Cost Estimation for Autonomous Delivery Projects

Software estimation has been the industry's biggest unsolved problem for decades. No cone of uncertainty, planning poker, or velocity-based projection has reliably predicted when software will be done or what it will cost. The fundamental issue is that traditional estimation tries to predict human effort, which is inherently variable and influenced by factors that cannot be quantified in advance.

From effort estimation to capability estimation

Autonomous delivery changes the estimation model. Instead of asking how long will it take a team to build this, the question becomes what is the scope of capability the system needs to deliver, and what are the governance constraints. The autonomous system's throughput is more predictable than human throughput because it is not affected by meetings, sick days, context switching, or motivation.

The new estimation variables

• Scope complexity: how many bounded contexts, service boundaries, and integration points
• Governance overhead: how many human review gates and approval cycles are required
• Constraint density: how many policies, compliance requirements, and architecture rules apply
• Integration depth: how many external systems must be connected and tested
• Novelty factor: how much of the system requires patterns the system has not generated before

The best estimate is a working increment delivered early enough to validate assumptions. Autonomous delivery makes the first working increment available in days rather than months, making estimation less critical and validation more frequent.