Claude Code now supports artifacts
Preview in-progress work as a live, interactive artifact built from full session context — and share it with your team.
20 articles · 5 categories
What happened in AI — Jun 18, 2026
Thursday, Jun 18, 2026
In 30 seconds
- Claude Code adds live, shareable artifacts built from full session context.
- Anthropic documents seven methods for steering Claude — skills, hooks, subagents and more — with their context cost and authority.
- Microsoft launches Scout at Build 2026, an always-on "autopilot" agent built on OpenClaw.
- GitHub Copilot ships an agent finder; OpenAI adds enterprise spend controls and usage analytics.
- OpenAI's reasoning model helps clinicians find 18 new diagnoses in unsolved rare-disease cases; GPT-5.5 Instant upgrades ChatGPT health responses.
- Enterprises can now centrally provision MCP connectors via Okta, and FERC's large-load interconnection ruling reshapes AI-factory grid access.
Agent tooling did the heavy lifting today. Anthropic shipped live, shareable artifacts inside Claude Code and published a field guide to the seven ways you can steer Claude's behavior, while GitHub turned on an agent finder for Copilot and Microsoft used Build 2026 to launch Scout — an "always-on autopilot" that works autonomously on a user's behalf. The throughline for platform engineers: the agent surface is consolidating into named primitives (skills, hooks, subagents, finders, autopilots) you're now expected to wire together rather than invent.
OpenAI spent the day in healthcare. GPT-5.5 Instant powers a revamped health-and-wellness mode in ChatGPT, a reasoning model helped clinicians land 18 new diagnoses in previously unsolved rare-disease cases, and Mass General Brigham put out a new benchmark for evaluating AI on everyday patient care — a reminder that the evaluation layer is racing to catch up with the deployment layer.
Underneath the product news, the plumbing kept tightening: enterprise spend controls and usage analytics for ChatGPT, centrally managed MCP-connector authorization through Okta, and a FERC ruling on large-load grid interconnection that directly shapes how fast AI factories can come online.
Agent Tooling & Dev Platforms 7 items
A dense day for agent infrastructure: Anthropic, Microsoft, and GitHub all shipped, and the open-source corner kept filling in memory and browser primitives.
Steering Claude Code: skills, hooks, subagents and more
A map of the seven methods for instructing Claude's behavior, each with its context cost and authority — useful for anyone designing agent harnesses.
Microsoft Scout: enterprise autopilot built on OpenClaw, announced at Build 2026
Scout is an always-on agent in a new category Microsoft calls Autopilots — agents that work autonomously on a user's behalf.
Agent finder for GitHub Copilot now available
Copilot gains a way to discover and route to specialized agents — another step toward an agent marketplace inside the IDE.
Memharness: bi-temporal memory for AI agents, in one SQLite file
Show HN project packing agent memory with bi-temporal versioning into a single SQLite file — a lightweight take on the persistent-memory problem.
Agentbrowse: drive any website from the terminal, built for AI coding agents
A terminal-driven browser automation tool aimed squarely at coding agents that need to operate the web.
The Roadmap to Mastering AI Agent Evaluation
A practical walkthrough of how to evaluate agents — timely as the tooling explosion outpaces the eval discipline.
Enterprise Controls & Governance 2 items
The boring-but-load-bearing layer for platform teams: cost controls and centralized connector authorization.
New usage analytics and updated spend controls for ChatGPT Enterprise
OpenAI adds spend controls and usage analytics so organizations can manage costs as they scale ChatGPT.
Centrally manage authorization for MCP connectors
Admins can provision MCP connectors org-wide through their identity provider (starting with Okta), with access granted automatically on first login.
AI in Healthcare & Science 5 items
OpenAI pushed hard into clinical use cases, and the evaluation community responded with new benchmarks for everyday care.
Improving health intelligence in ChatGPT
GPT-5.5 Instant strengthens ChatGPT's health and wellness responses with better reasoning, context, and physician-informed evaluations.
Using AI to help physicians diagnose rare genetic diseases in children
Researchers used an OpenAI reasoning model to identify 18 new diagnoses in previously unsolved rare-disease cases.
New benchmark evaluates AI for everyday patient care
Mass General Brigham introduces a benchmark for measuring AI performance on routine patient-care tasks, not just exam-style questions.
AINews: Midjourney Medical — scan your organs like you step on a scale
The only bootstrapped frontier lab announces its second product, moving into consumer medical imaging.
Agentic AI, Biology, and What Remains Human
A reflective essay on where agentic AI fits in biology research and where human judgment still matters.
Compute & Infrastructure 3 items
The physical layer behind the agents — grid access, national compute build-outs, and AI in advertising — mostly via NVIDIA.
How FERC's large-load interconnection actions help address grid stress and affordability
A FERC ruling on large-load interconnection directly affects how quickly AI factories and semiconductor fabs can connect to the grid.
France advances Europe's AI future with NVIDIA technologies
A year after GTC Paris, France reports progress on AI factories, national compute capacity, and open frontier models.
At Cannes Lions, NVIDIA partners reshape advertising and marketing with AI
NVIDIA's ecosystem pitches autonomous operations for the advertising and marketing industry.
Voices & Engineering Notes 2 items
Long-form perspectives — an investor's playbook and a deep dive on change-data-capture at scale.
The Professor of Outputmaxxing — Anjney Midha, AMP
A profile of the investor behind rounds in Anthropic, Mistral, Black Forest Labs, and Periodic Labs, and the AMP thesis.
Write-Ahead Intent Log: A Foundation for Efficient CDC at Scale
Engineers explain how Debezium hit limits under peak load and how a write-ahead intent log enabled efficient change-data-capture across heterogeneous databases.