Product & AI Blog
Notes on building AI products, making trade-offs, and deciding what comes next.
20 articles in AI & LLMs
AI Product Evaluation Checklist: What I Would Check Before Launch
Review an AI feature with task-specific tests, clear release blockers, human-calibrated grading and practical recovery checks.
4 min read · AI & LLMs
RAG vs Fine-Tuning: Start With the Answer That Went Wrong
Diagnose missing knowledge and inconsistent behaviour before choosing RAG or fine-tuning. Includes an OTT assistant decision table.
4 min read · AI & LLMs
Voice AI Latency: Why a Fast Answer Can Still Feel Like a Bad Conversation
Evaluate voice AI beyond model speed with response-start delay, interruptions, corrections and recovery in a learning-coach example.
4 min read · AI & LLMs
AI Feature Adoption Metrics: The First Click Is Not the Result
Measure AI adoption with clear denominators, a worked funnel and task-success checks, rather than treating generated responses as value.
4 min read · AI & LLMs, Metrics & Analytics
The Security Questions Enterprise Buyers Ask About AI Products
Questions enterprise security and procurement teams ask about AI products, covering data use, retention, model providers, access control and incident handling.
4 min read · AI & LLMs
The Blank Prompt Problem: Onboarding Users to an AI Feature
Why an empty AI input box leads to low adoption and how to fix it: starter prompts, task-based entry points, examples from the user's own data and progressive guidance.
4 min read · AI & LLMs
From AI Pilot to Production: Why Enterprise AI Pilots Stall
Why enterprise AI pilots get stuck and how to plan one that can scale: agreed success criteria, real users, integration, ownership and a decision date.
4 min read · AI & LLMs, Execution & Delivery
Build vs Buy for LLMs: Hosted API, Open-Weight Model or Something in Between
How to choose between a hosted LLM API and running an open-weight model: quality, cost at scale, data control, latency, team capacity and switching cost.
4 min read · AI & LLMs, Strategy & Roadmaps
The Recommendation Cold Start Problem: What to Show Someone You Know Nothing About
Product approaches to the recommendation cold start problem for new users and new content: onboarding signals, editorial picks, context and fast learning.
4 min read · AI & LLMs, Metrics & Analytics
LLM Guardrails as Product Requirements: Deciding What the Model Must Not Do
Turn AI safety concerns into product requirements: listing what the model must refuse, where checks run, prompt injection, and how to test guardrails.
4 min read · AI & LLMs
AI Agent or Workflow? Choosing How Much Freedom to Give the Model
When to build a fixed LLM workflow and when an autonomous agent makes sense, judged by predictability, cost, debugging and what happens when a step fails.
4 min read · AI & LLMs
Voice Agent to Human Handoff: Designing the Escalation Path
When and how a conversational voice agent should hand over to a person: escalation triggers, passing context, what the caller hears and how to measure it.
4 min read · AI & LLMs