Part 2 of the Agentic Software Factory series · Read Part 1 →
My last article described a factory that builds software — a Manager Agent orchestrating a roster of specialists, humans at the control points, governance in the pipeline. But I left something out, and a few of you caught it.
A factory that only builds is half a factory.
Software that ships is not finished. It has to run. And in most organizations, the moment code reaches production, the agents go quiet and the humans take back over. Pagers. Dashboards. War rooms at two in the morning. We automated the building and left the running to exhaustion.
The half we keep forgetting
We have spent a decade calling this DevOps, but the name gives away the seam. It is still two words bolted together. Build on one side, run on the other, connected by a handoff and a hope.
The Agentic Software Factory stopped at deploy. The DevOps Agent shipped the release and the loop went cold. Everything after that — the incident, the rollback, the root cause, the fix — landed back on people.
AIOps, before and after agents
AIOps is the name for applying AI to operations: watching systems, spotting anomalies, finding root cause, sometimes remediating. Most of what carries the label today is analytics bolted onto monitoring. Useful dashboards. Smarter alerts. But fundamentally, it is still a tool waiting for a human to act.
The agentic shift changes operations the same way it changed development. Not better dashboards. Agents that act.
The operations roster
If the Factory has a build roster, operations has its own:
- Observability Agent — watches the whole system, separates signal from noise, and knows what normal looks like.
- Incident Agent — diagnoses, correlates, and proposes or executes remediation within policy.
- Reliability Agent — tracks error budgets, surfaces fragility, and pushes hardening work upstream.
- Capacity Agent — manages scale and cost, anticipating load rather than reacting to it.
Each maps to a function we currently staff with tired people and runbooks. It is the same routing discipline we apply to models themselves — matching each task to the tool built for it.
The move that matters: closing the loop
Here is where it stops being faster operations and becomes something new.
The operations agents do not fix and forget. They report to the Manager Agent. A recurring production fault is not just remediated — it becomes a new work package routed back into the build pipeline. The thing that broke at two in the morning becomes a requirement by the morning standup, already in flight.
Build and run stop being two phases. They become one loop.
Build and run as one continuous loop, coordinated by the Manager Agent.
The flywheel is the whole point
Once the loop closes, something compounds. Every incident is training data. Every remediation teaches the build agents what fragility looks like before it ships. The system that operates your software makes the system that builds it smarter — and it does this continuously.
One number for the whole lifecycle
In the first article, Enterprise Workflow Yield (EWY) — a single figure for how much governed value the factory actually delivers — measured the build. Closing the loop lets it measure everything.
Operational yield — reliability, time to recovery, change failure rate — joins build yield under one figure. The Manager Agent stops optimizing half a lifecycle and starts optimizing the whole one. You can finally ask one question: how much governed value is this system delivering, end to end?
EWY across the loop: when build yield and operational yield sit under one metric, the Manager Agent can make trade-offs no human dashboard ever could.
A word of caution
There is a catch, and it is a serious one. The moment an agent rolls back a payment service on its own at two in the morning, the comfortable questions get sharp. Who approved that? What was the agent allowed to do without a human? Who answers for it if it was wrong?
That is not a footnote — it is a governance question, and it is the subject of the next piece in this series.
Where this leaves us
The Factory that only builds is impressive in a demo. The Factory that builds, runs, and learns from running is a business that gets harder to beat every quarter.
At Blue Avanti, this is the work: helping organizations close the loop deliberately, with the orchestration and the metrics to run it as one system rather than two. If your build side and your run side still meet at a handoff, that seam is where the opportunity is.
What does the loop look like in your organization, and where does it break? I would like to hear.