Insights
Practical notes on production systems.
Paul Titov & Co publishes engineering notes from studio practice: AI in real products, commerce funnels, PWA, measurement, and the systems that have to work before spend or models can help. These articles are written from delivery, not vendor documentation.
7 Oct 2026
How to get AI to recommend your product, and agents to use it
AI assistants can bring buyers to your products. Coding agents can add your libraries to new projects. For that to happen, the system has to find the product, see where it fits, and take the next step.
18 Sept 2026
Hoverwise: PWA with a ledger that refuses to guess
Shared Expenses without the math headache — a live PWA whose ledger stores integers, never caches balances, and treats iOS install as product work.
5 Sept 2026
Why More Traffic Won’t Fix a Broken E-Commerce Funnel
Paid traffic multiplies whatever the store already does. If checkout, tracking, performance, or order data leak, more acquisition spend scales the leak.
27 Aug 2026
When not to use an AI agent
A practical test for deciding when an agent adds value — and when rules, search, a queue, or an ordinary application service will be safer.
25 Aug 2026
RAG inside an existing Laravel product
How to add source-aware AI answers to a mature PHP product without rebuilding identity, permissions, data ownership, or operations around a chatbot.
22 Aug 2026
Build the evaluation set before the AI demo
Why representative cases should exist before prompt tuning, model comparison, and stakeholder demonstrations — and what those cases need to contain.
19 Aug 2026
Permissions and citations before a smarter prompt
A production assistant needs controlled access and inspectable evidence before prompt polish. Here is how those controls fit into the product.
16 Aug 2026
Document extraction needs a human review queue
A reliable document workflow separates extraction confidence from business consequences and gives operators a precise way to review exceptions.
13 Aug 2026
Model choice is not AI architecture
OpenAI, Anthropic, Gemini, Azure, Mistral, and local models are replaceable components. The durable system is the job, data, controls, evaluation, and operations around them.