AgentVet.ai

← All issues

AgentVet Weekly — June 8, 2026

June 8, 2026

From the founderThe thread I keep pulling on this week is trust boundaries, who controls what an agent can touch, and how the platforms hosting those agents are starting to encode that into their architecture. GitHub's strain from agentic load, Willison's careful work on agentic text editing, and OpenAI's Lockdown Mode all point to the same uncomfortable truth: the industry built agent capabilities faster than it built agent guardrails, and the correction is now underway.

Latent Space

GitHub's plan for Agents — Kyle Daigle, GitHub

GitHub is rearchitecting core platform infrastructure to handle the load and behavior patterns that agentic coding tools like Copilot have introduced at scale — a signal that agent-driven developer workflows are straining systems built for humans. If you're building coding agents or tooling that sits on top of GitHub, Daigle's roadmap tells you where the platform is heading and what constraints to design around. Worth reading before you commit to any GitHub-dependent agent architecture.

Simon Willison

datasette-agent-edit 0.1a0

Simon Willison is open-sourcing an early plugin that lets an AI agent make targeted edits to existing text — Markdown, SQL, SVG — rather than regenerating content wholesale. The challenge he's solving, diffing and applying partial edits reliably, is a concrete implementation problem anyone building document-editing agents will hit. The 0.1 alpha is rough by design, but the approach is worth studying.

Simon Willison

OpenAI Help: Lockdown Mode

OpenAI's Lockdown Mode lets users harden their ChatGPT account against unauthorized changes — think social-engineering attacks that try to alter system prompts or account settings. For agent builders, this matters because it signals that OpenAI is treating account-level security as a first-class product concern, which will shape what programmatic access patterns stay permissible. Buyers deploying ChatGPT-based agents to sensitive users should evaluate whether enabling this by default makes sense.

TLDR AI

Anthropic Oceanus leaks 🤖, ChatGPT Dreaming 💭, recursive self improvement 🚀

Coverage of a leaked Anthropic model codenamed Oceanus, OpenAI's dreaming-style background processing feature, and developments in recursive self-improvement research. Thin on sourced detail given the excerpt, but Oceanus is worth tracking if you're evaluating Anthropic's model roadmap for long-running agent workloads.

TLDR AI

Anthropic IPO filing 📄, OpenAI on AWS ☁️, Perplexity search code 🔍

Anthropic has filed for an IPO, OpenAI is expanding its presence on AWS, and Perplexity is releasing search-related code. The AWS-OpenAI move matters operationally for teams already running infrastructure on AWS who want tighter model integration without multi-cloud complexity.

Get AgentVet Weekly in your inbox

The latest in agentic and applied AI, every week.

Subscribe