project-development

Run the task through the task-model fit tables: proceed if it involves synthesis, subjective judgment with rubrics, natural language output, and error tolerance; stop if it requires precise computation, real-time responses, or perfect accuracy. Always validate with a manual prototype by testing one representative input in the model interface before building automation.

Use the five-stage structure: acquire, prepare, process, parse, render. Keep the LLM call isolated in the process stage, store intermediate outputs as files per item for idempotency, and parallelize the process stage with ThreadPoolExecutor. The included pipeline_template.py script implements this pattern.

Default to single-agent pipelines for batch processing with independent items. Escalate to multi-agent only when parallel exploration is required, the task exceeds a single context window, or specialized sub-agents demonstrably improve benchmark quality. Choose multi-agent for context isolation, not role anthropomorphization.

Multiply item count by tokens per item by price per token, then add a 20-30% buffer for retries and failures. The pipeline template includes an estimate stage that counts prompt characters, approximates tokens, and projects cost in dollars before you commit to the expensive process stage.

LLMs do not follow format instructions perfectly, so brittle parsers break on minor variations. Design prompts with explicit section markers, format examples, and rationale disclosure, then build parsers with flexible regex, sensible defaults for missing sections, and error logging instead of crashes.

Not necessarily. The Vercel d0 case study showed reducing from 17 tools to 2 raised success from 80% to 100% while cutting execution time 3.5x. Test whether scaffolding helps or constrains the model, and add complexity only when production evidence proves it necessary.