A developer is building Soil Doctor, an agricultural advisory system whose retrieval-augmented generation pipeline combines semantic embeddings, BM25 lexical search, and cross-encoder reranking over a…
Anthropic's prompt caching documentation states that Claude Code re-sends the full conversation context on every turn because the model keeps no state between requests, and cache reads are billed at r…
Towards AI co-founder and head of community Louis-François Bouchard advised in newsletter LAI #144 that AI teams change the generation prompt and the judge prompt in separate evaluation runs, because …
Engineering teams deploying generative AI agents with direct write access to Electronic Health Record systems see scheduling failures within weeks because clinical appointment scheduling is a multi-di…
A model upgrade to GPT-5.6 (Sol, Terra, Luna) dropped one team's prompt cache hit rate from roughly 91% to near zero because the new caching model only stops at the end of an eligible message, not at …
OpenAI's GPT-6 Sol and Luna models, launched September 22 at API prices 50% below their GPT-5.6 counterparts, support explicit prompt-cache breakpoints through the Responses API that cut cached input …
A developer built transcriptcut, a transcript-based video editor, after deciding Descript's $24 monthly subscription was too expensive for a workflow limited to recording, transcribing, cutting unwant…
TypeSafe's Jev AI is a "System One" decision model that takes state plus typed questions and returns choices, scores, yes/no judgments, probabilities and confidence signals for software to act on dire…
An agent's decision to search a data store or answer from model memory usually rests on a single hand-written description that goes stale as users add and delete files, according to an analysis priced…
Claude Code connects to external tools such as browsers, databases, and GitHub through MCP (Model Context Protocol), which supports two transport types: HTTP for already-deployed remote servers and st…
A developer who failed a system-design interview for a sales-agent memory layer built CogLayer, a working implementation of the Mem0-style hybrid retrieval plus LLM-judgment pipeline described in arXi…
TypeSafe shipped a model called Jev on 15 September 2026 that returns typed answers with probabilities instead of generating text, a category TypeSafe calls a System One model. Jev accepts a state and…
Agentic AI's second wave will let domain experts at manufacturers, logistics providers, banks and retailers build their own software directly, shifting enterprises from passive software consumers to a…
The second part of an AI fundamentals explainer on transformer attention mechanisms details four widely adopted techniques — Rotary Position Embedding (RoPE), the KV cache, Flash attention, and cross-…
A product manager with more than 10 years of experience has spent two evenings a week since June running a weekly "rehearsal" for a one-person business, giving each candidate idea its own AI workspace…
Vision LLMs combine three components — a vision encoder such as a Vision Transformer (ViT), a projector, and a language model — to accept both an image and text as input and respond in text, powering …
A writer tested three ways of using AI on the same decision note — cleaning up a self-written draft, having the model tighten ugly notes, and handing over the conclusion entirely — and found that the …
Google's open-source agentic orchestration runtime Google AX, published on September 20, 2026, drops the destination port field at the gRPC protobuf wire-format boundary, allowing sandboxed agents cle…
An essay by an unnamed author argues that AI systems such as GPT can search the web and hold large context windows but still fail at the upstream problem of deciding what to retrieve before making a j…
Agentic AI and generative AI differ in that agentic systems perceive their environment, plan multi-step action sequences, and execute those plans with minimal human intervention, while generative AI p…