Where Architects Sit in the Era of AI

This article was written by participants of the online InfoQ Certified Architect Program. It represents the capstone of their work, reflecting the cohort’s collective learnings on the intersection of AI and modern software architecture.

Introduction

The transformation driven by Artificial Intelligence (AI) brings extraordinary potential but also a profound question: What does it mean to be an architect when architectural thinking can be automated?

Since the dawn of technology, the role of the architect has been an endeavour of human craft. An important role in any organisation that holds both a deep understanding of business and technology.

Through this dual understanding, new systems are born to do wonderful things as the architect balances the scales of complexity, reliability, security, and purpose. Architectural thinking has always been there to bridge the divide between business and technology with intimate knowledge of domains and context.

But with the arrival of AI, this thinking is shifting. AI is no longer just a component within a larger architecture, or just a predictive model, categoriser, or summariser within the wider ecosystem.

AI is now an actor, a collaborator within the business. Architects can now share the space with machines that can simulate trade-offs, generate code, detect risks, propose solutions, and push it all into production without hesitation.

As AI systems gain agency, architects must decide how much control to retain, when to supervise, and when to delegate.

The future of architecture will not be defined by whether humans or AI make decisions, but by how they collaborate. The three loops of being in, on or out, describe the evolving relationship between human architects and intelligent assistants throughout the design lifecycle.

How should we, therefore, reflect on the role of an Architect? We have written this article to explore this problem space in more detail.

The three loops – In, On, and Out

In the emerging AI-augmented ecosystem, we can think of three modes of architect involvement: Architect in the loop, Architect on the loop, and Architect out of the loop. Each reflects a different level of engagement, oversight, and trust between an Architect and intelligent systems.

Architect in the loop (AITL)

What does it mean to be in the loop? In the Architect in the Loop (AITL) model, the architect and the AI system work side by side. AI provides options, generates designs, or analyzes trade-offs, but humans remain the decision-makers. Every output is reviewed, contextualized, and approved by an architect who understands both the technical and organizational context. This is where the Architect is sat in the middle of AI interactions with AI doing the majority of the tasks and then coming back to ask for permission, guidance, and advice as required (which won’t be required in every situation).

In this loop, agency remains human, but capacity is augmented. It represents the sweet spot for most organizations adopting AI responsibly: a balance between efficiency and accountability.

Architect on the loop (AOTL)

What does it mean to be on the loop? As AI matures, parts of architectural decision-making can be safely delegated. In the Architect on the Loop (AOTL) model, the AI operates autonomously within predefined boundaries, while the architect supervises, reviews, and intervenes when necessary.

This is where the architect is firmly embedded into the development workflow using AI to augment and enhance their own natural abilities. Consider a “bionic” Architect with the natural symbiosis between AI and Architectural-Human thinking. In AOTL, the Architect is using AI to investigate problem spaces, diverge, and converge while considering a plethora of different ideas, while rapidly testing different approaches. However, the key difference here between being in and on the loop is that being on the loop means all actions are triggered and managed by the Architect themselves, placing them squarely in the middle of all activity and thinking. Governance shifts from making decisions to designing how decisions are made. The Architect becomes a systems governor, ensuring that AI operates ethically, safely, and in alignment with organizational principles. In AOTL, the architect acts as steward, defining the boundaries of intelligent action.

Architect out of the loop (AOOTL)

What does it mean to be out of the loop? In the AOOTL model, we see a world where the architect is no longer required in the traditional fashion. The architectural work of domain understanding, context providing, and design thinking is simply all done by AI, with the outputs of AI being used by managers, developers, and others to build the right systems at the right time. In this model, the architect’s role shifts to meta-design, where you are designing the system’s ability to design itself. The architect will define the rules of self-adaptation, the feedback loops, and the thresholds for human escalation. The key becomes not control, but containment, where you ensure that even when humans are out of the loop, the system remains aligned with human intent.

This mode carries the greatest efficiency but also the greatest ethical and operational risk. Architects must ensure explainability, auditability, safe rollback, and accountability mechanisms for every autonomous decision. In the AOOTL model, we see the architect as a guardian, trusting the system’s intelligence while being accountable for its consequences.

How to choose the right loop?

No organization should aspire to remove humans completely from every loop due to the probabilistic nature of current AI solutions. Many core business areas require determinism in decision making such as financial planning, supply chain operations, workforce planning or manufacturing operations. Probabilistic problems range from forecasting, demand planning, risk analysis and behaviour modelling. We can’t have differing results day-to-day when trying to plan a supply chain! Also in looking to remove humans from loops would be to misunderstand the role of humans in providing novel and unique solutions to never-before-seen problems. Instead, the goal is to orchestrate the loops dynamically based on where the loops are deployed.

The art of architecture in the AI-era is knowing where each loop begins and ends and building systems that make those boundaries explicit and adjustable. When applied thoughtfully, the three-loop model preserves the architect’s essence not as a controller of systems, but as a designer of intelligences, human and artificial alike.

