Your AI agent said “done.” How do you know it worked? ↗
Evaluate AI agents by their final state, test retries and partial failures, and measure quality and cost per task that actually passes.
EXPLORE / ERNANI.DEV
Ideas born from practice. Explore the topics and find your next read.
Evaluate AI agents by their final state, test retries and partial failures, and measure quality and cost per task that actually passes.
Design an API change monitor with Firecrawl, evidence, dates, and validation, turning page collection into a practical B2B micro-SaaS opportunity.
MCPs and extensions expand what AI can do. Learn to verify provenance, permissions, data destinations, and contracts while distinguishing different risks.
Google and Shopify point to a new shopping interface. Drawing on my integration work at Oombe, I explain why catalogs, pricing, inventory, and authorization remain central.
A campaign can accelerate sales before operations are ready. Here is what I would review in inventory, orders, notifications, and fulfillment before Black Friday.
An analysis of continuity, integrations, and everyday decisions: what I would assess in Claude Duet and how I would design a pilot with verifiable results.
Useful agents need understandable boundaries. An integration example shows how to connect proposals, approvals, and execution while keeping control of operations.
Two lollipops illustrate a software decision: protecting data starts with collection, permissions, and disposal. Real experiences, OWASP, and MITRE.
GTA VI is the starting point for a conversation about AI, REA, PS5 jailbreaking, PC ports, and preservation. A look at the possibilities and limits of this transformation from someone who develops systems, integrations, and Unity games.
FROM IDEA TO PRACTICE