Insights
Scritti pratici su AI, tokenizzazione e regolamentazione europea — per chi deve metterci la firma.
What the 2025 predictions got wrong, the EU AI Act deadline that moved in July 2026 and the one that quietly passed, and where the money actually went — with dates, numbers and sources.
26 ago 2026 · 7 min di lettura
Seven workflows a two-to-ten person team can run today, what each one realistically saves, and the three that are usually not worth the setup.
02 ago 2025 · 4 min di lettura
The 2025 automation playbook assumed smarter models were the bottleneck. Two years of production data says otherwise — here is what separates the automations that survive from the ones that get cancelled.
01 ago 2025 · 6 min di lettura
Real estate was supposed to be tokenization's flagship use case. Treasuries took roughly $15bn on-chain and property around $200m. Here is why, and the narrow cases where it still works.
20 lug 2025 · 6 min di lettura
Three regulations, three different clocks, and several deadlines that have moved. What is enforceable today, what was deferred, and how to tell which rules your product is actually under.
15 lug 2025 · 5 min di lettura
Three certifications, three different buyers, and a new one for AI that European procurement has started asking about. What each proves, what it costs, and the order to do them in.
15 giu 2025 · 5 min di lettura
Altrove
Note di ingegneria pubblicate sul sito personale di Abhishek — architettura degli agenti, RAG, MCP e conformità all'AI Act europeo.
26 lug 2026
When an off-the-shelf AI tool beats a custom agent, when it doesn't, and the four questions that decide it — from someone who profits from 'build' but will tell you to buy.
09 giu 2026
An honest, docs-grounded comparison of LangGraph and CrewAI — mental models, state and durability, human-in-the-loop, protocols, and which one to actually ship to production.
18 mag 2026
Anthropic defines how to structure agent workflows. LangGraph provides the memory primitives. Nobody connects the two. This post does — a practical design framework for matching each Anthropic agent pattern to the right LangGraph memory architecture.
17 mag 2026
Your LangGraph agents forget everything the moment a session ends. Here's how to fix it — a production guide to agent memory architectures, Mem0 vs Zep vs LangMem benchmarks, token cost reduction, and GDPR compliance.
08 mag 2026
Most agent memory systems store facts and retrieve them. Production memory does more: it learns behavioral patterns, promotes them to rules, compacts history without losing signal, and scores retrieval with tier-aware decay. Here's the full architecture.
03 apr 2026
Real cost breakdown of building AI agents — LLM API pricing, infrastructure, vector stores, and development time from 4 production agents.
