Unlock the Potential of Generative AI in Enterprise Technology

New Patterns of Work with Gen AI

In enterprise technology, a fundamental evolution is underway in how tech teams operate. Beyond using gen AI tools for individual productivity boost, leaders are restructuring entire processes and workflows. Two new human-AI interaction patterns are emerging: the "factory" and the "artisan".The "factory" model deploys autonomous gen AI-enabled agents to handle end-to-end work in predictable, routine processes like log monitoring or legacy code migration. Early results show organizations can modernize code nearly twice as fast by orchestrating these agents.The "artisan" pattern uses gen AI tools as assistants for non-deterministic processes that require human judgment and ingenuity, such as cost management and vendor sourcing. Each enterprise technology domain and use case may demand the right human-AI team model. Leaders face the challenge of blending these approaches seamlessly to create a synchronized workflow. To support this, a framework and governance strategy are essential to guide efforts and define the division of labor between human and AI.

New roles and skills are also needed. For AI-led "factory" tasks, supervisors are required to oversee audit mechanisms and ensure accuracy and trust. In human-led processes, experts need to expand their skills, like rapidly iterating solutions. As the adoption of these patterns spreads, technical debt is expected to decrease, allowing staff to focus on innovation. This leads to increased innovation scale and faster IT capabilities with lower costs, driving a strategic reallocation within the enterprise technology portfolio.

Shifting IT Architectures

IT architectures are transforming from traditional application-focused to multiagent architectures. Tech leaders oversee hundreds or thousands of distinct gen AI agents that communicate and collaborate to achieve common goals.Super platforms are the next generation of third-party business applications with built-in gen AI agents, acting as commodities for rapid service. AI wrappers enable enterprise services to communicate with third-party services via APIs without exposing proprietary data. Custom AI agents are internally developed by fine-tuning pretrained models with proprietary data.The choice of platform strategy depends on factors like proprietary data. Super platforms give vendors access to data, while AI wrappers secure internal data. Designing and managing multiagent architectures requires different considerations from application-centric ones. Modular frameworks guide the development of reusable agents, and architectures include various types of agents like orchestration and communicator agents. A central enterprise technology team provides the platform with data layers and tools, while data science and business teams contribute agents for specific problems. The priority shifts to continuous improvement of gen AI agents and underlying data sets.

Organizational Structure and Workforce Development

As automation and AI-human collaboration expand, tech leaders are reshaping organizational structures. In domains relevant to the artisan team model, mid-level employees take on integrated roles spanning strategy and execution. In factory model domains, there is a flattening of the organization with fewer junior roles and a need for supervisors.Communication becomes a critical skill to ensure effective engagement. Senior and junior expert roles change as AI automates tasks. A flatter and potentially leaner organizational model may emerge. This leads to a skills revolution with considerations for workforce and career development. Questions arise about upskilling staff, training senior experts for new roles, creating talent pipelines, and leveraging gen AI without relying on it. Clear baselines and metrics are needed to measure the impact of gen AI on workforce planning.Infrastructure costs also shift as staff productivity rises. Tech leaders need to focus on compute spend to support gen AI agents and manage risks. Evaluating activities for AI-led processes and conducting cost-optimization reviews during agent development can help control costs. Monitoring compute spend during operation is crucial to avoid runaway costs. Using a FinOps as code approach provides real-time insights for effective cost management.

Full adoption of this new enterprise technology model is a long-term goal that requires more than just tools. It depends on understanding where to implement factory and artisan patterns, designing agent architectures, and preparing for talent, cost, and risk implications. Starting with a few domains can help build organizational capabilities and gain efficiencies as they expand.