Insights & Updates
Explore insights from our engineering team on software development, AI, cloud architecture, and industry best practices.
Master Claude API integration for enterprise applications. Learn authentication, streaming responses, tool use, vision capabilities, and best practices for building reliable AI-powered features with Anthropic's most advanced models.
Jishu Labs
CTO
January 20, 2026
13 min read
Featured Article
122 articles · page 3 of 11
A rate card lists what each role costs per unit of time. It looks like a pricing artefact and functions as a planning one - it is how a scope estimate becomes a number someone can approve.
Serving a 7B model is roughly 10-30x cheaper than a frontier model for tasks where accuracy is equivalent. The engineering question is which tasks those are, and how to find out without guessing.
A PRD states what should be built and why, precisely enough that engineers can build it without guessing. In a world where implementation is cheap, the specification became the constraint.
Chat history is not memory. A memory layer is durable, retrievable state about decisions, preferences and facts that survives past the context window. What belongs in one, and what should stay in a log.
A code smell is a surface indication that usually corresponds to a deeper problem. The word 'usually' is the whole point, and it is the part that gets dropped when smells become lint rules.
You usually do not need a dedicated vector database. A practical pgvector schema covering chunk modelling, tenant filtering before search, index choice, and the re-embedding problem nobody plans for.
Four measures of software delivery performance: deployment frequency, lead time, change failure rate and time to restore. They work because they are hard to game together - and they break the moment one becomes a target.
Long context windows and better tool use took work away from retrieval-augmented generation. RAG did not become obsolete — its job got narrower. A decision framework for when to retrieve, when to load, and when to call a tool.
Acceptance criteria define the conditions under which work is considered done. Written well they end scope arguments before they start; written badly they are a second, vaguer copy of the ticket title.
An embedding is a list of numbers representing meaning, so that similar things sit close together. What they are, how similarity search uses them, and the practical decisions — dimensions, chunking, distance metric — that determine whether retrieval works.
Transcription became accurate and cheap, which made it a commodity. The value moved to what happens after the transcript - and to knowing which problems speech is genuinely the right interface for.
Prompt wording stopped being the bottleneck around 2025. The harder problem — deciding what lands in the context window at every turn — got a name. What context engineering is, and the four things it covers.
Get the latest insights on software development, AI, and technology trends delivered to your inbox.
Get consistent results. Collaborate in real-time.
Build Intelligent Apps. Work with Jishu Labs.