Providing a safety net

The non-determinism of LLMs, on which AI agents are built, necessitates foundational governance. This is best implemented through policy-as-code rules. A core risk emerges when AI not only handles individual tasks but also orchestrates the entire workflow: the established process, which serves as a harness for ensuring quality, could be at risk. Automated non-AI rules must ensure that minimum guardrails are met, balancing the flexibility of an agent-orchestrated workflow with the reliability needed. 

A critical component of the agentic architecture involves specialized AI agents dedicated to validation and compliance. They are set up with a quality assurance goal and aim to challenge the work of other agents.

Ensuring trust and accountability

For effective governance and continuous improvement, agents must generate audit records logging their performance and success. Clear accountability for the results of agents lies with the human team and hence they must be able to monitor and make corrective actions.

A central trust register could be maintained for all agents. This register would provide access to the most reliable agents or transparently indicate the level of risk associated with deploying one.

Reliability is monitored through metrics such as the agent task completion rate. These metrics are essential for detecting and acting on behavioral drift, which must be actively managed.

AI FinOps is another critical component given the substantial costs of running LLM-powered agents. The organization will need a robust model to accurately track these costs and inform decision-making, particularly concerning the return on investment and potential workforce implications.

The culture change: The most neglected shift

The shift to an agentic SDLC is fundamentally a cultural transformation. This is, unfortunately, an area that’s often overlooked in earlier change initiatives. There’s a danger this new transformation will suffer the same fate.

This transformation rests on a foundation of strong software engineering excellence. The integration of agents requires embedding them into the team's culture, establishing a feedback loop where humans and agents give and receive input, which will inevitably modify human behavior. To accommodate the necessary pace, faster decision-making is required, making close cross-functional collaboration and mob programming essential practices. Engineers will see their roles evolve from creators to governors, a transition that may prove challenging for some individuals.

The shift to an agentic software development lifecycle (SDLC) necessitates a change in engineering mindset and team culture.

Shifting the engineering mindset

Professional value moves from syntax mastery to problem definition. The ideal developer transforms from the "hero developer" to the "system thinker". Engineers' primary responsibility evolves from being the main code authors to becoming the architects and auditors of the system.

Cultivating a new team culture

Based on the changes to the operating model described above and the nature of AI agents, team culture is affected in the following ways:

  • Healthy skepticism must be cultivated: teams need to trust agents sufficiently while remaining critical with their outputs. 

  • Silos must be reduced in favor of collaborative practices like mob programming and swarming.

  • Leaders need to cultivate a culture where errors are embraced, treating unsuccessful AI experiments as valuable learning opportunities.

Although culture change cannot be prescribed, anticipating the target picture helps to design the SDLC, so it’s not contradicting.

Conclusion

The transition to an Agentic SDLC is not merely a technological upgrade but a holistic organizational journey that touches every facet of an organization: architecture, talent, team structure, governance, and culture. Success therein depends on an integrated strategy that addresses all these dimensions in parallel - that’s applied system thinking.

By embedding autonomous AI agents as essential contributors, organizations can manage increasing complexity and achieve unprecedented speed. This requires a proactive talent transformation, evolving engineers from coders to architects and auditors, supported by a critical capability building strategy. Concurrently, team topologies must adapt toward smaller, more focused stream-aligned teams. To prevent chaos from non-deterministic agents, robust governance is mandatory. Most critically, the shift demands a profound culture change.

The Agentic SDLC is the future of software development. Organizations that embrace this integrated approach - adapting their people, processes, and culture alongside their technology - will be best positioned to harness the full potential of this transformation.