AI tools speed routine coding and prototyping at Netlight but create errors, context gaps and integration costs that increase the value of human supervision and orchestration. Firms that build effective orchestration and tooling layers — and hire for system-design and coordination skills — are likely to capture disproportionate productivity gains.

Main Finding

Experienced IT professionals at Netlight are already integrating AI tools into everyday development work, changing how tasks are done rather than replacing developers outright. However, current AI capabilities create new frictions (quality, context, coordination, governance) that make human roles of supervision, orchestration, and integration more valuable. The authors argue we are moving toward “AI orchestras” — multilayered systems of specialized AI components coordinated by humans — which will reshape work, firm organization, and the economics of IT production.

Key Points

• AI is being used as a practical assistant in coding, debugging, prototyping, and knowledge retrieval rather than as a fully autonomous developer.
• Common uses include: generating code snippets, suggesting fixes, accelerating routine tasks, surfacing design patterns or documentation, and scaffolding prototypes.
• Major limitations and frictions reported by practitioners:
  • Errors and hallucinations: output can be incorrect, incomplete, or misleading.
  • Context gaps: AI lacks full project context, design rationale, and long-term constraints.
  • Integration cost: AI outputs often need human revision, testing, and integration into existing systems.
  • Quality, security, and IP concerns: generated code may introduce vulnerabilities or licensing ambiguity.
  • Tooling and workflow mismatch: current AI tools are not always well integrated into team processes or CI/CD pipelines.
• Human roles shift toward oversight, curation, specification, and orchestration of multiple AI components and tools.
• The “AI orchestra” concept: future development will involve coordinated ensembles of specialized AI agents (code generation, test generation, dependency analysis, security scanning, etc.) whose outputs must be orchestrated by humans and higher-level controllers.
• This orchestration increases demand for skills in system design, AI tooling, and coordination — not just coding.

Data & Methods

• Basis: a field study of experienced IT professionals at Netlight Consulting GmbH (a Swedish-founded, internationally operating IT consultancy).
• The paper draws on first-hand empirical work with practitioners to document concrete AI uses, difficulties, and evolving practices.
• Methods (as reported): qualitative, practice-focused analysis of how AI is embedded in day-to-day IT development; evidence appears to come from interviews, observations, and analysis of workflows and tools used by experienced consultants.
• Limitations of the evidence:
  • Single-firm focus (Netlight) may limit generalizability across industries, firm sizes, or geographic contexts.
  • Fast-evolving AI toolchains mean findings could shift as models and integrations improve.
  • The study emphasizes practitioner experience and may not quantify productivity or labor-market outcomes.

Implications for AI Economics

• Task reallocation and complementarities: AI substitutes for routine coding work but complements higher-order tasks (system architecture, integration, orchestration). Demand will rise for complementary skills (coordination, specification, AI-tooling expertise).
• Productivity measurement: Standard metrics may understate productivity gains or shifts because AI changes the mix of tasks and introduces new coordination costs; careful measurement of output quality and integration time is necessary.
• Wage and skill dynamics: Skilled developers who can orchestrate AI and manage complex workflows may see increased wage premiums; mid-level routine tasks face downward pressure or require upskilling.
• Organizational change and returns to scale: Firms that build effective orchestration layers and integrate AI across pipelines may capture outsized gains, increasing winner-take-all dynamics and concentration in IT services and platforms.
• Markets for complementary goods: Demand will grow for orchestration platforms, testing/verification tools, secure code-generation services, and team-level integrations—creating new market segments and monopolistic-leveraging opportunities for dominant tool providers.
• Employment and task scope: While AI may reduce time on coding chores, it may expand demand for roles that supervise AI ensembles, audit outputs (security, compliance), and maintain long-term system health.
• Policy and governance: Issues of liability, IP, security, and certification of AI-generated code become salient. Regulators and firms will need standards for provenance, testing, and accountability in AI-assisted development.
• Research needs: Quantitative studies to measure net productivity effects, longitudinal work on skill trajectories, and comparative analyses across firm types and industries to assess general equilibrium effects.