By Krasimir Baylov, Managing Partner, Intway
When Productivity Becomes Misleading
Modern IT organizations are delivering faster than ever. AI-assisted development, low-code platforms, and automation pipelines have significantly increased how quickly features can be produced. Tasks like code generation and infrastructure provisioning that once took weeks can now be done in hours.
And yet, a paradox exists. Despite this speed, many projects still face delays. Quality issues arise more often. Teams report a heavier cognitive load and ongoing interruptions. The feeling of “always being busy” has not gone away; it has only intensified. The root of this contradiction lies in a concept that has long been present but is now impossible to ignore.
This is invisible work. Popularized in books like “The Goal” and later applied to IT in “The Phoenix Project,” invisible work refers to the effort that uses up capacity without being formally recognized in plans or metrics. What was once an unseen inefficiency is now becoming the main constraint in AI-driven systems.
Defining Invisible Work in Modern IT Systems
Invisible work can be defined as the effort needed to sustain delivery that isn’t explicitly shown in plans, metrics, or architectural artifacts. It occurs at all levels of an organization and shows up in several forms.
Coordination work involves aligning teams, managing dependencies, and constantly negotiating to push work forward in complex settings.
Cognitive work includes decision-making under uncertainty, maintaining mental models of systems, and understanding trade-offs that are rarely documented.
Unplanned work appears as incidents, urgent fixes, and interruptions that disrupt planned delivery.
Quality compensation involves debugging, rework, and validation, often caused by earlier shortcuts or uncertainty.
These activities rarely show up in architecture diagrams or delivery dashboards. Yet, they determine whether systems work reliably in real-world conditions.
From Code-Centric to Control-Centric Work
Historically, most effort in software delivery focused on writing code. Architecture, while vital, often played a supporting role to implementation. AI is fundamentally changing that balance.
In AI-first environments, code generation is no longer the main bottleneck. Instead, the focus shifts to:
- validating generated outputs
- controlling system behavior
- defining clear operational and architectural boundaries
AI does not eliminate work; it redistributes it into forms that are less visible and harder to measure. The result is a shift from code-centric work to control-centric work, where understanding, verification, and governance become the key activities.
Invisible Work in AI-First Applications
AI-first applications, where a large portion of code is generated instead of written, introduce new categories of invisible work that are often overlooked.
Prompting as Hidden Engineering
Prompt design is naturally iterative. It involves experimentation, refinement, and ongoing adjustments. Despite its significance, this effort is rarely tracked as development work, even though it directly affects system behavior and quality.
Validation as the New Development
As AI-generated output increases, the need for validation also rises. Verifying correctness, spotting hallucinations, and testing edge cases demand significant effort and in-depth expertise. In many cases, the effort needed to trust the system exceeds the effort required to create it.
AI-Generated Technical Debt
AI can quickly produce functional code, but often lacks strong architectural coherence. This leads to a new kind of technical debt – subtle, dispersed, and hard to detect early. Since delivery seems fast, this debt builds up quietly until it results in instability or lower maintainability.
Governance and Compliance Overhead
AI introduces new governance requirements, such as data privacy, model usage rules, and regulatory constraints. Ensuring compliance adds another layer of invisible work, often separate from traditional development workflows.
Architectural Boundary Definition
One of the most crucial and least visible tasks is defining what AI should and should not do. Determining system boundaries, setting up guardrails, and identifying where determinism is needed are fundamentally architectural issues.
Architecture is no longer just about designing systems. It is about ensuring systems behave predictably.
Invisible Work in Team Dynamics and Enterprise Architecture
Invisible work is not just a technical issue. It is deeply rooted in how teams and organizations function.
The Hidden Load in Teams
Teams continuously engage in alignment, translation, and coordination across different areas. Dependencies must be negotiated, knowledge must be shared, and assumptions must be clarified. Much of this work occurs informally and is never reflected in delivery metrics.
The Architect’s Invisible Burden
Enterprise architects bear a significant invisible load. Their work includes assessing trade-offs, maintaining conceptual integrity, and connecting business with technology. This work is largely cognitive. It involves thinking, anticipating, and negotiating. These activities are essential but seldom visible.
Governance as Stabilization Work
Governance is often viewed as unnecessary overhead or bureaucracy. However, from a systems perspective, governance represents stabilization work. These methods ensure consistency, compliance, and long-term sustainability.
When this work is overlooked or minimized, organizations may gain short-term speed but lose long-term resilience.
Why Invisible Work Matters More in AI-Driven Enterprises
AI boosts both productivity and complexity. While it speeds up output, it also raises uncertainty, variability, and the chances for unintended behaviors. As a result, invisible work increases in both volume and significance. Validation becomes more challenging. Governance becomes more crucial. Cognitive load rises as systems get harder to fully grasp.
In this context, invisible work is no longer a secondary concern; it becomes the main limit on throughput, quality, and trust. Organizations that fail to see this shift risk optimizing for visible output while compromising system integrity.
Practical Strategies for Dealing with Invisible Work
To effectively address invisible work, organizations must first recognize its presence and then incorporate it into their operating models.
Measure What Was Previously Ignored
Track unplanned work, interruptions, and validation efforts. Without visibility, these activities will continue to distort capacity and planning.
Redesign Metrics
Move beyond output-based metrics like velocity. Incorporate measures of stability, flow efficiency, and rework. These give a clearer picture of system health.
Formalizing AI Governance
Set clear standards for validation, define acceptable use boundaries, and assign ownership for AI-generated artifacts. Governance must evolve alongside technology.
Reduce Cognitive Load by Design
Simplify architectures, reduce unnecessary dependencies, and prioritize clarity. Minimizing complexity reduces invisible work directly.
Treat Invisible Work as First-Class Work
If an activity is necessary for delivery, it must be planned, resourced, and governed properly. Ignoring invisible work does not eliminate it; it just shifts the burden elsewhere.
The New Architecture Imperative
Industry is entering a new phase where the challenge is no longer just building systems, but managing their behavior in increasingly dynamic and automated environments.
Invisible work is the system behind the system. It enables delivery to function in practice, not just in theory. Organizations that fail to recognize it will continue to optimize for speed while sacrificing quality, resilience, and trust. Those that embrace it will build systems that not only deliver faster but also endure longer. In the age of AI, the real bottleneck is no longer what we can see; it is what we choose not to.