Blog
Insights, tutorials, and updates from the Fanpino team.

77% of Manufacturers Use AI. Over Half Aren't Sure Their Systems Are Ready.
In a survey Deloitte cites, only 6% of manufacturers used agentic AI and 24% expected to within two years. Most factories aren't held back by AI, though. They're held back by data that isn't clean, real-time, or structured enough to act on.

79% Say They're Adopting AI Agents. Most Never Reach Production
In PwC's survey, 79% of executives said their companies are adopting AI agents, yet two-thirds said fewer than half their staff actually use them. The gap isn't model quality. It's trust.

Cite or Refuse: Why an Internal AI Assistant Should Sometimes Say "I Don't Know"
A chatbot that confidently answers a question your documents never covered isn't being helpful; it's manufacturing a wrong answer with the same tone as a right one. The fix isn't a better prompt. It's an architecture that can't fabricate a citation.

The Real Reason University IT Systems Don't Talk to Each Other
It's rarely a technical limitation. A university's website, faculty records, service desk, and insurance system usually don't fail to integrate; they were never designed to, because each was bought by a different office in a different year.

Why Some Organizations Still Choose to Self-Host Email in 2026
Cloud email is the default answer for a reason. For organizations with 10,000 mailboxes, data-residency rules, or a currency that makes dollar-billed SaaS expensive, self-hosting is still the more rational choice, if you get the details right.

The Real Cost of Running a Repair Shop on WhatsApp and a Notebook
"Is my phone ready?" shouldn't require a phone call. Three places repair shops quietly lose money without noticing, and why they all trace back to the same root cause.

Generative Engine Optimization: How to Get Your Site Cited by ChatGPT and AI Overviews
Ranking #1 on Google no longer guarantees an AI cites you. In Ahrefs’ data, only 12% of the URLs AI assistants cite rank in Google’s top 10 for the same query. Here's the concrete technical checklist we used to make our own site retrievable by ChatGPT, Perplexity, and Google's AI Overviews.

You Can't Reconstruct Compliance at Inspection Time
In most organizations nobody knows exactly which revision of which standard was the reference, and they find out at handover.

Why Your Multi-Thousand Document Archive Is Effectively Unusable
Storage isn't the problem; nobody knowing what's in there is. And an answer without a citation is worse than no answer.

We Have an ERP: What Would an MES Add?
Your ERP knows what the order is and what it costs. But "where is part B-214 and who welded it?" is answered somewhere else.

Six Things to Check Before Buying a Helpdesk System
What separates a good helpdesk from a bad one isn't the feature list; it's a few structural decisions that show their consequences a year later.

How a 7M-Parameter Model Outscored Much Larger LLMs on ARC-AGI (Is Bigger Still Better?)
The usual assumption in AI is that bigger is better. Yet a 7-million-parameter model, TRM, outscored LLMs like Gemini 2.5 Pro and o3-mini on the ARC-AGI-1 puzzle benchmark, and its 27M-parameter predecessor HRM beat o3-mini there too. Here is what the papers and ARC Prize's own re-tests actually show.
Last updated: