Figma is moving AI design from outputs to operating methods
The important change is not prettier generated screens, but reusable tools built from team context and design judgment.
13 min
记录 AI 工具、开源应用、自托管部署和日常开发中的具体问题。
Bug reporting is a high-frequency workflow that deserves specialized automation
The value is not a generic Jira bot, but a governed skill chain for submission, confirmation, and statistics.
10 min
The bottleneck in AI side income is problem selection
A pragmatic view: tools matter only after a person has a real problem, audience, and repeatable delivery path.
10 min
Logback, Log4j2, and the Real Cost of Production Logging
A practical evaluation guide for Java logging, focused on choosing Logback or Log4j2 for production services, operating boundaries, failure behavior, and production readiness.
5 min
D2 Makes Architecture Diagrams Maintainable When Teams Treat Them Like Code
A practical evaluation guide for D2, focused on diagram-as-code documentation, operating boundaries, failure behavior, and production readiness.
6 min
ByteChef Becomes Interesting When Agent Workflows Need Governed Execution
A practical evaluation guide for ByteChef, focused on AI workflow automation, operating boundaries, failure behavior, and production readiness.
6 min
Flutter is a coordination-cost decision, not just a cross-platform framework
A product-operations view of Flutter: the real gain is reducing multi-platform delivery drag, not pretending one codebase removes all platform work.
5 min
Spring Boot and Apache Tika Need an Ingestion Pipeline, Not a Utility Method
A practical evaluation guide for Apache Tika in Spring Boot, focused on document ingestion services, operating boundaries, failure behavior, and production readiness.
5 min
The microservices rollback is really an operations story
Teams are not rejecting modularity. They are rejecting a deployment model that turns every feature into distributed operations overhead.
5 min
Mistral OCR 4 Should Be Evaluated by Structure, Provenance, and Downstream Use
A practical evaluation guide for Mistral OCR 4, focused on structured document AI, operating boundaries, failure behavior, and production readiness.
5 min
Claude Code Slash Commands Work Best as a Small Development Control Surface
A practical evaluation guide for Claude Code commands, focused on repeatable coding-agent workflows, operating boundaries, failure behavior, and production readiness.
5 min
All-POST APIs Are an Operations Contract, Not an API Religion
A practical evaluation guide for all-POST API design, focused on API style decisions, operating boundaries, failure behavior, and production readiness.
5 min
Hermes MoA is decision hygiene for agents
MoA should be used as an escalation path for risky agent work, not as the default way to make every answer longer.
5 min
Team Skills Are the Operating Layer for AI Product Engineering
A team Skill system defines what agents read, produce, verify, and write back so AI-assisted product engineering becomes governable.
5 min
SGLang Should Be Judged as an Inference Operations Layer
A practical evaluation guide for SGLang, focused on LLM inference serving, operating boundaries, failure behavior, and production readiness.
6 min
OpenSpec for AI Coding: A Control Layer Before the Patch
OpenSpec turns vague AI coding requests into reviewable Change artifacts, giving long-running agents a safer control surface.
6 min
Local Coding Agents Are Infrastructure, Not Just a Model Choice
A local coding agent stack needs operational boundaries around model serving, harness behavior, filesystem access, telemetry, and rollback.
6 min
Firecrawl turns the web into an agent data ingress layer
A product-operations view of Firecrawl: search, scrape, crawl, structure, and screenshot web data so agents can run recurring research and monitoring workflows.
5 min
Loop Engineering is governance for persistent coding agents
A product-operations take on loop-driven development: the real challenge is not making agents run longer, but defining completion, risk zones, stop rules, and accountability.
6 min
Hermes MoA is best understood as decision hygiene for agents
A product-operations view of Hermes MoA: multi-model reasoning is useful when it becomes a review checkpoint for high-risk agent work, not a default way to make every answer longer.
8 min