Write the problem hypothesis
For [user], the agent loses or retrieves [information] incorrectly across [scope], causing [wrong action or repeated work].
You can name one user, one situation and one measurable consequence without proposing a feature.Hard-coded `new InMemoryRunner(...)` prevents injection and custom runner implementations
Evidence observed in google/adk-java, a Agent Infrastructure project.
6 comments5 positive reactions328 days openProject Radar 89
needs review
Start the no-account validation tool4 guided steps · private browser notes · export anytime
REPEATED ACROSS 6 PROJECTS
Related friction appears in 6 independent repositories. This is stronger than one backlog item, but still requires direct user validation.
Project strength and demand confidence are measured separately.SOURCE EVIDENCE
Read original GitHub Issue ↗Start with what users actually said
Reporter context: In the current implementation of , the runner instance is created directly via constructor: This introduces a structural limitation: the code forces a specific runner implementation and does not allow the caller to provide their own or override behaviour. Why this is problematic No ability to inject alternative Runner…Excerpted from the public Issue. Read the complete thread before interpreting it.
Recruit: Recruit teams with live traces, tool calls or deployment constraints from a real agent workflow.
Guardrail: Test observable state transitions, permission boundaries and recovery from partial failure.
Interview five agent-platform engineers or technical operators
- Which information must persist?
- For how long and within what scope?
- Show the last wrong or missing recall.
- What data must never cross boundaries?
- How do you correct memory today?
Run the smallest experiment
Test one constrained memory scope with a small real dataset and a written expected-recall benchmark.
The system retrieves the right evidence across repeated sessions without leaking or inventing context.Test observable state transitions, permission boundaries and recovery from partial failure.Make a build decision
- Build: repeated pain and active commitment
- Narrow: pain is real but the audience or job differs
- Stop: weak frequency or no behavioral proof
Why this brief exists
Information has value only when it changes action. This page turns one public signal into a bounded validation exercise. It is a research aid, not proof of demand, investment advice or a product recommendation.