DiffusionGemma
Simon Willison on DiffusionGemma, Google's return to diffusion-based language models after last May's fast-but-abandoned Gemini Diffusion preview.
14 articles · 5 categories
What happened in AI — Jun 10, 2026
Wednesday, Jun 10, 2026
In 30 seconds
- Anthropic published a guide to building with Claude managed agents.
- A wave of open-source agent infrastructure landed: Interbase (long-running goals), SafeAgentDB (per-branch isolated DBs), and AgentMeter (agent cost tracking).
- Xcode 27 added Gemini to its agentic coding toolset; AWS detailed agents that auto-optimize Trainium/Inferentia kernels.
- Microsoft open-sourced pg_durable for durable workflows inside PostgreSQL; Azure APIM shipped a unified model API with MCP content safety.
- Simon Willison flagged DiffusionGemma, Google's renewed push on diffusion language models.
- Jeremy Howard floated a provocative scheme for slowing recursive AI self-improvement at the top lab.
The day's center of gravity was the scaffolding around AI agents rather than the models themselves. Anthropic published a guide to building with Claude managed agents, a fresh crop of open-source projects tackled the unglamorous parts of running agents in production — long-running goals (Interbase), isolated per-branch databases (SafeAgentDB), and cost visibility (AgentMeter) — and the tooling story stretched from the IDE, where Xcode 27 folded Gemini into its agentic coding toolset, down to the silicon, where AWS detailed agents that hand-tune Trainium kernels.
Underneath the agent layer, infrastructure had a busy day. Microsoft open-sourced pg_durable to run durable workflows natively inside PostgreSQL, Azure API Management shipped a unified model API plus content-safety policies that now cover MCP tool calls, and a new Python library, llmbuffer, went after prompt-cache hit rates for big cost savings — a reminder that as agents proliferate, the economics and plumbing get most of the attention.
Models weren't entirely quiet: Simon Willison flagged DiffusionGemma, Google's return to diffusion-based language models after last year's Gemini Diffusion preview, and Latent Space covered the aftermath of Anthropic's Fable 5 launch and its controversial usage terms. Jeremy Howard, meanwhile, offered the day's sharpest provocation on how labs might voluntarily slow recursive AI self-improvement.
Frontier Models & Research 2 items
A new diffusion language model from Google, and the fallout from Anthropic's latest frontier launch.
[AINews] Anthropic Claude Fable 5 — Mythos but Safe, with Controversial Terms
Latent Space covers the aftermath of Anthropic's Mythos-class Fable 5 launch, where the model's debut was overshadowed by controversial usage policies.
Building & Orchestrating Agents 4 items
From managed-agent guides to long-running orchestration and IDE-to-silicon tooling — the day's strongest thread was how to actually build and run agents.
Building With Claude Managed Agents
Anthropic walks through building with Claude managed agents, its framework for standing up agentic workflows.
Show HN: Interbase – Long-running AI goals and aliases for any model
An open-source CLI agent built around long-running agent workflows and model-agnostic aliases for persistent goals.
Xcode 27 expands agentic coding toolset with Gemini integration
Apple's Xcode 27 broadens its agentic coding capabilities by integrating Google's Gemini models.
Stop hand-tuning kernels: How Neuron Agentic Development accelerates AWS Trainium optimizations
AWS introduces Neuron Agentic Development — AI agents and skills that automate kernel optimization for Trainium and Inferentia.
Agent Cost, Safety & Memory 4 items
The practical concerns of agents at scale — what they cost, how to isolate them, and how to manage their context.
Show HN: AgentMeter – Know what your AI coding agents cost
An open-source tool for tracking and surfacing the cost of AI coding agents.
SafeAgentDB – Isolated databases for every AI agent branch
A project that gives each AI agent branch its own isolated database, reducing the blast radius of agent actions.
Show HN: Llmbuffer – Python library for cache-optimized LLM conversation history
A Python library that maximizes LLM prompt-cache hits via dynamic context, compaction, and tool-output truncation — pitched at roughly 10x cost savings.
Beyond Prompting: Context Engineering and Memory Management for AI Systems at Scale
Adi Polak on the architecture needed to move from stateless prompts to state-aware, context-rich AI agents.
Platforms & Infrastructure 3 items
Durable workflows, model gateways, and enterprise deployments — the plumbing that carries AI into production.
Microsoft Open-Sources PostgreSQL Extension for In-Database Durable Execution
Microsoft's pg_durable runs durable workflows natively inside PostgreSQL, removing the need for an external orchestration system.
Azure API Management Ships Unified Model API and MCP Content Safety at Build 2026
Azure APIM's unified model API lets clients speak one format while it transforms requests to Anthropic, Vertex AI, and other backends, with content-safety policies now covering MCP tool calls.
From data to decisions: how LSEG is scaling trusted AI
OpenAI on how LSEG scales trusted AI across its global business — accelerating insights, shrinking release cycles, and empowering 4,000 employees.
Commentary 1 item
The day's sharpest take on AI's trajectory.
Quoting Jeremy Howard
Jeremy Howard floats a scheme for slowing recursive AI self-improvement: the lab with the top-ranked model agrees not to use it for frontier AI work, while everyone else gets access.