Build notes › topic

AI & agents

What it takes to put a language model in front of real customers in a WhatsApp thread: what it is allowed to say, what it is allowed to do, how it is stopped from answering on top of a colleague, and what it does when it should not answer at all.

Covered here: Cloudflare Workers AI · 24-hour service window · 7-star customer experience · agent decision tree · agent_decision_tree · agent-eval v1.0 · agentic tool use · AI permissions · AiSensy alternative · allow-list enforcement

• 23 September 2026 · Campaign reliability & auto-responder triage

  The WhatsApp broadcast that started talking to itself: Eliminating auto-responder loops

  Out-of-office auto-responders and BSP system error messages trigger infinite bot ping-pong loops on WhatsApp campaigns. Meta Cloud API provides no is_bot header, so standard CRMs either loop indefinitely or mute real customers. Here is how we engineered deterministic correspondent triage, context-aware thresholding, and prompt refusal gates in Klaros.

• 22 September 2026 · Thought leadership & customer context

  WhatsApp Is Not a Marketing Channel. It Is a Living Room.

  The businesses that treat WhatsApp like a billboard optimize for messages. The businesses that treat it like a living room optimize for memory, context and what happens next. Explains the transition from Net to Spear to Living Room, the true cost of forgetting, and how Klaros grounds decisions, authorized execution, and verified outcomes on top of conversation memory.

• 8 September 2026 · AI architecture & maturity framework

  Klaros Pragmatic AI Maturity: A Situational Agentic Framework for WhatsApp CRM

  Modern AI engineering focuses heavily on autonomous multi-agent swarms and self-improving prompt harnesses. However, in commercial WhatsApp CRM messaging, unconstrained swarms carry massive risks of rate limits, hallucinated commitments, and high latency. This engineering note details the Klaros Situational AI Maturity Framework: an 8-level fail-closed priority decision tree in agent-decision.mjs, a $0 zero-token Jaccard Q&A cache in learned-answer.mjs, a fact-fenced RAG responder in ai-auto-reply.mjs, and a pre-validation JSON repair engine in json-repair.mjs for open local LLMs (Ollama/Qwen). Evaluates YC Harness Club principles against production CRM requirements.

• 7 September 2026 · commerce & platform

  Building an Enterprise Edge-Native PIM Master for Precious Commodity Commerce

  Generic PIM systems assume static product pricing and flat catalog models, failing when applied to precious metal commodities, gold/silver purity formulas, and heavy multi-channel catalog syndication. This engineering note details how we built an enterprise-grade, edge-native PIM Master on Cloudflare Workers, D1, and R2. We cover our dynamic live spot-rate pricing engine with a 36-hour stale-rate circuit breaker, parent-child matrix schema migration, 4-pillar fault-tolerant binary sheet and remote image ingestion pipeline, rotatable Google Merchant Center XML feed syndication, and zero-dependency 60 FPS DOM virtualized catalog table.

• 1 September 2026 · AI evaluation & reliability

  Engineering Klaros AI WhatsApp Agent with Diligent Evaluation

  Nuts and Bolts of Building a Production Grade Voice and Text Product Orchestrator. How we moved from basic prompt engineering to an enterprise-grade evaluation harness (agent-eval v1.0), F0-F5 Cloudflare infrastructure fidelity, micro-confidence voice safeguards (<0.70 handoff), synthetic mutation testing (20/20 detected), and an 8-step controlled rollout protocol (0% to 100%) backed by canonical RFC 8785 JCS SHA-256 evidence.

• 19 August 2026 · AI architecture & safety

  Klaros Chatbot As-Is AI Architecture: Fail-Closed Design and Zero-Token Grounding

  A first-hand developer perspective on our existing WhatsApp chatbot architecture. Outlines our choices behind the "Rules First, LLM Tail, Human Override Always" priority tree in agent-decision.mjs, human staff takeover yield, fact-fenced auto-responder in ai-auto-reply.mjs, zero-token learned answers cache in learned-answer.mjs ($0 token cost), pure JS BM25 retrieval in retrieve.mjs, and D1 token cost accounting.

• 19 August 2026 · updated 21 August 2026 · AI architecture & roadmap

  Klaros WhatsApp Autonomous AI Agent: To-Be Architecture for a 7-Star Experience

  A candid developer architectural roadmap challenging our own past design decisions, now carrying a correction. The hybrid sparse-dense retrieval this note proposed had already been written and registered ahead of every other strategy, and it had never executed once: the Cloudflare Vectorize binding it depended on does not exist on the worker, so every request took its fallback branch and re-served plain BM25 while reporting success. The audit that found it also found the real defect, which was more mundane and worse: ingest deletes the 173 question-and-answer pairs the site publishes in its own structured data, then asks a ranking function to reconstruct them from the prose. Dense retrieval is deferred in favour of contextual BM25 and reranking.

• 15 August 2026 · agentic design

  The WhatsApp assistant that could explain booking but not book

  At 13:00 we shipped a screen that cross-references what our assistant can explain against what it can actually send. It reported scheduling as explainable and not sendable. At 13:56 somebody typed “Book a meeting for me” and was told a person would pick it up, while a working scheduler sat one function away. No corpus entry fixes that: a request for an action is not a question. What closed it, why the enforcement lives in an allow-list rather than the prompt, and the shadowed question that would have denied a feature we had shipped four hours earlier.

• 13 August 2026 · conversational AI

  The WhatsApp auto-reply that answered a hello with a price list

  A contact of five weeks typed “Hello” and received an explanation of our licensing. The model was not hallucinating: the thread query selected the oldest 40 rows of a 46-message conversation, so it was shown messages 21 to 40 and never saw the greeting it was answering. Every timestamp had been stripped too, so a pricing exchange from three weeks earlier read as the previous turn. Investigating one bad reply surfaced seven more ways the same assistant was failing to listen, including voice notes answered with total silence, “cancel my order” recorded as an unsubscribe, and a customer who asked to speak to a person and was ignored twice.

• In progress

  The half of a sale that nobody demos

  Every platform shows you the payment landing. Almost none show you the twenty minutes afterwards: who gets told, who packs it, what the buyer hears while they wait, and what happens when it ships four days later and the messaging window has closed.

• In progress

  Where the ask ends and the shop begins

  A payment request is messaging machinery. A product library is a commerce application, and the part most likely to be replaced by a real one. On finding that seam in our own code, and making it a documented extension point instead of a function that happened to sit in the middle of a file.

• In progress

  What a feature looks like when nothing consumes it

  Contact limits computed on every request and enforced nowhere. A tier ladder that is marketing copy at runtime. On the difference between building a capability and shipping one, found in our own codebase.