Batch Size as the "Gravity" of Software Delivery: Why AI Systems Fail Like Human Teams
This article argues that batch size acts as a fundamental "gravity" in software delivery, causing AI agent swarms to fail in similar ways to human teams when tackling large, complex tasks. It posits that many perceived human-related failures in software projects are actually systemic issues stemming from large batch sizes and poor structural practices. The solution lies in adopting Continuous Delivery principles, emphasizing small batches, deployment automation, test automation, and robust monitoring.
Read original on The New StackThe "Gravity" of Batch Size in Software Delivery
The core premise of the article is that "batch size" is a fundamental law governing software delivery, much like gravity. Just as gravity affects all physical objects, batch size impacts the complexity and failure rate of software projects, irrespective of whether they are executed by human teams or multi-agent AI systems. Large batches increase coordination complexity and communication overhead, leading to failures that are often mistakenly attributed to human factors.
Key Takeaway: Batch Size & Complexity
The article highlights that increasing the batch size (the amount of work processed together) significantly increases the "gravitational pull" of complexity. This makes delivery harder, whether using human teams or AI agents, indicating that the challenges are systemic rather than purely human-centric.
Why Multi-Agent AI Fails in Familiar Ways
Experiments with multi-service backend systems built using AI agents reveal that coordinating multiple agents for complex tasks often outweighs the benefits of work division. The resulting failures mirror those of large human projects: communication misalignment, coordination overhead, and an inability to deliver working software reliably. This suggests that the architectural and process challenges of large-scale software development are inherent to the complexity itself, not just human fallibility.
Fixing the System of Work: Continuous Delivery Principles
Instead of blaming individuals or AI agents, the focus should be on fixing the underlying system of work. The article advocates for practices aligned with Continuous Delivery to reduce batch sizes and manage complexity effectively. These include:
- Deployment Automation: Makes releases repeatable and reliable, reducing wide-reaching failures and enabling more frequent, smaller deployments.
- Test Automation: Increases confidence in deployable software, catching bugs early and reducing embarrassing production issues.
- Monitoring and Alerting: Provides real-time insights into system health in production, allowing teams to proactively address problems and maintain system stability.
Implementing these practices reduces batch size, lowers complexity, and solves communication and coordination problems that become insurmountable in large work batches. This approach is effective regardless of who or what is writing the code, reinforcing that good engineering practices are universally applicable.