"Should we deprecate our REST API in favor of GraphQL?"
4 models · 1 round · reconciled into one memo.
Keep REST as the system of record. Add GraphQL only where clients demonstrably suffer from over-fetching. A hard deprecation forces every partner to migrate on your timeline, which is where these projects stall.
A hybrid layer is the pragmatic path. GraphQL's flexibility is real for mobile, but it shifts complexity into resolvers and caching. Do not deprecate REST until GraphQL has carried production traffic for at least two quarters.
The migration cost is dominated by integration rewrites and observability, not the API surface itself. Run both; freeze new REST endpoints once GraphQL covers the core schema, but leave existing REST intact indefinitely.
GraphQL reduces round-trips for the mobile client, which is the strongest concrete win here. But a full deprecation is premature without usage data on which REST endpoints partners actually depend on.
Estimated by the admin model from conceptual analysis; cross-checked against vocabulary overlap.
The panel converged 4-of-4 against full deprecation and toward an additive hybrid layer, citing migration cost (integration rewrites, observability) as exceeding platform savings. Mobile over-fetching is the clearest case for GraphQL.
REST remains the partner contract
Hybrid layer is lower-risk than a full cutover
Pace of any eventual REST sunset
Models diverged on whether to freeze new REST endpoints once GraphQL ships.
Run GraphQL alongside REST; do not deprecate until it has carried core production traffic for two quarters.
- GraphQL covers the core schema before any REST freeze
- Partner usage data gathered on REST endpoints before deprecation talk
One panelist holds that maintaining both layers doubles surface area and the hybrid period should be hard-capped.
Recommendation: do not deprecate REST. Introduce GraphQL as an additive layer for the specific clients that benefit (mobile, partner integrations), and keep REST as the stable default. The panel found that a full migration's cost — rewriting integrations, retraining the team, and rebuilding observability — exceeds the platform savings within any reasonable horizon.
Where the panel converged: a hybrid layer is lower-risk than a cutover; GraphQL helps over-fetching on mobile; REST should remain the contract for existing partners.
Open disagreements you should weigh: the timeline (18 vs 24 months vs indefinite), and whether to freeze new REST endpoints once GraphQL lands.