit for call centers and 35% for other back-office functions over five years. The Philippines faces parallel risks: BPO contributes 8.5% of GDP and employs 1.7 million, with contact centers accounting for 83% of revenue — highly vulnerable to conversational AI. The concept of “premature deprofessionalization” haunts both nations: AI may cap the share of high-quality white-collar jobs before developing nations reach higher income status. In India alone, an estimated 400,000 junior developer jobs face disruption risk by 2030.

Infrastructure determines everything. High-income countries (17% of global population) account for 87% of notable AI models, 86% of AI startups, and 91% of venture capital funding. They control 77% of global data center capacity; low-income countries hold less than 0.1%. Internet usage stands at 93% in high-income nations versus 27% in low-income countries. Only 5% of populations in the poorest countries possess basic digital skills.

Yet bright spots exist. Over 40% of ChatGPT’s global traffic now originates in middle-income countries. Africa’s AI market is projected to grow from $4.5 billion to $16.5 billion by 2030. Rwanda’s Babyl telemedicine platform has registered 62% of the adult population and delivered over 5 million virtual consultations. In Ghana, the Rori AI math tutor delivered learning equivalent to an extra year of schooling at $5 per student via WhatsApp. In Ethiopia, 380,000 micro-enterprises accessed $150 million in AI-powered uncollateralized credit.

The UNDP’s December 2025 report warns that without strong policy action, AI could reverse decades of narrowing development inequalities. The IMF projects that AI-driven growth in advanced economies could be more than double that in low-income countries. The “premature deindustrialization” risk is real: as AI-powered automation makes reshoring profitable, developing countries lose the cheap-labor advantage that was their primary path to industrialization.

But “Small AI” — mobile-first, low-infrastructure applications designed for everyday devices — represents perhaps the most promising pathway. The leapfrog pattern is not hypothetical; it is already happening in healthcare, education, and financial services across Africa and Southeast Asia.

A CEO’s Confession

I want to be honest about something that most business leaders avoid saying publicly: the tension between delivering returns and managing this transition responsibly is not theoretical for me. It is the hardest part of my job.

When we evaluate an AI deployment at Ibizai, we look at the 5-year NPV. Not just the financial projection — what optionality does it create, what does it foreclose, and what is the compounding effect if we execute well? But we also ask: who benefits and who doesn’t?

That second question is not altruism. It is risk management. The 55% of employers who regret their AI-driven cuts learned this the hard way. They optimized for the short term — reduced headcount, captured efficiency gains — and then discovered they had lost institutional knowledge, damaged morale, and undermined the very adaptability they needed as the technology continued to evolve.

My conviction, refined through building a fintech in São Paulo that served 12,000 micro-enterprises and through the evidence assembled in this article, is that augmentation-first is not just the ethical choice — it is the strategically superior one. The EY data supports this: the organizations that reinvested AI gains rather than cutting headcount are the ones building compounding advantages.

But I will not pretend the choice is simple. We operate in competitive markets. The company that underinvests in AI efficiency loses market share. The company that deploys AI without regard for its workforce loses something harder to measure but equally essential: the trust, creativity, and institutional knowledge that no model can replicate.

The discipline is holding both simultaneously — and accepting that the right answer is not always the comfortable one.

The Entry-Level Crisis

If there is one finding in this research that keeps me awake, it is the entry-level pipeline crisis.

Goldman Sachs found that employment among 22-to-25-year-olds in AI-exposed occupations fell 16% from late 2022 to mid-2025. Among young software developers specifically, the decline was nearly 20%. The Burning Glass Institute documented that entry-level software development postings requiring three years or less of experience dropped from 43% to 28% of listings between 2018 and 2024. In the UK, tech graduate roles were cut 46% between 2023 and 2024, with a projected additional 53% drop by the end of 2026. Computer science graduates now face 6.1% unemployment.

This is not a labor market correction. It is a structural change in how careers begin.

When AI absorbs the tasks that traditionally trained junior professionals — the research, the drafting, the analysis, the “grunt work” that built skills, judgment, and professional networks — we are not just eliminating positions. We are eliminating the apprenticeship pathway through which the next generation of senior professionals was forged.

And the new roles that AI creates are not accessible to the people being displaced. 77% of new AI jobs require master’s degrees. The displaced retail clerk, the eliminated administrative assistant, the junior developer whose role was absorbed by Copilot — these workers cannot simply transition to AI engineering.

The data shows a simultaneous widening of two gaps: the organizations that adopt AI grow more productive, while the pipeline of workers who can sustain and advance those organizations in the future grows thinner. This is a 5-year problem masquerading as a hiring efficiency. We will feel it when the current cohort of mid-career professionals reaches leadership roles and discovers there is no bench behind them.

