We run production services with monitoring, SLOs, alerting, and postmortems. We do all that to watch the system, improve it, and act quickly with a short feedback loop. With coding agents, we review the final output and monitor token usage and spending, but rarely measure whether new context, skills, or tools improve the output. We almost never review the process, and how can you improve the process if you don't measure?
This talk makes an attempt at introducing a method for monitoring and improving ADLC, making the process visible by scoring it end to end: intent clarity, context quality, local tool calls, CI runs, PR outcome, and how the change survived production. Each failure signal maps to a treatment (context, skills, tools, guardrails), giving engineers a way to improve ADLC with the same rigor they apply to production systems.
This talk has been presented at AI Coding Summit Asia 2027, check out the latest edition of this Tech Conference.
