Ten distinct tasks · Performance and operations
Measure and reduce wasted work
Payloads, caching, polling, loading, observability and task-level usage. Choose a specific task for its worked example, adaptable agent brief and acceptance checks.
10 guides found
Reduce database egress in an AI app: measure payloads and repeated reads
Measure outgoing payload volume and request frequency by operation, then reduce unnecessary fields, rows and repeated reads. A worked example, agent brief, failure checks and acceptance evidence.
Cache a public API response in an AI app: freshness, variants and invalidation
Cache only data whose visibility and freshness contract permit reuse. A worked example, agent brief, failure checks and acceptance evidence.
Improve Core Web Vitals in an AI website: diagnose the actual page experience
Measure loading, interaction and layout stability on representative pages, then repair the largest supported bottleneck. A worked example, agent brief, failure checks and acceptance evidence.
Optimize images in an AI-generated website: dimensions, formats and useful detail
Serve images at appropriate dimensions and quality for their actual display role. A worked example, agent brief, failure checks and acceptance evidence.
Reduce a JavaScript bundle in an AI app: active dependencies and deferred features
Inspect what enters the client bundle and when it is needed. A worked example, agent brief, failure checks and acceptance evidence.
Reduce database query payloads in an AI app: select the facts the screen needs
Define the screen’s data contract and request only needed rows and fields. A worked example, agent brief, failure checks and acceptance evidence.
Fix excessive polling in an AI app: visibility, backoff and shared results
Identify every polling owner, choose a freshness interval that fits the task and stop or reduce work when it is not useful. A worked example, agent brief, failure checks and acceptance evidence.
Show progress for AI runtime jobs: confirmed stages instead of invented percentages
Expose the job’s actual stage, elapsed time and last confirmed event. A worked example, agent brief, failure checks and acceptance evidence.
Monitor errors in an AI-built application: actionable signals and bounded noise
Collect meaningful client and server failures with revision and request context, then group recurring signatures. A worked example, agent brief, failure checks and acceptance evidence.
Measure AI agent cost per task: accepted work, retries and verification
Track the actual model and runtime work associated with a task, including retries and verification. A worked example, agent brief, failure checks and acceptance evidence.