Build vs Buy Decision

Forty-five minutes on a decision you have been going round on for weeks.

Vector database. Agent framework. Observability stack. Managed inference or self-hosted. Build the retrieval layer or buy it. Which model provider to standardise on, and what it costs to switch off it later.

Bring your shortlist and your real constraints: team size, latency budget, compliance, and whatever you have already committed to that we cannot undo. I will tell you which way I would go, what the decision actually turns on, and what it costs you to reverse if you pick wrong.

I have lived with most of these choices in production across AWS, Azure and GCP. Mostly that means I know which ones hurt eighteen months later.

You leave with the decision written down and the reasoning behind it, so you can defend it to your team without relitigating the whole thing.