Field notes — week of October 5, 2026

Sierra announced the Personal Agent Protocol with Meta on October 6, an open standard for how a consumer's personal AI agent authenticates with a business and what access the consumer grants it, with Genesys, Instinct, Rocket, Shopify, Stripe and Walmart listed as industry partners and a v0.1 spec planned for later in October. Decagon separately open-sourced the Personal Agent Consent & Trust Protocol (PACT) on October 6, co-developed with Instinct and built on the Agent2Agent protocol and OAuth 2.0, and joined the Sierra/Meta-led working group. Anthropic also released Claude Haiku 5.5 on October 7 at $0.10/$0.50 per million input tokens depending on prompt length, scoring 72.4% on OSWorld 2.1 versus 15.7% for Haiku 4.5.

Field notes — week of October 5, 2026 What changed in AI-support and AI-localization tooling this week. Every item was checked against its source before it entered this list. - Sierra - announced the Personal Agent Protocol with Meta https://sierra.ai/blog/introducing-personal-agent-protocol Oct 6 , an open standard for how a consumer's personal AI agent authenticates with a business and what access the consumer grants it; Genesys, Instinct, Rocket, Shopify, Stripe and Walmart are listed as industry partners, with a v0.1 spec planned for later in October. - Decagon - open-sourced the Personal Agent Consent & Trust Protocol PACT https://decagon.ai/blog/introducing-the-personal-agent-consent-trust-protocol-pact Oct 6 , co-developed with Instinct and built on the Agent2Agent protocol and OAuth 2.0, letting a business verify which personal agent is contacting it and what the customer authorized it to do; Decagon also joined the Sierra/Meta-led working group. - Decagon - was named a Cool Vendor in Gartner's 2026 Coolest Vendor Innovations in CRM report https://decagon.ai/blog/decagon-named-a-gartner-cool-vendor announced Oct 7, from a Gartner report dated Sept 14 . - Anthropic - released Claude Haiku 5.5 https://www.anthropic.com/claude-haiku-5-5 Oct 7 , its cheapest and fastest small model yet, priced at $0.10/$0.50 per million input tokens depending on prompt length, with an adjustable effort setting and a large jump on computer-use benchmarks 72.4% on OSWorld 2.1 versus 15.7% for Haiku 4.5 . - OpenAI - extended Ultrafast mode to GPT-6.1 Sol in the Responses API https://developers.openai.com/api/docs/changelog Oct 8 , opening the low-latency service tier to all API users for the first time and adding EU data residency alongside the existing US option. - Intercom - added Macro Admin Mode https://www.intercom.com/changes/en/159696-macro-admin-mode Oct 9 , letting teammates with shared-macro permissions view other people's personal macros and either promote the useful ones to shared status or delete the rest. - DeepL - launched DeepL for Sovereign Markets https://www.deepl.com/en/blog/building-for-europes-sovereignty-needs Oct 6 , a translation environment hosted and operated entirely within the EU for public-sector and enterprise customers, initially in DACH, France and EU institutions, carrying SOC 2 Type II, GDPR, ISO 27001 and BSI C5 certifications.