The binding constraint is attention
With one operator and several products, the scarce resource is not compute or even time — it is attention, and specifically the number of things that can be actively watched. A system needing routine human checking consumes a fixed share of a small budget permanently.
This inverts some standard advice. Engineering that would be over-investment for a team with an on-call rotation is straightforwardly correct when there is no rotation.
Prefer boring and self-healing
The correct default is the option with fewer moving parts, even at some cost in elegance or efficiency. Managed over self-hosted. Fewer services over more. Retries and graceful degradation over alerting, because an alert is only useful if someone can act on it, and at 3am nobody can.
Systems that fail into a degraded-but-functioning state are worth a great deal more than systems that fail loudly and correctly, when the audience for the loud correct failure is asleep.
Cost ceilings are not optional
With several products running, a runaway loop in one is a genuine financial event before anybody notices. Hard budget ceilings with graceful degradation are not a nice-to-have; they are the mechanism that makes it safe to not be watching.
The same is true of anything with unbounded retries or unbounded context growth. If it cannot be bounded, it should not run unattended.
Standardise ruthlessly across products
The compounding advantage of a portfolio is reuse — but only where things are actually the same. The same deployment approach, the same telemetry shape, the same eval harness structure, the same cost-attribution scheme. Each product that deviates costs a permanent context-switch tax on every future visit.
The reverse is also true: consistency means a fix or an improvement made once can be applied everywhere, which is the closest a solo operator gets to leverage.
Why this makes for better client work
Operating your own systems produces a specific kind of knowledge: what actually breaks, what it costs when it does, and which engineering earns its keep. It is a different kind of understanding from having advised on architectures that other people then have to maintain.
It also enforces honesty. When you carry the operational burden of your own recommendations, you stop recommending things that are impressive to build and unpleasant to run.