AI & Workforce Impact

AI's impact on work is the most contested topic in the field — predictions range from "10% of tasks automated across every job" to "most white-collar work replaced within 10 years." This page covers what the evidence actually shows, which roles and tasks are most exposed, and how practitioners should think about this.

What the Research Says

Several rigorous studies have attempted to measure AI's labour market effects. Key findings as of 2025–2026:

• Goldman Sachs (2023): Estimated 300M jobs globally "exposed" to automation, with 18% of work tasks potentially automated by AI. But "exposed" ≠ "eliminated" — most exposed roles see task substitution, not job elimination.
• McKinsey Global Institute (2024): Projected that 12 million US occupational transitions by 2030 are attributable to AI (out of a workforce of ~160M). Most displaced workers move to adjacent roles, not permanent unemployment.
• MIT/NBER studies (2024–2025): Found that AI coding assistants (GitHub Copilot) increased developer productivity 55–26% on specific tasks, but overall employment in software development continued to grow — demand expanded with supply.
• World Economic Forum (2025): Estimated 85M jobs displaced globally by 2025 (ahead of projection) but 97M new roles created — net positive, but with significant geographic and skill mismatch.

The consistent finding: AI substitutes tasks more readily than jobs. Most jobs are bundles of tasks; rarely does AI automate the entire bundle.

Most Exposed Roles and Tasks

Augmentation vs Displacement

The augmentation vs displacement framing captures two distinct outcomes for the same technology:

• Augmentation: A software developer uses Claude to write boilerplate 3× faster, freeing time for architecture and design. Output doubles; headcount stays flat. The developer is more valuable, not less.
• Displacement: A company realises that 10 junior writers can be replaced by 2 senior writers using AI, since AI handles the first draft and structural work that juniors did. The juniors are let go.

Both outcomes coexist. Which one applies depends on:

• Task structure: How separable is the AI-substitutable portion from the rest of the role?
• Firm strategy: Does the company capture AI gains as productivity improvements (same output, lower cost) or as capability expansions (more output, same cost)?
• Market demand elasticity: If AI makes legal services 3× cheaper, does demand for legal services increase 3× (absorbing the freed capacity) or does total spend stay constant (eliminating 2/3 of lawyers)?

What About Software Developers Specifically?

Given the audience for this site, it's worth being direct: software development is among the most AI-exposed knowledge work categories. By 2025:

• Agentic coding tools (Claude Code, Cursor, Copilot Workspace) can complete multi-file changes from a single high-level prompt
• Scaffold generation, boilerplate, test writing, and documentation are largely automatable
• Junior developer tasks (implementing a spec, writing CRUD endpoints, fixing well-described bugs) are increasingly handled by AI

However, employment in software development has continued to grow through 2025. The reason: AI increases the productivity of senior developers, allowing them to build more — expanding the total amount of software being built. AI has democratised the ability to ship software (via tools like Lovable, Bolt.new, and Claude Code), creating demand for software in markets that previously couldn't afford custom development.

The likely outcome: fewer junior developers needed per senior developer, but more software projects getting built overall. The mix of skills that matters is shifting toward architecture, system design, product judgment, and AI orchestration — away from routine implementation.

How to Think About Reskilling

The most durable professional advantage in an AI-augmented world comes from skills that are hard to automate and that AI amplifies:

• Taste and judgment: Knowing what good output looks like — the ability to evaluate AI-generated work critically — is scarce and valuable. AI produces infinite drafts; the skill is choosing and improving the right one.
• Domain depth: AI makes a generalist more productive, but it amplifies deep domain expertise the most. A doctor using AI produces better medicine than a non-doctor using the same AI. Domain knowledge provides the evaluation capacity AI lacks.
• Orchestration and prompt engineering: Getting AI systems to do complex multi-step work reliably is a learnable skill. This is not just "prompt hacking" — it includes system design, evaluation, and workflow architecture.
• Communication and stakeholder management: AI doesn't manage client relationships, navigate organisational politics, or communicate bad news. These remain fundamentally human.
• Learning agility: The half-life of specific AI tool skills is short. The most durable skill is the ability to learn new tools quickly — staying current with a fast-moving landscape.

See also: AI Strategy → AI Workforce Impact & Change Management — practical change management, reskilling strategy, and how to manage AI adoption across your organisation.

Checklist: Do You Understand This?

• AI substitutes tasks more readily than entire jobs — most roles are task bundles, rarely fully automatable
• High-exposure tasks: routine content, data entry, boilerplate code, tier-1 support, basic translations
• Lower-exposure work: novel judgment, high-stakes relationships, creative direction, skilled trades
• Augmentation and displacement coexist — which applies depends on task structure, firm strategy, and market demand
• Software development is AI-exposed but employment has grown — demand expands with productivity
• Durable skills: taste and judgment, domain depth, AI orchestration, communication, learning agility