$ today_in_10_seconds
Guidewire shipped an agent framework for insurers that treats policy and claims context as the product, not the model. Meanwhile the EU's transparency rules went live on Sunday, and AWS quietly stopped letting new customers touch Bedrock Agents. Three different companies, one message: the agent platform layer is being rebuilt around context, identity, and proof.
$ lead_story
Guidewire ships an agent runtime for insurance
Guidewire launched its Agentic Framework in the new Qusar release yesterday, letting insurers build, deploy, and manage AI agents directly on Guidewire Cloud Platform, with per-task model selection and secure real-time access to policy, claims, and billing data and workflows. Qusar includes prebuilt claims and underwriting agents, plus Developer Assistants that Guidewire says ship features more than 40% faster than generic coding assistants — those assistants understand Gosu, Jutro, and APD configuration, which no general-purpose coding agent does. The concrete surface is unglamorous and useful: voice-based first-notice-of-loss guidance, claim summaries, policy-change support. Read past the insurtech framing and this is a vertical software vendor deciding that the defensible layer isn't reasoning — it's owning the schema, the workflow, and the audit trail the agent has to touch.
// why_it_matters: The moat isn't the model. It's who controls the data the agent can't function without.
$ also_worth_attention
TikTok's agentic AI strategy points to a new era of ad tech
TikTok is betting ad tech shifts toward agents that connect systems, recommend actions, and execute workflows rather than just generating content — the June launch of Agentic Hub, a marketplace built on TikTok for Business MCP, is the vehicle. HubSpot, Wix, Innovid, Kochava and a dozen others have already published Skills covering campaign creation, creative generation, catalog management and performance diagnostics, and advertisers can skip the Skills entirely and point their own agents at the MCP server. The stated upside is fewer intermediaries, because agents automate the negotiation and the transaction. Note the distribution model: MCP isn't a protocol here, it's a storefront.
link →The EU AI Act started biting on Sunday
Article 50 transparency and information obligations became generally applicable and enforceable across the EU on 2 August 2026 — if your product talks to users, or generates text, images, audio or video, the duties apply whether or not the system is high-risk. The same date switched on penalty enforcement for general-purpose model providers, and the Commission can now fine retroactively for obligations dating back to August 2025. The high-risk regime did not arrive: the Digital Omnibus on AI, signed 8 July, pushed stand-alone Annex III systems to 2 December 2027 and embedded systems to 2 August 2028. If you've been telling yourself you have until 2027, that's only true for one chapter — your agent still has to disclose that it's an agent, today.
link →AWS closed Bedrock Agents to new customers
Bedrock Agents — launched November 2023 — is now Bedrock Agents Classic, in maintenance mode and shut to new customers as of 30 July 2026, with AgentCore as the recommended migration target. Existing agents keep running, but the model catalog is frozen: every future model generation skips Classic entirely. AgentCore offers two paths — a config-based harness that handles compute, memory, identity and observability, or code-defined agents on any framework, including Strands, LangChain, the OpenAI Agents SDK or the Claude Agent SDK. A flagship agent product went from launch to historical document in 32 months. Price your framework bets accordingly.
link →
$ bigger_picture
The trend behind the headlines
Look at what all four stories are actually about and it's the same short list of primitives. Google made Agent Runtime and Agent Identity generally available last week — a native IAM type built on open standards that enforces least privilege, binds access to the agent runtime to mitigate token theft, and produces non-repudiable auditing of every agent action. Guidewire is selling context. TikTok is selling distribution. AWS is rebuilding the runtime. Brussels just set the floor for disclosure. Nobody in this list is competing on whether the model can reason — that fight is over. The competition has moved to identity, permission scope, and whether you can prove after the fact what your agent did and on whose authority. That's not a governance tax bolted on at the end. On this evidence, it's the product.
Agent Identity (Gemini Enterprise Agent Platform)
Worth reading even if you'll never deploy on Google Cloud. Each agent gets a unique SPIFFE-formatted ID supported directly in IAM, as a granular alternative to shared service accounts. It also automatically manages the identity lifecycle to eliminate dormant credentials — the failure mode nobody budgets for until an offboarded engineer's agent is still running. Treat the docs as a spec for what agent identity should do, then check what your own stack actually does.
$ subscribe --free
Get the next edition in your inbox
// DAIly Agent covers agentic AI three times a week
// free, no spam, exit anytime