Agents that can pay: building Restock with Stripe's Link and Managed Deep Agents
Restock pays for real purchases from Slack using Stripe's Link over the Machine Payments Protocol on Managed Deep Agents.
32 articles · 4 categories
The finishable daily brief
Thursday, Oct 8, 2026
32 articles · 4 categories
read top to bottom · then stop
In 30 seconds
Claude Haiku 5.5 is reported to beat GPT-6 Luna at the same price, and Anthropic also launched a Cyber Mission and a new Usage Policy.
For builders, agent tooling got more operable: payments through Stripe's Link, crash reporting for coding agents, and Spotify's multi-agent guardrail patterns.
Agents are gaining payment rails, failure telemetry, and documented automation patterns, which moves them from demos toward operable systems.
Restock pays for real purchases from Slack using Stripe's Link over the Machine Payments Protocol on Managed Deep Agents.
Claude's reference implementation of an agent automation, with the common failure modes called out.
Spotify Ads Manager runs multi-agent systems on Google ADK Java, with domain ownership and deterministic guardrails.
Show HN: crash reporting for coding agents, the Crashlytics model applied to agent failures.
OutSystems opens its Agent Experience platform to Claude Code and Cursor.
Smaller and specialized models are the day's theme: a cheap Haiku release, task-specific tuning tooling, and open decision models.
Claude Haiku 5.5 is reported to beat GPT-6 Luna at the same pricing.
LangSmith Fine-Tuning and the SmithTune CLI train specialized models without hand-built data pipelines.
Cloudflare's Clef open-weight 9B and 27B models choose between predefined options instead of generating text.
Prime Inference offers hosted serving for frontier open models.
Labs are tightening usage rules and cyber defenses while AI-assisted attacks and influence operations keep surfacing.
Anthropic publishes its 2026 Usage Policy with a summary of what changed.
Anthropic launches its Cyber Mission for defensive security work.
Goodfire deploys probe-based cyber monitors for Kimi K3 and GLM 5.3 open-weight models.
OpenAI disrupts two AI-enabled influence operations built on false-front journalists and a think tank.
A hacker in their 20s used AI in attacks on South Korean financial institutions.
Enterprise agent platforms and capital flows are consolidating around a few vendors and a well-funded Chinese lab.
Google launches the Gemini agent, a single universal work agent, at Gemini at Work 2026.
Oracle uses ChatGPT Work and Codex for recruiting, engineering, and operations workflows.
Anthropic commits $150 million over three years to the federal Genesis Mission for AI-driven science.
DeepSeek reportedly nears a $12 billion round backed by Tencent and CATL.
Ecosia replaces Mistral with a Chinese AI model.
You are caught up for this edition