This Week in Brief
A new book claims under 10 % of shoppers will pay extra for green goods, so companies should view sustainability as a driver of innovation, not an expense.
Sustainability Strategy Shift
The book *Clean Winners: Sustainability Strategy That Puts Customers First* asserts that less than 10 % of consumers will pay a premium for green products.
В книге утверждается, что менее 10 % потребителей готовы платить больше за экологически чистые продукты.
Firms should treat sustainability as an innovation catalyst, not a cost center [S5].
Autonomous‑Ship Buzz
Austin’s Saronic sent drone boats to engage an Iranian submarine and unveiled plans for a $3 billion shipyard that could create 10 000 jobs [S6].
AI & Tools
Loop engineering is gaining traction as Anthropic and OpenAI switch from prompts to loops; developers can now automate PRs, incident triage, and design reviews with the `/goal` command [S1]. Claude Code has shifted to a Rust‑based Bun v1.4.0 “canary,” delivering roughly 10 % faster performance on Linux and simplifying version verification [S16]. Federated learning exploded 100× since 2021; Flower Labs’ framework now supports very large models, opening new use cases across data silos [S17].
Backend & Infrastructure
**TL;DR:** The sqlite‑utils team patched `table.transform()`. It now throws a `TransactionError` if used inside an active transaction with foreign keys on, stopping silent changes to related rows [S15].
Fix Overview
- **Problem**: Calling `table.transform()` while a transaction is open and FK checks are enabled could silently delete or alter rows that reference the transformed table.
- **Solution**: The function now raises a `TransactionError` under those conditions. This forces developers to finish the transaction before transforming, preserving referential integrity.
Impact
- No new features added; just stricter safety.
- Existing code must avoid calling `table.transform()` inside transactions when FK enforcement is on.
Frontend & Web
**TL;DR:** No new frontend‑specific announcements were found this week.
Worth Your Time
- **Loop Engineering Overview** – A quick guide to chaining AI agents into productive loops, plus real‑world use cases [S1].
- **Claude Code Rust Bun Update** – Performance gains and version‑check tricks for startups using Claude Code [S16].
- **Flower Labs Federated Learning** – Why adoption jumped 100× and how new large‑model support powers production federated learning [S17].
My Take
Loop engineering is a double‑edged sword. It promises massive automation, but most developers see significant drift and constant human intervention. Без надёжного мониторинга команды начинают гоняться за багами вместо того, чтобы повышать продуктивность. Такая гонка в колесе сжигает токены. Сам попал в похожую ситуацию, когда AI чрезмерно полировал код и сжигал токены. Пришлось прописывать ряд условий.