Can Societies React in Time?

The most sobering finding across all the research is the speed mismatch between AI advancement and policy response. Legislative and institutional processes move linearly while AI capabilities advance exponentially.

What exists is promising but insufficient. Singapore’s SkillsFuture Initiative has reached over 600,000 individuals, offering S$500 lifetime credits for all citizens over 25 plus S$4,000 top-ups for workers over 40, with training allowances up to S$3,000 monthly for 24 months during career transitions. The U.S. has activated over 90 federal AI policy actions. The EU AI Act provides the world’s first binding legal framework. France has committed €10 billion to AI.

UBI experiments show mostly positive results. The Stockton, California SEED program saw recipients actually increase full-time employment. Kenya’s GiveDirectly trials demonstrated increased entrepreneurial activity. Finland reported wellbeing improvements. But scaling remains the challenge.

The Mauritius model offers a compelling alternative: guaranteeing displaced workers on-the-job retraining at state expense, with wage insurance covering any income gap during transition. The principle — protecting workers, not jobs — is exactly right.

What is missing is structural. No country has dedicated “technological displacement” social insurance. U.S. workforce development is “chronically underfunded” compared to peer nations. 90% of teachers have never received AI training. Education systems still operate on a model inherited over a century ago. Robot taxes exist in only one country — South Korea, and only as reduced tax deductions.

History’s warning is sobering. During the Industrial Revolution, real wages stagnated for decades even as output per worker rose. The Luddite Rebellion displaced approximately 50,000 textile workers. Stabilizing social contracts emerged only after nearly 200 years of upheaval. The 1990s–2000s computerization wave produced “job polarization” that contributed to an estimated $79 trillion in income flowing from the bottom 90% primarily to the top 1%.

AI may compress these timelines — but whether from two centuries to two decades or two years remains uncertain. The adoption constraints are real: only 9.3% of U.S. companies have used generative AI in production, only 1% of companies have reached “mature” deployment, and Scale AI’s Remote Labor Index shows a 2.5% end-to-end automation rate. The “suddenly” phase may still be years away — but waiting for it to arrive is not a strategy.

Practical Recommendations for Individuals

The evidence points to clear actions that professionals can take now.

Build AI fluency, not just AI skills. The demand for AI fluency in job postings grew sevenfold in two years — from roughly 1 million workers in 2023 to 7 million in 2025. You do not need to become an AI engineer. You need to understand how AI tools apply to your specific domain, where they excel, and where they fail. The AI skills wage premium is 56% and rising.

Invest in what AI cannot do. The OECD framework shows AI at Level 2 in social interaction, metacognition, and physical manipulation. Complex judgment, emotional intelligence, stakeholder management, cultural nuance, creative direction, and ethical reasoning remain distinctly human advantages. Double down on these.

Adopt a portfolio approach to your career. The reinvention data shows that new roles emerge at the intersection of domain expertise and AI capability. Professionals who combine deep industry knowledge with AI fluency will command the highest premium. Do not abandon your domain — augment it.

Document your value in augmented terms. Start measuring and communicating your productivity in AI-augmented workflows. “I produce X with AI assistance in half the time” is the professional narrative of 2026. The organizations that track AI-augmented productivity will reward those who demonstrate it.

Develop verification and oversight skills. As AI reaches the 70–80% capability range on OWCI and GDPval, the premium shifts from producing work to validating, refining, and contextualizing AI-generated work. The role of “human in the loop” is becoming a defined professional skill, not a stopgap.

Build your network deliberately. If AI eliminates the apprenticeship pathway, professional networks become even more critical for career development. Mentorship, cross-functional exposure, and industry communities cannot be replaced by any model.

Practical Recommendations for Organizations

Adopt an augmentation-first strategy. The EY data is unambiguous: 96% report productivity gains, only 17% reduced headcount, and 55% of those who made cuts regret them. Augmentation preserves institutional knowledge, maintains workforce adaptability, and builds compounding advantages. Displacement should be the last resort, not the first instinct.

Measure the right things. Track AI-augmented productivity by role and team, not just aggregate cost reduction. Monitor workforce composition changes — especially at the entry level. Document the skills being developed alongside the skills being automated.

Redesign the entry-level pipeline. If AI absorbs junior-level tasks, organizations must deliberately create new pathways for early-career development. This means structured AI-human apprenticeships, rotation programs that expose junior staff to the work AI cannot do, and mentorship models that compensate for the loss of learning-by-doing.

Invest in reskilling before you need to. The organizations that reinvested AI gains into reskilling (38% per the EY survey) are building the workforce that will sustain their AI advantage. Reskilling after displacement is expensive and slow. Reskilling during augmentation is efficient and builds loyalty.

