top of page

AI as a fair growth engine, if we choose it

A dawn aerial of a river city where a bridge glows with golden light trails forming a balanced-scale silhouette linking both sides.

Dek. Artificial intelligence can amplify inequality or rebalance opportunity. This essay sets out the best outcomes we could plausibly achieve, the real risks, and a leadership playbook to steer toward an economy that works for everyone.


What if the next wave of productivity growth did not just raise profit margins but lifted wages across the income distribution and opened new ladders for people who missed the last digital boom? It is possible. The IMF estimates that about forty percent of current jobs worldwide are meaningfully exposed to AI, with sixty percent in advanced economies. Exposure can mean augmentation or displacement, and the difference will hinge on the choices we make in the next few years (IMF, 2024).


Why this matters for managers and founders is straightforward. The early productivity results from real firms are not hype. In customer support, access to a conversational assistant increased resolutions per hour by about fifteen percent, with the most significant gains for less experienced staff. In software work, a controlled trial found more than fifty percent faster task completion with an AI partner programmer. In professional writing, the time to complete tasks fell by forty percent while the quality increased (Brynjolfsson et al., 2025; Peng et al., 2023; Noy and Zhang, 2023).


The question is not whether AI will move the needle. It will. The question is whether we allow it to deepen divides or use it to rebalance our economy.


I. What changed in the market


Takeaway. This wave scales decision-making and creative tasks.


Unlike past automation focused on routine manual tasks, modern AI reaches into non-routine cognitive work. That means higher wage roles are exposed, at least in part. OECD analyses show that exposure is widespread across white-collar occupations, with skills such as management and business process design increasingly in demand alongside AI literacy (OECD, 2024).


Adoption is also riding on a small number of global platforms. A handful of cloud providers now capture a large share of enterprise infrastructure spend. Estimates for 2025 show quarterly cloud revenues approaching $100 billion, with the three largest providers consistently dominating. This concentration matters because compute and data access are gatekeepers to capability and cost (Synergy Research Group, 2025; OECD, 2025)


Policy, too, has shifted. The EU AI Act entered into force in August 2024, with significant obligations phased in through 2026, while NIST issued a generative AI profile to guide risk management. Governments convened safety summits from Bletchley to Seoul to align on testing and accountability for frontier systems (European Commission, 2024; NIST, 2024; UK Government, 2023; AP News, 2024).


Video explaining the EU AI Act

So the playing field is different. The technology targets both the head office and the shop floor. The infrastructure is concentrated. The rulebook is emerging in real time.


II. The best outcomes we could plausibly achieve


Takeaway. Augment people, spread tools, and you can get broad productivity by narrowing gaps.


Field and lab evidence points to three bright spots.


First, AI can compress performance gaps. In call centres, new or lower-skilled agents improved the most when assisted by AI. That is what you want if your aim is inclusive productivity growth, and it is consistent with early evidence of reduced wage dispersion within occupations exposed to AI (Brynjolfsson et al., 2025; OECD, 2024).


Second, AI can unlock the capabilities of small firms. Suppose a three-person business can draft policy-compliant contracts, produce serviceable code, segment customers, and prepare data dashboards in a morning. In that case, the returns to scale enjoyed by large incumbents are less automatic. McKinsey’s macro estimates are only that, estimates, but they show order-of-magnitude potential value across customer operations, marketing, software, and R&D at scale (McKinsey, 2023).


Third, it can change who gets to participate. In professional writing and consulting tasks, generative tools reduced time and increased quality across skill levels. In software, the most significant gains often occur among less experienced developers. That is a route to faster learning and progression if firms pair tools with coaching and fair evaluation (Noy and Zhang, 2023; Dell’Acqua et al., 2023; Peng et al., 2023).


None of this happens by magic. It happens when leaders design for augmentation, not simple substitution.


III. The inequality traps in plain view


Takeaway. Left to drift, AI can widen gaps through capital deepening, superstar dynamics, and uneven transitions.


Start with market power. The cloud and foundation model markets are highly concentrated. That can tilt the balance of bargaining power toward providers and away from smaller customers and creators. Competition authorities are watching, but the economics still favour scale (OECD, 2025; Synergy Research Group, 2025).


Then consider labour markets. A large global share of jobs is exposed to AI. Exposure in advanced economies is particularly high because cognitive tasks dominate. Without policy and firm choices that share gains and support mobility, exposure risks translate into wage pressure, mainly where rents are dissipated by targeted automation (IMF, 2024; Acemoglu and Restrepo, 2024)


The regional picture is uneven as well. Some economies have a high share of jobs complementary to AI and the infrastructure to exploit that, while others may lag. OECD work suggests that regions already ahead in skills and digital capacity will pull away without intentional diffusion strategies (IMF, 2025; OECD, 2024).


Finally, skills and adoption are lumpy. Many firms are still figuring out where AI truly helps. In the BCG study, people excelled when using AI for tasks within its frontier, but performance suffered when they leaned on it for tasks beyond that, and did not check its work. Without coaching and workflow redesign, inequality of outcomes can rise within the same firm (Dell’Acqua et al., 2023).


IV. Reconciling conflicting numbers


Takeaway. Different studies measure different things; do not mix them.


You will see very different figures in the press. One set estimates potential GDP effects from broad adoption over a decade. Another measures actual task-level productivity in a given setting. A third studies wage dispersion within and between occupations.


I switched to Thinking to reconcile these. I use IMF figures for the global scale of exposure because they summarise task exposure across countries. I use peer-reviewed or working paper evidence for measured productivity effects inside firms. I use the OECD for early signs of wage dispersion within occupations. These are complementary, not contradictory, lenses (IMF, 2024; Brynjolfsson et al., 2025; OECD, 2024).


