$ cat briefs/daily/2026-07-01.md
2026-07-01
July 1, 2026 (Wed)
2 top stories · 1 paper pick · 3 watch items · 3 new wiki pages
+3new pages
[01]
Top Stories
1. Google ADK 2.0 GA + Agents CLI — agent tooling gets a full stack
- ADK 2.0 reaches General Availability with two flagship additions: graph workflows (fan-out/fan-in, loops, human-in-the-loop, nested) and a Task API for structured agent-to-agent delegation (source)
- New Agents CLI covers the full lifecycle in one tool: scaffold → evals → deploy → observability → publishing; works with Claude Code, Cursor, Gemini CLI, and Antigravity (model-agnostic)
- ADK Kotlin (Beta) joins Python, Go, Java — enables on-device mobile agents coordinating with backend Python agents
- Karpathy: 89% of teams have observability, only 52% have evals — Agents CLI directly addresses this
- Why it matters: This is the first model-agnostic, end-to-end agent toolchain from a Tier-1 lab. Unlike Anthropic Managed Agents (Claude-native) or Azure Agent Mesh (Azure-native), ADK 2.0 + Agents CLI is explicitly designed to run any agent framework. The standardization battle for the agent-infrastructure layer has begun.
- → Google ADK (Agent Development Kit), Google DeepMind
2. Anthropic Claude Science — AI workbench for researchers (June 30)
- Announced at "The Briefing: AI for Science" livestream (June 30, 10am PST); no new model — runs Claude Opus 4.8 (source)
- Connects to 60+ scientific databases with prebuilt toolkits for genomics, protein structure, and chemistry
- 50 funded projects: up to $30,000 in API credits each for postdoc/grad students; applications open through July 15
- Connects the dots: Allen Institute/HHMI partnerships (Feb), Coefficient Bio acquisition ($400M, April), John Jumper hire (Nobel Chemistry 2024, June 19)
- Why it matters: Claude Science is the operationalization of Anthropic's science bet — not a demo or a capability claim, but a deployable product with 60+ database integrations and a funded research program. The workbench-over-model approach directly targets researchers who need workflow, not benchmarks. Anthropic vs. Google's Gemini for Science (30+ databases, May 2026) is now a concrete product race in the research market.
- → Claude Science, Anthropic, John Jumper
[02]
Paper Picks
Agents-A1: Scaling the Horizon, Not the Parameters — arXiv:2606.30616
- TL;DR: A 35B MoE model from SJTU matches 1-trillion-parameter models (Kimi-K2.6, DeepSeek-V4-pro) on agentic benchmarks by scaling the agent horizon (trajectory length and ability breadth), not raw parameters. Average trajectory: 45K tokens. Training: full-domain SFT → domain-level teacher models → multi-teacher routed on-policy distillation.
- Key results: SEAL-0 56.4, IFBench 80.6, HiPhO 46.4, FrontierScience-Olympiad 79.0 — leading vs. 1T-class models. Weights publicly available on HuggingFace.
- Why read it: The claim that the primary agentic scaling axis is trajectory length/ability breadth (not model size) is directly relevant to agent infrastructure design. Combined with Google ADK 2.0's graph-workflow long-horizon execution support and Anthropic's Dynamic Workflows (1,000 subagents), a convergent picture emerges: the industry is racing to build the infrastructure for exactly what Agents-A1 shows works.
- → Scaling the Horizon, Not the Parameters: Reaching Trillion-Parameter Performance with a 35B Agent, Agents (LLM Agents)
[03]
Watch
- GPT-5.6 Sol broader rollout ~July 2: The June 2 Executive Order's 30-day federal assessment window closes around July 2. OpenAI confirmed a Cerebras launch (750 tok/sec) is also coming "in July." This is the most-watched AI event of the week. → GPT-5.6 Sol (and Terra, Luna)
- Grok 5 still absent: xAI's flagship missed its Q2 deadline (June 30 passed). July is now the stated target but no date given. 6T-parameter MoE; the monthly cadence announcement (June 28) is the only concrete signal. → xAI
- Gemini 3.5 Pro still in limited preview: GA still July target; "cleared for July launch" per TechCrunch June 29. The 2M-token context window remains the most significant gap-filler given Fable 5's ongoing suspension. → Gemini 3.5 Pro
[04]
New in Wiki
- Google ADK (Agent Development Kit) (new — agent tooling platform page; review recommended for completeness on graph workflow specs)
- Claude Science (new — Anthropic science workbench; more detail needed on specific database names and funded project outcomes)
- Scaling the Horizon, Not the Parameters: Reaching Trillion-Parameter Performance with a 35B Agent (new — Agents-A1 horizon-scaling paper; open weights available)
[05]
Updates
- Anthropic: Added Claude Science launch (June 30) to Recent Activity + Models & Products
- Google DeepMind: Added ADK 2.0 GA + Agents CLI (June 30) to Recent Activity
- Agents (LLM Agents): Added Google ADK 2.0 + Agents CLI section and Agents-A1 to Key Papers