Thursday, October 22, 2026

Pinned fresh every weekday morning

1. Story 1

   Anthropic launches Claude Duet, its always-on agent

   Anthropic launched Claude Duet, an agent that keeps working on tasks even when the user is offline. It runs on Claude Opus 5.5 and connects to Gmail, Slack, and Stripe through more than 6,000 connectors. It asks for approval before sensitive actions.

   Why it matters: Always-on agents are becoming the default. If you run one, set its approval rules before it touches money, customer data, or your inbox.

   Source: The Decoder

2. Story 2

   AI models found a critical PHP flaw affecting millions of websites

   OpenAI's GPT-5.6 Sol found a critical PHP vulnerability as part of the Patch the Planet initiative, and security firm Wiz found the same flaw independently using Google's Gemini 4 Argon. The flaw could affect millions of websites.

   Why it matters: If your business site runs on PHP, patch it and ask your host how they watch for flaws like this. AI finds these holes faster than humans do now.

   Source: The Decoder

3. Story 3

   Google Cloud now offers OpenAI models alongside Gemini and Claude

   Google Cloud is the first major cloud provider to offer Gemini, Claude, and OpenAI's GPT models on a single platform. At launch, GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna are available in private preview for enterprise customers.

   Why it matters: You may be able to run three vendors' models on the bill you already pay. Ask your Google rep about access before shopping for a new provider.

   Source: The Decoder

4. Story 4

   Inception's Mercury 3 claims a speed lead over small GPT and Claude models

   Inception says its Mercury 3 diffusion language model runs four to eight times faster than GPT-6 Luna and Claude Haiku 5.5, and beats both on eight of twelve benchmarks. There are no independent tests yet.

   Why it matters: Speed matters when an agent makes dozens of calls per job. Test the fast, cheap models on your own workload before trusting the claims.

   Source: The Decoder

5. Story 5

   Replit switches its default AI provider to OpenAI

   Replit replaced Anthropic with OpenAI as the preferred provider for Replit Agent, citing cost, performance, and partnership reasons. Users can still choose Claude and other supported models.

   Why it matters: The default model in your favorite tool can change any month. Keep your prompts and workflows portable so a vendor swap never breaks your week.

   Source: TLDR

6. Story 6

   Ramp rebuilt its agent platform in six weeks with agent-written code

   Ramp built Laminar, its own agent platform, in six weeks with a small team. The code was written entirely by AI agents using the strongest models available. Ramp says good engineers are still needed to keep the platform healthy.

   Why it matters: Agents can now do a serious share of the build, but someone still has to own the result. That someone is a person on call.

   Source: TLDR

7. Story 7

   AI's hidden storage bill: what inference leaves behind

   MIT Technology Review looks at the storage costs AI inference racks up behind the scenes. Running models at scale quietly adds a second bill on top of compute.

   Why it matters: If you run AI at real volume, ask about your storage costs, not just compute. The bill has two halves.

   Source: MIT Technology Review