Peter Pedross, CEO & Founder, PEDCO – SAFe Fellow
Agentic Engineering fundamentally changes the role of governance. What traditionally existed as policies, process descriptions and quality manuals increasingly becomes executable engineering behavior. Rules become machine-readable, evidence is generated continuously and governance evolves from approving work to enabling better decisions. Skills, rules and hooks must be derived from the organization’s process model and aligned with its Lean Quality Management System and overall way of working, ensuring that humans and autonomous agents operate according to the same principles, guardrails and decision boundaries. In the age of autonomous systems, the central challenge is no longer controlling execution, but creating the decision environments in which teams and AI systems can act quickly, confidently and responsibly.
Every engineering organization has governance. The difference lies in where that governance exists.
Traditionally, governance has been documented in policies, process descriptions, quality manuals and architectural guidelines. These documents describe how software should be developed, which reviews are required, which quality standards apply and which regulatory obligations must be fulfilled. For decades this approach was entirely reasonable because people interpreted these documents. Engineers read the guidelines -> Architects reviewed the designs -> Quality managers verified compliance. The effectiveness of governance therefore depended largely on human experience and discipline.
Agentic Engineering changes this relationship fundamentally. Autonomous agents cannot “interpret” organizational intent in the same way experienced engineers do. They require governance that is explicit, structured and increasingly executable. This represents a profound shift. Governance is no longer something that exists outside the engineering process. It becomes part of the engineering system itself.
- Standards improve our own processes
- Processes and Policies evolve into rules.
- Rules become machine-readable.
- Rules activate skills.
- Skills invoke tools.
- Tools generate evidence.
- Evidence continuously validates whether engineering decisions remain aligned with organizational objectives.
In other words, governance moves from documentation into execution. This transition can already be observed across modern engineering organizations.
- Security policies automatically trigger vulnerability scans before software is deployed.
- Architectural constraints prevent unauthorized dependencies from entering the codebase.
- Quality gates validate test coverage before changes are merged.
- Infrastructure policies ensure cloud resources comply with organizational standards.
One of the most significant shifts may occur in the role of governance itself. For many years, governance in large engineering organizations primarily focused on approval. The central questions were often operational in nature.
- Do we have sufficient capacity?
- Do we have budget?
- Have the necessary stakeholders approved the initiative?
- Can we realistically deliver it?
Agentic Engineering changes this balance fundamentally. As implementation becomes increasingly inexpensive and engineering throughput increases dramatically, the limiting factor is no longer execution capacity. The new question becomes considerably more difficult:
Does it actually make sense for us to build this?
Not only from a business perspective. But from a technical, economic, architectural and organizational perspective. This changes the role of leadership. The central task is no longer controlling execution. It becomes the creation of clear decision-making environments within which teams can act quickly, confidently and responsibly. In other words, governance gradually shifts from approval governance toward decision-making governance. This transition has several practical consequences.
Human review gates remain essential, even in highly autonomous engineering organizations. Artificial intelligence can accelerate analysis, implementation and validation, but some decisions continue to benefit from human judgement. Architectural trade-offs, security implications, regulatory considerations, scalability concerns and long-term product strategy remain fundamentally human responsibilities.
Equally important are explicit decision guardrails. Teams and autonomous systems need to understand which decisions they can make independently and which require additional review by architecture, security, compliance or portfolio functions. Interestingly, organizations that make these boundaries explicit often experience both higher autonomy and stronger governance at the same time.
The concept of readiness also begins to evolve. A traditional Definition of Ready often focused on implementation prerequisites. For AI-assisted development, organizations increasingly require something closer to a Definition of Ready for decisions themselves.
- What problem are we trying to solve?
- What value hypothesis are we testing?
- Which risks are already known?
- Which assumptions are we making about architecture, data or scalability?
- Under which conditions would we stop investing further effort?
Similarly, the Solution Intent evolves once again. Rather than becoming a heavier governance document, it becomes a lighter and more dynamic decision artifact.
- Why are we building this capability?
- Which constraints apply?
- Which assumptions are considered stable?
- What outcomes are we actually trying to achieve?
Perhaps one of the more interesting consequences concerns evidence itself. Traditionally, evidence was created after implementation had finished. Increasingly, evidence becomes part of implementation. Tests, reviews, documentation, traceability information and compliance records emerge continuously as engineering work happens rather than being reconstructed afterwards.
The same observation applies to the Definition of Done. As autonomous systems assume a growing proportion of implementation work, organizations gain the opportunity to expand their quality expectations considerably. Documentation, traceability, test evidence, architectural validation and regulatory considerations can increasingly become part of everyday delivery rather than exceptional activities.
Finally, Lean Portfolio Management unexpectedly moves closer to the center of the engineering organization. When implementation becomes cheap, ideas become abundant. Portfolio management therefore becomes less about allocating scarce implementation capacity and more about ensuring that organizational attention remains focused on the initiatives that create the greatest value.
Perhaps this is one of the defining governance challenges of the Agentic Engineering era.
- Not deciding what can be built.
- Deciding what deserves to be built.
From Process Models to Executable Governance
Engineering governance is becoming operational. The same principle extends naturally into Agentic Engineering. Autonomous agents should not rely on static documentation that must be interpreted for every task. Instead, they should operate inside an environment where organizational knowledge is already embedded into their daily activities. Skills, rules and hooks do not emerge in isolation. They must be derived from the organization’s process model and aligned with its overall way of working. The same architectural principles, quality standards, approval mechanisms and decision boundaries that guide human engineers must also guide autonomous systems. This introduces a new requirement for engineering organizations. Processes can no longer exist solely as documentation intended for human interpretation. They increasingly need to become explicit, structured and machine-consumable representations of organizational behavior.
- Skills define what an agent is capable of doing.
- Rules define what an agent is allowed to do.
- Hooks determine when additional validation, human interaction or specialized workflows become necessary.
Together they form an engineering environment that continuously guides autonomous behavior without interrupting engineering flow. Importantly, this should not be misunderstood as replacing human decision-making. The objective is not to remove humans from engineering. The objective is to ensure that humans participate where they create the greatest value.
- Routine decisions can increasingly be delegated.
- Strategic decisions remain human.
This creates an engineering organization where responsibilities become clearer rather than less visible. One observation repeatedly emerged during our own engineering work. Whenever governance remained isolated inside documents, autonomous systems produced inconsistent results. Whenever governance became executable, engineering quality improved while development accelerated. The reason is surprisingly simple. Autonomous agents perform best when expectations are unambiguous. Explicit boundaries reduce uncertainty. Structured knowledge improves consistency. Continuous feedback enables learning. Engineering organizations, therefore, face a new challenge. The question is no longer whether governance exists. The question is whether governance actively participates in software development.
Organizations that continue treating governance as documentation will increasingly struggle to scale autonomous engineering.
Organizations that transform governance into an executable engineering capability will enable humans and autonomous agents to collaborate within the same trusted operating environment.
The next chapter builds upon this idea by introducing the concept of Context Engineering—the discipline that connects architecture, knowledge, governance and engineering artifacts into a shared decision-making environment for both humans and autonomous systems.