Run the 5-year NPV on workforce decisions. Cutting 50 junior positions saves money today. But who manages the AI systems in three years? Who provides the judgment, context, and institutional knowledge that makes AI outputs reliable? The short-term savings often create long-term costs that do not appear on the quarterly slide.

Be honest about the timeline. Goldman Sachs projects a 10-year adoption timeline. This is not a sprint. Organizations that pace their transition, communicate transparently with their workforce, and build adjustment capacity will outperform those that lurch between panic and inaction.

Sector-Specific Recommendations

Finance:

• Deploy AI for routine transaction processing and fraud detection, but preserve human judgment for client relationships and complex risk assessment
• Prepare for the 200,000 job restructuring Bloomberg Intelligence projects — begin transition planning now, not when cuts are announced
• Monitor the gap between AI-generated analysis and the senior judgment that interprets it

Healthcare:

• Leverage the 11 million worker shortage as a framework for AI deployment — position AI as filling gaps, not displacing caregivers
• Invest in AI-assisted diagnostics (AI doubles malignant nodule detection rates) while maintaining human clinical oversight
• Prioritize administrative automation (medical transcription, coding, scheduling) to free clinical time

Legal:

• Use the 240 hours annually saved by AI-assisted lawyers to expand access to legal services, not just to boost revenue per partner
• Redesign paralegal roles around AI oversight and quality assurance rather than eliminating them
• Address the hallucinated case citations problem systematically before it becomes a liability issue

Education:

• Scale AI tutoring models that demonstrate double the learning gains — but as supplements to human teachers, not replacements
• Redirect the 5.9 hours per week saved by AI tools toward mentorship, student support, and the relational work that defines great teaching
• Redesign curricula to develop the human skills (judgment, creativity, empathy) that AI cannot replicate

Manufacturing:

• Address the 500,000 unfilled jobs with AI-augmented training and human-robot collaboration rather than pure automation
• Invest in predictive maintenance AI (25% cost reduction, 30% less downtime) as a high-ROI starting point
• Plan for the assembly line workforce declining from 2.1 million to 1.0 million by 2030 — begin transition programs now

Retail:

• Deploy AI chatbots for the 70–80% of standard inquiries they handle well, but invest in human staff for the experiential and complex interactions that drive customer loyalty
• Learn from the self-checkout backlash: automation that degrades customer experience is not a net gain
• Prepare for the 11% cashier decline through 2033 with retraining into customer experience and logistics roles

Creative Industries:

• Invest in AI-augmented creative direction — the roles that are growing even as individual creator positions decline
• Establish clear policies on AI use in creative work, following the precedent set by the SAG-AFTRA and WGA agreements
• Recognize that 26% of creative tasks can be automated, but creative judgment, cultural sensitivity, and original vision remain human domains

Best Practices Checklist

For Individuals:

• Assess your role’s AI exposure using the Anthropic Observed Exposure framework — understand which of your tasks are augmented, which are at risk, and which are uniquely human
• Develop working proficiency with at least two AI tools relevant to your domain within the next 90 days
• Build a personal portfolio that demonstrates AI-augmented productivity, not just traditional output
• Invest in skills that the OECD framework identifies as Level 1–2 for AI: social intelligence, metacognition, ethical reasoning, cultural nuance
• Join professional communities focused on AI adoption in your specific industry
• If you are early-career, actively seek mentorship and cross-functional exposure to compensate for reduced apprenticeship opportunities

For Organizations:

• Conduct an AI exposure audit across all roles, distinguishing between tasks that will be augmented and those that may be displaced
• Establish an augmentation-first policy with clear criteria for when displacement is justified
• Redesign entry-level roles to include AI-human collaborative learning, not just AI-automated output
• Allocate a specific percentage of AI efficiency gains to workforce reskilling (benchmark: 38% per EY survey)
• Create internal AI fluency programs accessible to all employees, not just technical staff
• Monitor workforce composition quarterly, with specific attention to entry-level hiring and mid-career retention
• Run 5-year NPV analyses on all workforce reduction proposals, including the cost of lost institutional knowledge

For Leaders:

• Communicate transparently about AI’s impact on your organization — silence breeds anxiety and turnover
• Engage with policy discussions on workforce transition — the Mauritius model of “protecting workers, not jobs” is a framework worth advocating for
• Support industry-wide apprenticeship redesign initiatives — no single company can solve the entry-level pipeline crisis alone
• Measure and report on the distributional impact of your AI investments — who benefits, who is at risk, and what you are doing about it

Public Policy Recommendations

Redesign social insurance for technological displacement. No country has dedicated insurance for structural technological disruption. Existing unemployment systems were built for cyclical job loss. We need transition insurance that covers reskilling, wage gaps, and geographic mobility for workers displaced by AI — modeled on the Mauritius approach of protecting workers, not jobs.

