AI, Work, and the Speed of Reality: A Field Guide for Leaders and HR
The loudest story right now says AI will wipe out jobs within 18 months. Some panic. Some cheer. Few look at the data and the real constraints that decide what actually happens inside companies.
This isn't blind optimism or knee-jerk pessimism. The only useful approach is cross-disciplinary: technology, economics, history, and philosophy. As Hayek warned, a single lens creates bad conclusions - and bad policy.
Reality as Competitive Advantage
Consider two companies. One reads radiology images. The other processes medical insurance claims. The first is a fixed problem: consistent inputs, converging answers, more data makes everyone equally good over time. The second is a living system: changing rules, disputes, shifting incentives. You only learn it by doing - and by getting things rejected until you adapt.
That operational "scar tissue" is the moat. AI speeds up learning when rules are stable. It can't speed up regulators, competitors, or the pace at which the real world throws surprises. In complex systems, learning speed is limited by reality, not compute.
The Adoption Gap: Recursive Technology ≠ Recursive Adoption
Models train models; the tech improves itself. Extrapolating that to the economy is the mistake. Adoption is gated by things that don't move at software speed:
- Physical capital and long build cycles (fabs, data centers, logistics)
- Energy availability and grid constraints
- Regulatory approvals and compliance
- Organizational change - the slowest variable of all
Look at U.S. manufacturing construction spending: roughly $75B to $240B between 2021 and 2024. That's historic. And it still takes years, not months, to turn concrete and copper into capacity.
Productivity Shocks Expand Demand
Historically, AI-like shocks cut marginal costs and expand output. Keynes predicted 15-hour workweeks by now. He missed the elasticity of human desire. When costs fall, we don't sit still - we ask for more, better, and newer.
PC prices dropped roughly 99.7% since 1980. That didn't end work; it created the internet, mobile, cloud, e-commerce, streaming, and millions of jobs that didn't exist before. The same logic applies to cognitive work.
Recent data points the same way: new business formation in the U.S. remains unusually high, and demand for software roles has stabilized to around 2019 levels after the post-pandemic correction. Technology is restructuring work, not deleting it wholesale.
"Will AI Replace Us?" is the Wrong Question
The useful question is: how do we stay valuable as AI amplifies output? The answer is to move up a layer - from tasks to systems, from execution to direction, from static playbooks to dynamic judgment.
Core work for the AI-assisted professional:
- Systems design and solution architectures
- Strategy creation that agents can execute
- Business understanding translated into concrete plans
- Skill-building alongside AI
- Critical thinking to steer AI in the right direction
- Deep research with agents to solve real problems
- Metrics, orchestration, monitoring, and governance of systems, agents, and subagents
AI widens access to capability. It does not equalize judgment. Scar tissue - earned by contact with the real world - remains scarce and defensible.
Practical Playbook for HR and Management
- Redesign roles and ladders: Shift job architectures toward outcomes, system stewardship, and cross-functional problem-solving. Update competencies and compensation to reflect agent-assisted work.
- Skill taxonomies and learning paths: Map critical skills (prompting, agent orchestration, data literacy, compliance) and build continuous learning loops tied to performance.
- Governance first: Set policies for data use, model selection, validation, bias controls, and audit trails. Define approval gates for AI in high-risk workflows.
- Pilot where rules are stable: Target static or semi-static processes first (claims triage, routing, QA checks, documentation). Use fast cycles to prove value and de-risk adoption.
- Change management as a program: Communicate role impacts, set expectations, and create incentive structures that reward adoption and quality, not volume of keystrokes.
- Measure what matters: Cost per task, cycle time, error rates, rework, customer satisfaction, and risk incidents. Tie bonuses to improved quality and reduced risk, not just speed.
- Compliance and privacy: Lock down data classification, retention, and model access. Document human-in-the-loop guardrails for regulated decisions.
- Capacity planning: Coordinate with IT on compute, networking, and budget ceilings. Anticipate latency, throughput, and vendor lock-in risks.
- Workforce planning: Anticipate task-level automation. Retrain and reassign early rather than cutting late. Build internal "scar tissue" before competitors do.
For more on org change and governance, see AI for Management. For HR process redesign and people analytics with AI, explore AI for Human Resources.
The Underpriced Scenario: Abundance
Markets have already priced in doom: big selloffs when AI tools compress certain workflows. The hidden flaw is assuming demand is fixed. When cognitive costs fall, demand expands and new categories emerge.
Two useful ideas: "Ghost GDP" (growth that doesn't reach households) versus "Abundance GDP" (growth with falling service prices). The optimistic path doesn't require higher nominal wages - it requires cheaper healthcare admin, legal, accounting, education, and support. Households win when service prices drop faster than incomes.
Early signals are here: U.S. labor productivity has accelerated to its fastest clip in two decades. That is what a positive supply shock looks like. See the Nonfarm Business Productivity index for context.
What This Means
Three points line up. Reality creates friction that AI can't simulate - operational scar tissue remains a moat. Technology improves recursively, but adoption doesn't. And the most underpriced future isn't collapse; it's abundance built on cheaper cognition and expanding demand.
Your edge is judgment, coordination, and the willingness to earn scar tissue faster than others. Pair agents with people who think clearly, measure honestly, and iterate with the grain of reality. That's how teams stay valuable - and get more valuable - as AI scales.
Sources and References
- Beyer, David - Analysis on AI limits in complex systems and the concept of operational scar tissue.
- Citadel Securities - "Global Intelligence Crisis 2026": Recursive technology vs. recursive adoption and physical limits.
- The Kobeissi Letter - "It's Too Obvious. What If AI Doesn't Actually End The World?"
- Penrose, Roger - The Road to Reality: A Complete Guide to the Laws of the Universe. Knopf, 2005.
- Hayek, Friedrich - Writings on interdisciplinary economics and the risks of narrow specialization.
- Data series from FRED (Federal Reserve Bank of St. Louis): Business Applications (BABATOTALSAUS), Indeed Job Postings: Software Development (IHLIDXUSTPSOFTDEVE), Total Construction Spending: Manufacturing (TLMFGCONS), Nonfarm Business Real Output Per Hour (OPHNFB), PCE Price Index: Computers and Related Equipment (DIPERG3A086NBEA). See FRED for datasets.
Your membership also unlocks:
