Agentic Workflows, Citizen Developers, and AI Governance
An executive analysis of the shift toward agentic engineering, the rise of citizen developers, and the critical need for governance frameworks. This brief explores model routing strategies, token economics, and the emerging 'software factory' paradigm in enterprise AI adoption.
The Rise of the Citizen Developer
The integration of agentic AI tools is fundamentally altering the software development lifecycle (SDLC) by extending coding capabilities beyond professional engineers. Non-technical roles, including product managers and sales staff, are now deploying functional applications using agentic workflows. This shift creates a critical operational gap: these 'citizen developers' lack the expertise to manage security, backups, and compliance. Consequently, enterprises face a surge in unmaintained internal applications, necessitating the development of opinionated 'citizen developer platforms.' These platforms must provide default guardrails, automated CI/CD, and secure deployment pathways to mitigate the risks of uncontrolled software proliferation.
Strategic Model Routing and Economics
Token economics have become a primary constraint for enterprise AI adoption. The era of using frontier models for all tasks is ending due to unsustainable costs. A strategic approach involves hierarchical model routing, where complex planning and research are handled by high-capability models, while routine execution and classification are delegated to smaller, cost-effective models. This optimization is crucial for achieving positive ROI, particularly in non-engineering departments where productivity metrics are less objective. Furthermore, the shift toward per-token billing in regulated industries and the adoption of inference gateways are driving the need for granular observability and data governance.
Governance as an Enabler
Contrary to traditional views, governance is no longer a barrier to innovation but a prerequisite for scaling. Organizations are moving from blanket bans to structured 'agentic marketplaces' that allow teams to share skills and MCP connectors within controlled environments. This framework enables grassroots experimentation while ensuring security audits and compliance. The focus is shifting from preventing usage to empowering safe, auditable innovation, allowing companies to leverage institutional data and expertise through agentic workflows without exposing themselves to significant security risks.
The Software Factory Paradigm
The industry is converging on the 'software factory' model, where agents handle well-defined, verifiable tasks such as library upgrades and bug fixes. These tasks have constrained solution spaces, making them ideal for high-autonomy agentic execution. However, complex feature development still requires human oversight to navigate ambiguous requirements. The future of engineering involves dynamic 'loop engineering,' where the level of human involvement is adjusted based on task complexity. To fully realize this potential, the industry must overcome the bottleneck of local development environments by adopting cloud-based, containerized infrastructure that supports parallel agentic execution.
Key insights
-
Non-technical employees are deploying unmanaged software via agentic tools, creating a new class of 'citizen developers' who lack security and maintenance expertise. This leads to a proliferation of unmaintained internal applications with significant compliance risks.
Impact: Enterprises face increased technical debt and security vulnerabilities unless they implement opinionated platforms that automate guardrails, authentication, and CI/CD for non-engineers.
-
Effective AI governance requires centralized marketplaces for skills and MCP connectors rather than blanket bans. This allows for grassroots innovation while maintaining auditable controls and preventing security issues like injection attacks.
Impact: Organizations that build robust governance frameworks can scale AI adoption faster by enabling safe experimentation and standardizing best practices across teams.
-
Using frontier models for all tasks is cost-inefficient. Strategic routing that assigns complex planning to high-end models and routine execution to smaller models significantly reduces token spend.
Impact: Implementing hierarchical model routing improves ROI and makes AI adoption viable for cost centers outside of engineering, such as finance and customer success.
-
Agentic 'software factories' are most effective for verifiable, low-complexity tasks like library upgrades and bug fixes. These constrained solution spaces allow for high autonomy, whereas complex features still require human-in-the-loop feedback.
Impact: Focusing agentic efforts on verifiable tasks accelerates delivery and reduces error rates, allowing engineers to focus on high-value architectural decisions.
-
Local development environments are a major bottleneck for agentic workflows. The transition to cloud-based, containerized environments is essential to support parallel task execution and reliable end-to-end testing.
Impact: Adopting cloud-native development environments removes local resource constraints, enabling true 'software factory' capabilities and scaling agentic throughput.
Action items
-
Develop opinionated 'citizen developer platforms' that provide default security, authentication, and CI/CD pipelines for non-technical staff using agentic tools. Ensure these platforms enforce data access controls and automated updates.
Impact: Mitigates security risks from unmanaged internal applications and enables safe, scalable AI adoption across non-engineering departments.
-
Implement a centralized 'agentic marketplace' for skills and MCP connectors. Establish a promotion pathway that allows teams to experiment safely and share audited, high-quality skills across the organization.
Impact: Reduces skill sprawl and reinvention of the wheel while maintaining governance and security standards for AI workflows.
-
Deploy an inference gateway to manage token routing, observability, and data retention. Configure rules to route complex tasks to frontier models and routine tasks to smaller, cost-effective models.
Impact: Optimizes token spend, ensures compliance with data retention policies, and provides granular visibility into AI usage for ROI analysis.
-
Identify and automate verifiable, low-complexity tasks such as library upgrades, dependency management, and bug fixes using agentic 'software factories.' Define clear success criteria and verification steps for these tasks.
Impact: Accelerates routine engineering work and frees up senior engineers for high-value architectural and strategic tasks.
-
Transition development environments to cloud-based, containerized infrastructure. Ensure that end-to-end test suites can be executed from a clean Git checkout in a container to support parallel agentic execution.
Impact: Removes local resource bottlenecks and enables scalable, parallel agentic workflows, paving the way for full 'software factory' implementation.
Quotes
“I think that there's going to be a whole industry around how do we build the internal platform for these system developers?”
“The natural fit is like, I want to talk to a Fable or an Opus, depending on how big a story it is, depending on how big a feature it is.”
“We're getting a 10x boost in coding output lines of code or whatever but it's all being dissipated by either slamming into the wall up downstream or we're kind of pulling stories faster”