Ideas, field notes, and things worth sharing from the people building with AI.
AI ecosystem
Choose between retrieval, fine-tuning and a combined design by diagnosing the failure, inspecting available evidence and comparing operational responsibilities on the same evaluation cases.
Model Context Protocol
Design a support workflow that makes missing policy evidence visible, produces a useful bounded draft and routes unresolved questions through an authorized review path.
Model Context Protocol
Explore how an assistant can retain a useful proposal without turning it into an approved decision, using explicit status, authority, provenance and retrieval rules.
AI ecosystem
Design system prompts around explicit decisions, evidence boundaries and reviewable outputs, then test them with cases that expose ambiguity and unsupported assumptions.
AI ecosystem
Use an LLM to draft a focused weekly review checklist from documented requirements, recent changes and incident lessons, then have reviewers validate every item.
AI ecosystem
Design an expense-tracking pipeline that preserves original transactions, proposes categories, records corrections and separates uncertain suggestions from confirmed ledger data.
AI ecosystem
Design prompts around observable requirements, separate instructions from evidence, and test boundaries with a fictional customer-support example.
AI Evaluation
Two fictional classifiers both score 80% but route different requests incorrectly. Use the example to build an error ledger and a reviewable comparison.
AI ecosystem
Identifying and avoiding three frequent errors that waste developer time and degrade LLM output quality, with concrete examples and preventive patterns.
Agent Design
Discuss partial AI outputs using fictional workflows, explicit completion states, evidence boundaries and a small review exercise for product teams.
AI ecosystem
A community exercise for designing fixed scoring rubrics, testing AI output consistency across runs, and separating rubric judgment from model preference.
AI ecosystem
A practical guide for breaking down complex AI projects into achievable milestones, celebrating incremental progress, and creating positive feedback loops that drive continuous improvement.