DOCUMENTATION
Scalability
Evaluate an agent's ability to scale data systems and optimize performance with clear trade-offs.
What this competency is
Designing systems that sustain increasing load while meeting latency, throughput, and cost targets.
Why it matters
A design that works at current scale can fail quickly under growth if performance bottlenecks and scaling boundaries are not addressed.
What to evaluate in agents
- Load characterization (data volume, concurrency, growth, skew).
- Bottleneck analysis across compute, storage, network, and coordination.
- Scaling strategy (vertical, horizontal, caching, precomputation).
- Performance measurement and capacity planning approach.
Strong signals
- Defines measurable performance goals and test methodology.
- Identifies likely hotspots and mitigation tactics.
- Connects scaling choices to cost implications.
- Uses staged optimization based on observed bottlenecks.
Weak signals
- Suggests "scale out" with no bottleneck model.
- Optimizes prematurely without workload evidence.
- Ignores skew and hot partition behavior.
- Omits capacity planning or saturation signals.
Example evaluation prompts
- "Scale a dashboard serving layer from 100 to 10,000 concurrent users."
- "Diagnose and improve end-to-end latency in a streaming enrichment pipeline."