We’re witnessing the AI value delivery paradox: while AI tools allow engineers to generate code at unprecedented speeds, many CTOs have yet to see a corresponding impact on time-to-market, as highlighted in the recent Harvard and Jellyfish study. The reason is simple: deploying agentic AI on its own doesn’t accelerate the software value chain but often stresses it.

Strategic starting points for transformation

The answer depends on your organization's starting point. Typically, they fall into two categories:

  • AI-first digital transformation. If your digital transformation is ongoing or hasn't truly taken root, you face a foundational challenge. This requires rethinking digital architecture, moving from projects to products and shifting from functional to cross-functional teams. Upskilling in TDD, CI/CD and removing technical debt via the "inverse-Conway maneuver" is essential. An AI-first approach can compress this change so rapidly that catching up becomes highly worthwhile.

  • AI transformation. If you already have a mature digital estate, your challenge is unlearning. You’re faced with the sunk cost fallacy and change fatigue while needing to disrupt established architectures and practices.

Realizing the return on AI investments requires leaders to be honest about their current capabilities. AI assistants are powerful but expensive  and smart, AI-powered  product features are expensive to run. Getting the foundation right is critical to avoid speed at the cost of EBITDA, employee satisfaction and customer experience.

Think of living, adaptive systems: Tissue vs. cell

To prevent the organizational dam from breaking with agentic AI, we must think systemically. A lot of moving parts are in play during transformations. Borrowing terms from biology can help us differentiate the core components of a modern tech organization: the "cell" and the "tissue".

  • The cell. This is where products are developed. Thanks to generative AI, the inner loop of development (coding, testing and CI/CD) can be an agentic single, high-speed flow. This is the organ that does the primary work. 

  • The tissue. This is the connective substance that wraps around the cells. It handles cross-cutting concerns like identity, customer channels, security and common data. The goal of the tissue isn’t to introduce new bottlenecks but to provide a collection of smart value chain interfaces and automated guardrails that adapt to cell needs. 

The challenge is to design and run a tissue which is increasingly machine-readable. An AI agent in a cell might generate a perfect feature, but it cannot push until it satisfies the gated surface of the tissue (e.g., MFA or SOC2 compliance) in real-time, without human gatekeepers.