This week I dug into evidence that AI crossed a threshold: models that code, test, and reason are colliding with mature agentic plumbing—GLiNER2 for fast, deterministic, schema-bound IE/guardrails, and CASAFARI’s MCP data feed for real‑estate AI and analytics. The takeaway: use frontier models, wire them to clean data, lean into autonomous tools reshaping engineering, and prepare for power shifting from products to the builders.
AI
- ‘GLiNER2 for Agentic Information Extraction’: GLiNER2 evolves GLiNER from zero-shot NER to a unified encoder system for NER, relation extraction, structured (JSON) extraction, and text classification, built for agentic workflows. It creates rich context, runs fast on CPU with deterministic, schema-bound outputs, and rivals frontier models in zero-shot while being easily fine-tuned with few examples. Key uses: model routing, guardrails, hallucination detection, and prompt-injection defense.
- ‘Datos inmobiliarios para sus aplicaciones de IA’: CASAFARI MCP is an API-first, AI-native platform using the Model Context Protocol to build real estate AI apps. It offers clean, comprehensive property data across 20+ European countries and the UAE, enabling predictive models, deep analytics, and intelligent agents with seamless data flow to up-to-date listings.
Others
- ‘Something Big Is Happening’: Matt Shumer says AI has crossed a threshold: 2025 advances and models like GPT-5.3 Codex and Opus 4.6 show judgment, can code full products, test them, and help build successors, triggering accelerating progress. Knowledge work in law, finance, writing, and engineering will be reshaped within 1–5 years. Free tiers lag; adopt top models, use them deeply, stay adaptable, and build resilience. The upside (faster medical breakthroughs) is huge, but risks (misuse, deception, surveillance) are real.
- ‘Go Crazy, Folks, Go Crazy’: Benn Stancil argues AI is reshaping engineering via autonomous tools like Claude Code and agent swarms, while analysts lag. Despite calls for trust and governance, attention favors reckless, high-autonomy tools like Clawdbot that tap everything and iterate hard—someone will ship them. Either analysis is uniquely hard or analysts are timid. As software creation cost collapses, power may shift from products to creator personalities, mirroring art/content.