Eighteen months after the first GPT-5 release and ten months after GPT-5.5 brought extended reasoning to general availability, enterprise AI has moved from pilot programmes to material P&L line items. This is the institutional read on what the adoption curves actually show, where the productivity gains are concentrated, and which displacement signals are now visible in S&P 500 earnings transcripts.
ADOPTION CURVES: WHERE ENTERPRISES ACTUALLY DEPLOYED
Enterprise survey data from Q1 2026 shows AI deployment penetration above 70 percent in three functional areas: software engineering, customer support, and back-office knowledge work. Deployment in regulated functions such as legal, compliance, and clinical decisioning sits between 20 and 35 percent, gated by audit and explainability requirements rather than technical capability. Sales and marketing show high adoption of consumer-grade tools but low adoption of integrated enterprise systems.
Key Takeaway
The cost-per-task collapse
Frontier model API pricing has fallen roughly 95 percent for equivalent capability between GPT-4 launch and GPT-5.5 general availability. A standard customer-support ticket resolution that cost 0.42 USD in inference compute in early 2024 now costs under 0.025 USD on equivalent quality. This compression is the single most important variable behind enterprise economics: workloads that did not pencil out in 2024 are obviously profitable in 2026.
FUNCTIONAL BREAKDOWN: WHAT MOVES THE NEEDLE
Software engineering
Code-generation tools have produced measurable throughput gains across surveyed engineering organisations. The median acceleration sits at 32 percent for senior engineers and 48 percent for junior engineers. The gap is consequential: AI tooling is a more powerful productivity lever for less experienced engineers, which is reshaping early-career hiring and apprenticeship economics.
Customer support and contact centres
AI-first deflection now resolves 38 to 52 percent of support tickets without human escalation across surveyed enterprises. Cost per resolved ticket has fallen 60 to 75 percent. The labour implications are direct: aggregate US contact-centre employment is down 11.2 percent year-over-year as of April 2026 BLS data, the steepest single-year decline outside recessions.
Knowledge work and analyst functions
The most contested category. Survey-reported productivity gains are wide, ranging from 15 to 70 percent depending on task design. Where AI does the synthesis and a human does the judgement, gains are large and durable. Where the task is genuinely judgement-led, gains are smaller and noisier. The institutional read is that knowledge-work displacement is real but slower-moving than support displacement.
We are not reducing headcount across the board. We are reducing it in specific functions while expanding it in others. The net is roughly flat. The composition is fundamentally changing.
WHAT THE EARNINGS CALLS SHOW
A scan of Q1 2026 S&P 500 earnings transcripts shows AI mentions in 82 percent of calls. More telling: 31 percent of calls include specific productivity numbers, 24 percent include headcount commentary tied to AI, and 14 percent disclose discrete cost-savings figures. The disclosures cluster in financial services, professional services, and technology. They are notably absent from industrial and consumer staples, where deployment lags.
WHAT THE PRODUCTIVITY MARKETING GETS WRONG
Three claims appear in vendor pitch decks and do not survive scrutiny. First, that AI uniformly accelerates senior workers; the evidence shows junior workers benefit disproportionately. Second, that integration is easy; the median enterprise rollout takes 14 months from pilot to material deployment. Third, that ROI is immediate; the cost of integration, training, and workflow redesign typically takes 9 to 18 months to recover.
RISKS THROUGH YEAR-END 2026
- Model concentration risk: over 80 percent of enterprise AI workloads run on three providers. Outage or pricing changes have systemic productivity implications.
- Regulatory drift: EU AI Act enforcement and US state-level AI hiring laws are reshaping deployment in HR and recruiting workflows.
- Skills mismatch: labour markets are not yet reabsorbing displaced workers into AI-adjacent roles at the rate displacement is occurring.
Key Takeaway
METHODOLOGY AND DATA SOURCES
Adoption data from McKinsey State of AI 2026, Stanford AI Index 2026, and Slack Workforce Index. Labour data from US Bureau of Labor Statistics CES and JOLTS through April 2026. Earnings transcripts from public Q1 2026 filings. Editorial research; not investment advice or recommendation on staffing decisions.
Related coverage: AI Labor Displacement Q2 2026 Data, AI x Crypto Convergence 2026, Humanoid Robotics Commercialisation 2026.
Share this brief