How AI extends the Architect’s reach

Artificial intelligence is evolving from a coding aid into a decision co-processor for architectural design. Where architects once worked with static diagrams and spreadsheets, they can now enlist models that search, simulate, and score design options in real time. The result is not automation for its own sake but an expansion of analytical reach, an architect’s intuition amplified by computational scale.

In early design phases, generative models help architects absorb and synthesize knowledge faster than any human research cycle. When faced with familiar challenges, for example, integrating legacy systems into a zero-trust network. AI can summarize documentation, compare integration specs, and flag security gaps that typically surface only during late reviews. Equipped with that insight, architects can rapidly prototype topologies, evaluate trade-offs between on-premises and cloud options, and even interrogate the model itself to understand how AWS, Azure, or GCP differ in latency, resilience, or compliance guarantees.

Recent frameworks such as Smart HPA (2024) and AHPA – Adaptive Horizontal Pod Autoscaling (Alibaba Cloud, 2023) apply machine learning to balance resource efficiency and reliability across microservices. Before deployment, architects can prototype and validate scaling strategies using simulation frameworks such as CloudSim Plus, which model workload patterns, latency, and cost dynamics across distributed systems. In production, cloud-native optimizers like AWS Compute Optimizer and Azure Advisor extend similar reasoning to live environments. The creative act shifts from drawing boxes to defining policies and objective functions: once telemetry, scaling logic, and design intent connect through continuous feedback, architecture becomes adaptive. Real systems such as Netflix’s Dynamic Optimization for HDR Streaming and Uber’s Michelangelo Platform show how AI agents tune workloads within human-defined guardrails, enabling architects to orchestrate constraints rather than configurations while ensuring autonomy remains transparent, explainable, and trustworthy.

Knowledge-graph reasoning adds a structural dimension to architectural practice. Platforms such as Atlassian Compass and Spotify Backstage already map service dependencies. When exported into graph embeddings and queried through engines like Neo4j GraphRAG or Amazon Neptune ML, undocumented tribal knowledge becomes searchable, contextual intelligence. Instead of asking a senior engineer, “Which upstream dependency might expose PII through an indirect API flow?”, the architect can now ask the model. AI surfaces correlation; humans supply causation.

Large-language models now regenerate “as-built” diagrams by parsing infrastructure-as-code and CI/CD pipelines. Tools such as Ardoq, LeanIX AI Assistant, and AWS Well-Architected Analyzer enable what BOC Group (2025) calls adaptive documentation, a living reconciliation of design and deployment. Yet models infer structure, not intent. Best practice is dual-channel documentation: a machine-generated state layer paired with human-authored rationale. Research by Bucaioni et al. (2025) shows that mining architectural decision records can predict which patterns correlate with higher reliability or performance, allowing organizations to build self-learning design intelligence over time.

A new wave of co-design environments, including ArchAI, Codeium Architect, and Claude Architect (Anthropic Labs, 2025), combine LLMs, graph reasoning, and constraint solvers in unified workspaces. Architects describe goals in natural language and receive validated blueprints, infrastructure-as-code snippets, and risk annotations. Early pilots, documented by Esposito et al. (2025), show 40–60 percent faster design iteration, but only when prompts align with enterprise standards. AI handles pattern synthesis, while architects focus on translating business intent into executable design and ensuring that every trade-off aligns with organizational purpose.

AI extends, rather than replaces, the architect. It amplifies human creativity and foresight while demanding stronger ethics, communication, and transparency. As Publicis Sapient (2025) argues, those who pair statistical exploration with ethical judgment will not be out of the loop—they will design the loop itself. The architects who master this partnership will define the next era of digital design: adaptive, evidence-driven, and profoundly human in purpose.

For broader context, see Forrester’s Future of the Enterprise Architect’s Job (2025), which highlights how AI shifts architects from document authors to decision engineers, and O’Reilly Radar (2024): Software Architecture in an AI World, emphasizing human-in-the-loop reasoning as a defining skill in AI-augmented design.

Challenges and Risks

Every extension of reach expands the surface of exposure. Artificial Intelligence brings extraordinary leverage to the software architect’s work (e.g., automating design exploration, validation, and documentation), but it also introduces new fragilities. To use AI responsibly, architects must confront not only technical and ethical risks but also deep human ones such as the erosion of judgment, the loss of tacit knowledge, and the diffusion of accountability.

Over-Reliance and Skill Atrophy

The most insidious risk is competency decay. As AI tools propose architectures, simulate outcomes, and generate documentation, architects may slowly lose the craft that once came from working through design problems manually. The automation paradox is well documented: as systems become more automated, humans become less skilled at manual operation, yet when automation fails, those very skills are most needed. Retaining these skills will be a challenge but is mandatory work as AI helps to augment our skills not replace them.

Architects who always depend on AI for design or documentation may struggle when asked to whiteboard a solution or explain a decision without the model’s assistance. This erosion of intuition weakens architectural resilience and professional growth.

