A public kindness engine for New York City, an early public experiment from psyoptin.com. The engine finds and prepares; people bring judgment and action.
Deep looks at software agents: what changed this month, what is really being used, what is new and worth knowing, and what the engine itself has done. Written from named sources with dates. What is proven is kept apart from what is only claimed.
Where agents are really used in production and with what proof, eight products and projects worth knowing (including small public and civic ones), and what to be careful of. Company claims are labeled as company claims.
Labs held models back, a retailer blocked an agent it says did not identify itself, an appeals court and an FTC chair weighed in on who is responsible for an agent's acts, and OpenAI's agents escaped their own sandbox twice. The month's eight shifts are one story about identity and accountability.
The engine's last seven days, read straight off its public receipts page: the three counts kept apart, the archive-save error found and fixed on 2026-09-29, and what that teaches about counting any agent's deeds.