An interactive tool to explore cost and capability tradeoffs of AI delivery methods. This includes a constraint-based decision wizard, TCO calculations, and side-by-side comparisons.

What problem or question does this address?

Every AI-powered feature requires a provisioning decision: how will inference happen? Existing references cover the landscape, but may not translate well to project scoping decisions. Pulling these components together in a single tool allows for a guided exploration of options and the flexibility to adjust assumptions as needed.

What does this experiment actually do?

  1. Wizard — A set of common app development constraint questions (budget, privacy, capability, latency, reliability) filter delivery methods into viable / caution / eliminated tiers.
  2. Calculator — TCO estimation with adjustable assumptions, Observable Plot charts showing cost breakdowns, volume curves, and per-request costs
  3. Compare — side-by-side table of selected methods across operational, control, capability, and risk dimensions
  4. Guide — A basic, narrative decision guide

What signals are we looking for?

  • Can someone unfamiliar with the provisioning landscape use the wizard to narrow down to a few viable options in jsut a couple minutes?
  • Does the TCO calculator surface non-obvious cost differences (e.g., operations costs dominating inference costs at low volume)?
  • Is the tool useful enough that teams reach for it when scoping a new AI-powered feature?

A clear failure would be if the constraint filtering is too coarse (everything stays "viable") or if the TCO model is so simplified that teams don't trust the numbers.

What are the boundaries?

  • Keep this focused on surfacing the right magnitude of TCO differences between delivery methods, not connected to real billing data.
  • Scoped to the decision framework only; doesn't cover implementation guidance