Practices and Perspectives on Data Intelligence
What we've learned in the field — about data, about AI, and about making technology truly serve business.
Decoding Qwen3.8-Flash-Next: How 51B N-gram Embedding Reshapes the Inference Stack
Beyond the widely-misquoted 58B figure, the 51B N-gram Embedding module is the real story. We unpack its sparse-lookup architecture and address three persistent misconceptions — it is not built-in RAG, it does not retire RAG, and it will not deliver a step-change in general capability. A more measured engineering read on what this release actually changes.
Agents Hit 'Critical': Auditability Unlocks Enterprise AI
OpenAI's Astra is the first model to hit the Critical cybersecurity tier. For enterprises, the adoption gate isn't the score — it's agent auditability.
Data Agents Have Their Own Market Now — 3 Signals from IDC
On September 1, IDC published its first China Data Agent vendor assessment: 18 vendors entered, only four reached the Leaders quadrant, and the firm forecasts that 60% of China's top 500 enterprises will deploy enterprise-grade data agents by 2028. Data agents have officially become their own procurement category. Here are the three signals that matter and the four foundations to lay before going live.
Agents Hit the Runtime Wall: ServiceNow's 9×, Replica Cyber's Isolation, and How OntiCards Got Ahead
Enterprise agents made the same pivot three times in the last week of August: ServiceNow's production-agent customers grew 9× in nine months; Replica Cyber bolted an isolation engine onto high-risk agents; Kyndryl and Google Cloud used a semantic layer plus guardrail agents to lock Swiss Incore Bank's KYC into a governed boundary. Gartner says 40% of agentic projects will be cancelled by end of 2027 — not because agents fail, but because governance is still catching up.
Text-to-SQL in Production: From 90% to 25% — Why the Semantic Layer Is Non-Negotiable
Models that score 90%+ on academic benchmarks collapse to ~25% on real enterprise schemas. New arXiv research and dbt Labs production experiments agree: without a semantic layer, LLMs struggle to even find the right table.
Semantic Layer Is the New Data Stack for Agents — Oracle, Google, and the Benchmarks All Agree
Oracle wants agents to pick trusted reports, not write SQL. Google folds Measures + Graph into BigQuery. A SIGMOD 2026 semantic-layer-mediated NL2SQL agent pushes Spider 2.0 accuracy from 17% to 94.15%. Three very different players landed on the same idea in the same week.
Agent ROI Phase: Why Data Readiness Decides Success
SPD Bank has deployed over 2,500 financial AI agents across 440+ AI application scenarios. Jiangsu Bank's daily token usage grew 18x in six months, with AI applications displacing 1.2 million person-hours of manual work. As AI shifts from pilot to ROI measurement, data quality and query accuracy—not model size—decide who succeeds.
Making Data Speak: How OntiCards Built a Four-Layer Architecture
Upgrading data from 'fields lying in a warehouse' to 'knowledge assets business teams can directly ask about.' Here is how we decomposed that problem.
A Real E-Commerce Data QA Walkthrough: From Orders, Refunds, and Users to Business Answers
We took our internal test environment's e-commerce data source (5 tables) and ran the full OntiCards flow end-to-end. Here's the record, with real data.