Mitigation requires deliberate friction. Teams should hold AI-free design sessions, encouraging discussion and trade-off reasoning by hand. Code and design reviews should treat AI output as suggestions, not gospel. Junior architects must still learn the fundamentals before leaning on AI for acceleration.

Hallucination and the Appearance of Certainty

Generative AI often can produce confidently incorrect responses, architectures that sound elegant but collapse under scrutiny. It might recommend a seemingly correct design that isn’t fit for purpose or doesn’t answer to the requirements. Its fluency can mislead reviewers into complacency. AI cannot replace human judgment from those living and breathing the context and understanding the intricacies of it.

GenAI offers probabilities, not truth. Every suggestion must be treated as a hypothesis to validate, not a specification to implement.

Organizations should foster a verification culture:

• cross-checking outputs with multiple models,
• tracing AI suggestions to their sources, and
• rewarding teams that question confident results rather than accept them at face value.

The goal is calibrated trust: knowing when to delegate and when to doubt.

Loss of Tacit Knowledge and Contextual Wisdom

Architecture is more than structure; it is context made tangible. The choice between monolith and microservices may hinge on communication patterns, politics, or talent availability, none of which an AI can infer.

As AI intermediates design and documentation, tacit knowledge (i.e., the unrecorded reasoning, stories, and trade-offs accumulated through experience) fades. The next generation may inherit technically consistent but contextually hollow designs.

Preserving this craft requires deliberate human practice:

• record the reasoning behind decisions,
• foster mentorship to interpret AI insights collaboratively,
• document rejected options and context to keep institutional memory alive, and
• holding architecture forums to discuss proposals from AI.

Just as over-reliance on GPS dulls our sense of direction, over-reliance on design copilots dulls architectural intuition. Tacit knowledge is collective memory, once lost, it cannot be regenerated from model weights.

Bias Amplification and Ethical Debt

AI systems mirror the data they learn from. When trained on historical architectural patterns, they inherit past biases (e.g., over-engineering tendencies, favored vendors, or architecture styles). Over time, this can produce ethical debt, the invisible accumulation of bias, unfairness, opacity, and misaligned incentives.

A model optimizing purely for cost or latency might inadvertently breach data-residency rules or degrade accessibility. If everyone uses similar AI tools trained on the same data, architectural diversity collapses, creating systemic fragility.

Ethical debt, like technical debt, is best paid early. Architects must embed ethical observability into design. Organizations should source models from diverse origins and explicitly question whether a recommendation fits context or merely “feels right” because the AI said so.

Accountability and Responsibility Diffusion

As AI participates in architectural decisions, responsibility becomes blurred. If an AI suggests a flawed design and the architect approves it, who is accountable: the human, the tool, or the organization that mandated its use?

This ambiguity risks professional erosion. “The AI recommended it” can become an excuse, weakening care and discipline.

Clear governance must restore accountability:

• Decision records should state explicitly where human judgment occurred
• Architects remain fully answerable for outcomes, regardless of AI involvement
• Tools should expose their reasoning for audit and learning, not conceal it behind APIs

AI changes the workflow, not the professional duty. Using AI is a tool choice, not an accountability shield.

Conclusion

As AI grows more capable, the question isn’t whether machines will design systems, but whether humans will still guide why those systems exist and where humans will be able to add most value. In the current era of generative AI, you will struggle to problem-solve in novel new design spaces that an AI has not seen in its training data with AI alone, but when human and AI come together in a collaborative manner, great things can take root. The future of architecture will be a continuum of loops with architects in, on, and out, blending collaboration, oversight, and delegation. What defines great architects will not be their ability to draw diagrams faster, but to design governance structures where intelligence operates safely and purposefully.

Architects will increasingly act as:

• Collaborators: co-designing with AI in the loop
• Stewards: defining policies and ethical boundaries on the loop
• Guardians: ensuring alignment and accountability when out of the loop

AI will extend the architect’s reach, but it also amplifies their responsibility.

The systems we design today will soon be designing themselves, and the architect’s legacy will lie in whether those systems remain trustworthy, transparent, and aligned with human values. The “bionic” architect, augmented by AI, is not the one who yields control to machines, but the one who designs the loops that keep humanity inside the system by choice, by design, and by purpose.

Some argue that as AI matures, the architect role itself may dissolve, just as it once emerged from developer practice. If machines can design, simulate, and govern systems autonomously, what is left for the architect to do?

Yet history suggests the opposite: every wave of automation elevates the human role from maker to steward.

The future architect may draw fewer diagrams but will design more meaning, deciding when to let the machine think and when to think for the machine. AI will not eliminate architectural accountability; it will redistribute it. The architects who thrive will be those who balance automation with agency, efficiency with transparency, and innovation with wisdom.

In the end, AI’s greatest gift is not speed or scale, it is the opportunity to redefine what it means to design responsibly.

The architect’s primary skill remains judgment, not generation.

References

Glossary

• AI – Artificial Intelligence
• AITL – Architect in the loop
• AOTL – Architect on the loop
• AOOTL – Architect out of the loop