Reform education at every level. When 90% of teachers have never received AI training and education systems still operate on century-old models, the mismatch is dangerous. Policy should fund AI literacy across all education levels, redesign curricula to emphasize judgment, creativity, and ethical reasoning, and create pathways between displaced roles and emerging ones that do not require master’s degrees.

Close the accessibility gap in new roles. If 77% of new AI jobs require master’s degrees, the transition will calcify inequality rather than resolve it. Governments should fund accelerated credentialing programs, recognize experiential learning, and incentivize employers to create AI roles accessible at multiple education levels.

Invest in the entry-level pipeline. Tax incentives for companies that maintain or expand entry-level hiring in AI-exposed occupations. Subsidized AI-human apprenticeship programs. Public-private partnerships that create the new apprenticeship pathways the market alone will not build.

Address the development divide proactively. When high-income countries control 87% of AI models, 86% of startups, and 91% of venture capital, the risk of AI widening global inequality is not theoretical. International coordination should focus on digital infrastructure investment, technology transfer agreements, and “Small AI” initiatives designed for low-infrastructure environments.

Tax the transition, not the technology. Economists are right that taxing AI infrastructure would be self-defeating. But AI windfall gains — the disproportionate returns captured by firms that automate at scale — should fund the transition costs borne by workers and communities. The principle: those who benefit most from the transition should contribute most to its management.

Coordinate internationally. The fragmentation of AI governance — with the U.S. and UK refusing to sign the Paris AI Summit declaration and geopolitical rivalry fracturing cooperation — undermines every other policy response. AI’s impact is global; the response must be as well.

The Path Forward

I think about this transition across three time horizons.

In the near term (2026–2028), the augmentation phase dominates. Most organizations are still deploying AI for efficiency gains within existing workflows. The adoption constraints are real — only 9.3% of companies in production use, only 1% at mature deployment. The opportunity is to use this window to build the workforce, the institutions, and the policies that will sustain the transition when it accelerates. Organizations that invest in augmentation-first approaches, entry-level pipeline redesign, and workforce reskilling now will compound their advantages.

In the medium term (2028–2032), the capability gap closes. As OWCI and GDPval scores approach and exceed human expert parity, the economic pressure to automate rather than augment will intensify. This is when the entry-level crisis becomes visible, when developing economies feel the full force of reshoring, and when the adequacy of policy responses will be tested. The societies that invested during the augmentation window will navigate this phase. Those that waited will scramble.

In the long term (2032+), the reinvention takes shape. History’s pattern — 60% of today’s workers in roles that did not exist in 1940 — will reassert itself, but the new roles will require capabilities we are not yet developing at scale. The long-term winners will be the economies and organizations that treated this transition as a design challenge, not a crisis to be survived.

At Ibizai, we are building for all three horizons. Our mission — helping businesses understand, adopt, and benefit from AI — is not neutral on the question of how that adoption happens. We advocate for augmentation because the evidence says it works better. We focus on access because the data shows the transition’s costs fall heaviest on those with the fewest resources. We think in time horizons because the decisions made today compound for decades.

The intelligence we are building into our machines is extraordinary. The question that defines this era is whether we bring equivalent intelligence — strategic, ethical, long-term — to the decisions we make about the people those machines affect.

I believe we can. But believing is not enough. We have to build it.

---

Recommended Reading

• World Economic Forum, Future of Jobs Report 2025 — The most comprehensive global survey of employer expectations, projecting 170 million new roles and 92 million displaced by 2030. weforum.org

• Anthropic, Labor Market Impacts of AI: Observed Exposure — The first large-scale measurement of actual AI use mapped to occupational tasks, revealing the critical gap between capability and deployment. anthropic.com

• Erik Brynjolfsson, Danielle Li, Lindsey R. Raymond, Generative AI at Work — The landmark empirical study showing 15% productivity gains from AI assistance, with the largest gains for less experienced workers.

• Brookings Institution, Measuring US Workers’ Capacity to Adapt to AI-Driven Job Displacement — Identifies 6.1 million vulnerable workers and proposes targeted policy interventions. brookings.edu

• UNDP, The Next Great Divergence — Warning that AI could reverse decades of narrowing development inequalities without strong policy action.

• David Autor, New Frontiers: The Origins and Content of New Work, 1940–2018 — The foundational research showing that 60% of current employment is in occupations that did not exist in 1940.

• Scale AI, Remote Labor Index — The sobering reality check: AI agents achieve only 2.5% automation rate on real professional projects.

• Stanford University, HAI AI Index Report 2025 — Comprehensive tracking of AI capability advancement, investment, and policy developments worldwide.