How AI Could Reshape Computers and Computing
Wajih Alkhiami — July 20, 2026
Overview
AI is shifting from an optional application to part of computing’s underlying infrastructure. Personal computers increasingly include neural processing units that run AI workloads locally, reducing latency, limiting reliance on cloud services, and potentially keeping more sensitive data on the device. AI-capable PCs represented 14% of global quarterly shipments in the second quarter of 2024 and were projected to reach 43% of shipments in 2025.
Software development is changing too. Coding assistants are evolving into agents that can navigate repositories, run tests, use development tools, and complete multi-step tasks. Operating systems may eventually use AI to improve process scheduling, memory management, and security, while specialized system architectures could support many concurrent agents more efficiently.
Progress, however, is uneven. AI performs impressively on some coding benchmarks yet still struggles with basic tasks. Its effect on productivity also varies sharply by environment, making the future of computing promising but difficult to forecast confidently.
Boomer Perspective
The optimistic case begins with more capable, responsive computers. On-device AI could summarize documents, analyze local data, or assist creative work without repeatedly sending information to remote servers. Dedicated processors and increasingly efficient Arm-based designs may make these features faster and more power-conscious.
Developers could delegate repetitive work to agents while focusing on architecture and product decisions. Controlled studies have reported time savings ranging from 13.6% to 55.8%. One enterprise deployment reduced pull-request review cycles by 31.8% and increased code shipment volume by 28%. Used carefully, AI may also narrow skill gaps by helping people debug code and work with unfamiliar systems.
Efficiency is improving as well: AI hardware costs have declined by about 30% annually, while energy efficiency has improved by roughly 40% each year. AI’s power needs are also encouraging renewable-energy procurement, battery storage, and investment in new energy technologies.
Doomer Perspective
The cautious view is that faster generation does not guarantee better computing. In a randomized trial involving 16 experienced open-source developers, early-2025 AI tools increased task completion time by 19%, even though participants expected a 24% speedup. Mature codebases require organizational context, architectural judgment, and extensive verification that agents may lack.
Security risks grow when agents can access files, execute commands, or call external tools. Prompt injection can convert malicious content into unauthorized actions or data exposure. Repeated AI refinement can also degrade code: one study found a 37.6% increase in critical vulnerabilities after five iterations without human oversight.
Energy presents another constraint. U.S. data centers consumed about 176 TWh in 2023, or 4.4% of national electricity, and could reach 325–580 TWh by 2028. Cooling alone may consume 38%–40% of a facility’s electricity. Grid delays, equipment shortages, and rising local costs could limit expansion.
Balanced Analysis
Both perspectives are supported, but they describe different conditions. Optimists are strongest where tasks are structured, local processing improves privacy or speed, and organizations retain expert review. Pessimists are better supported in complex, high-risk environments where verification, security, and infrastructure costs can erase apparent gains.
The likely outcome is neither effortless automation nor inevitable failure. AI should make computers more adaptive and development more automated, but benefits will depend on permission boundaries, audit logs, human validation, secure agent-tool interfaces, and realistic energy planning. Only 25% of organizations had fully implemented comprehensive AI governance programs, suggesting that institutional readiness—not raw model capability—may determine how valuable the next generation of computing becomes.