V. A leadership playbook for equitable AI growth


Takeaway. Design the work, share the gains, build the commons, and harden the guardrails.


1. Design for augmentation. Start with tasks, not job titles. Map where AI can speed drafting, search, summarisation, or code suggestions, then redesign handoffs so people stay in the loop. Use pair programming or pair analysis as the default. This pattern produces the biggest and most inclusive gains in the field evidence (Brynjolfsson et al., 2025; Peng et al., 2023).


2. Share the gains. When teams move output or quality metrics through AI-assisted methods, raise wages or reduce hours while holding pay constant. Publish a simple shared gains formula so the default split is transparent. This keeps buy-in high and reduces the incentive to cut headcount where demand is growing.


3. Invest in mobility, not just skills. Training is necessary but not sufficient. Create internal pathways that let people move from shrinking tasks to growth tasks with real support and time. OECD guidance stresses the need to anticipate skill needs and build programmes at all levels (OECD, 2024).


4. Build common assets, pool data inside your sector where competition law allows. Contribute to open source tooling, evaluation datasets, and safety benchmarks. Small firms get better access, and everyone gets better quality assurance. Nationally, governments should invest in compute access for research and small enterprise use, as proposed in the OECD compute blueprint (OECD, 2024).


5. Strengthen guardrails and competition. Adopt a risk management framework aligned to NIST, with model and prompt logging, red team routines, and incident reporting. Support regulator capacity to test and enforce under the new EU regime and equivalents. Watch for vertical restraints and bundling in cloud and model markets (NIST, 2024; European Commission, 2024; OECD, 2025).


6. Fiscal policy that smooths the path. Where AI generates excess profits from market power or windfall timing, use mainstream tax instruments to fund transition support. IMF work argues for strengthening social protection and reviewing capital income taxation to broaden the gains from generative AI (IMF, 2024).


Evidence block

Methods. I compiled a small table of four widely cited studies that used field or controlled designs to measure task-level productivity effects from AI assistance. The sample spans customer support, writing, coding, and consulting tasks. I then computed and plotted the reported percentage improvements. Limits include small sample external validity and task specificity.


AI Productivity gains from selected studies


Wrap up

Action checklist for leaders

  • Map tasks and redesign workflows so people remain in the loop where judgment matters.

  • Create a shared gains formula that ties productivity improvements to wages or time.

  • Stand up a mobility programme with real projects and mentoring, not just courses.

  • Contribute to sector data and evaluation commons and lobby for compute access for small firms.

  • Adopt NIST-aligned risk controls and engage with regulators early on tests and disclosures.


From the professor’s desk

I recently sat in a meeting where a manager pitched AI as an instant headcount reducer. The numbers looked tidy. The risks looked invisible. We took a breath and asked a different question. What would it take to get the same productivity gains while increasing internal capability and maintaining trust? This choice changed the plan. Technology should not decide that outcome. People did.


References

Acemoglu, D. (2024). The simple macroeconomics of AI. MIT Shaping the Future of Work Initiative. https://shapingwork.mit.edu/wp-content/uploads/2024/05/Acemoglu_Macroeconomics-of-AI_May-2024.pdf Massachusetts Institute of Technology

Acemoglu, D., & Restrepo, P. (2024). Automation and rent dissipation: Implications for wages, inequality, and productivity [Working paper]. https://economics.mit.edu/sites/default/files/2024-05/Automation%20and%20Rent%20Dissipation%20-%20Implications%20for%20Wages%20Inequality%20and%20Productivity.pdf MIT Economics

AP News. (2024). AI companies make fresh safety promise at Seoul summit. https://apnews.com/article/2cc2b297872d860edc60545d5a5cf598 AP News

Brynjolfsson, E., Li, D., & Raymond, L. R. (2025). Generative AI at work. Quarterly Journal of Economics, 140(2), 889–932. https://academic.oup.com/qje/article/140/2/889/7990658 Oxford Academic

Dell’Acqua, F., Gino, F., Lakhani, K. R., et al. (2023). Navigating the jagged technological frontier: Field experimental evidence of the effects of AI on knowledge worker productivity and quality [Working paper]. Harvard Business School. https://www.hbs.edu/ris/Publication%20Files/24-013_d9b45b68-9e74-42d6-a1c6-c72fb70c7282.pdf Harvard Business School

IMF. (2025). How artificial intelligence will affect Asia’s economies. https://www.imf.org/en/Blogs/Articles/2025/01/05/how-artificial-intelligence-will-affect-asias-economies IMF

NIST. (2024). AI risk management framework: Generative AI profile. https://nvlpubs.nist.gov/nistpubs/ai/NIST.AI.600-1.pdf NIST Publications

Noy, S., & Zhang, W. (2023). Experimental evidence on the productivity effects of generative AI. Science, 381(6654), 187–192. https://www.science.org/doi/10.1126/science.adh2586 Science

Peng, S., Kalliamvakou, E., Cihon, P., & Demirer, M. (2023). The impact of AI on developer productivity: Evidence from GitHub Copilot. arXiv. https://arxiv.org/pdf/2302.06590 arXiv

Stanford HAI. (2024). AI Index Report 2024. https://hai.stanford.edu/ai-index/2024-ai-index-report Stanford HAI

White and Case. (2024). Long awaited EU AI Act becomes law after publication in the Official Journal [Client note]. https://www.whitecase.com/insight-alert/long-awaited-eu-ai-act-becomes-law-after-publication-eus-official-journal White & Case

1 Comment

Rated 0 out of 5 stars.
No ratings yet
Add a rating
Guest
2 days ago
Rated 5 out of 5 stars.

Very interesting

Like
bottom of page