What shipped in agentic AI, curated with a builder's note on each. Every entry is selected, verified against the primary source, and written by Jigar, not auto-generated. The fastest takes go out in the WhatsApp community first; this page is the archive.
Google shuts down antigravity-preview-05-2026; pin Gemini managed agents to the 09-2026 agent ID
The Gemini API deprecations table lists `antigravity-preview-05-2026` as shut down on October 5, 2026. The replacement is `antigravity-preview-09-2026`, released September 17. Remote-sandbox callers that only read `output_text` or `model_output` can swap the agent string. Teams that run `local_environment` or parse `function_call` steps must migrate tool names and argument shapes: file writes move from full-file `write_file` to `write_to_file` and line-range `replace_file_content`, reads become `view_file`, and parameters use PascalCase. Inventory every hardcoded 05-2026 ID and local tool parser today, then re-run sandbox evals on 09-2026 before traffic fails: /blog/gemini-antigravity-05-shutdown-managed-agent-cutover-october-2026.
via GoogleDetails
Tools October 5, 2026
GitHub secret scanning adds Pydantic AI gateway, Lovable, and Supabase credential detectors
GitHub now detects `pydantic_ai_gateway_api_key`, `logfire_token`, `lovable_api_key`, `supabase_oauth_access_token`, and `supabase_scoped_personal_access_token`. Lovable Labs joined the partner program, so public-repo `lovable_api_key` hits can be forwarded for revocation. Agent platform teams using Pydantic AI gateways, Lovable, or Supabase should confirm secret scanning is on, rotate any keys already in git history, and keep gateway tokens out of agent transcripts and MCP configs: /blog/gemini-antigravity-05-shutdown-managed-agent-cutover-october-2026 and /blog/agent-eval-sandbox-escape-checklist-july-2026.
via GitHubDetails
Open Source October 3, 2026
Aleph Alpha open-weights Kolibri-1, a 78B DE/EN MoE for on-prem agent and RAG workloads
Kolibri-1 (`Aleph-Alpha/Kolibri-1`) is Apache 2.0 on Hugging Face: 78B total parameters, 3.46B active per token, explicit reasoning effort, tool calling, June 18, 2026 knowledge cutoff, and a native 262,144-token context with quality checks reported to 1,048,576 tokens. Aleph Alpha recommends serving at or below 256K for latency-sensitive and complex tasks. Official card scores include LiveCodeBench v6 85.9, SWE-Bench Verified 66.4, and BFCL v4 overall 61.4 at high reasoning. Serve with `aleph-alpha-inference` and vLLM using the `kolibri1` reasoning and tool-call parsers. EU and DE/EN regulated teams should bench it as an on-prem worker or RAG agent, not as a drop-in GPT-6.1 Sol or Opus 5.5 planner: /blog/aleph-alpha-kolibri-1-eu-open-weight-agent-routing-october-2026.
via Aleph AlphaDetails
Tools October 2, 2026
GitHub Copilot retires Gemini 3.5/3.6 Flash, Kimi K2.7 Code, and Claude Opus 4.7
via GitHubDetails
Tools October 1, 2026
GitHub Copilot computer use enters public preview in CLI and the Copilot app
via GitHubDetails
OpenAI September 29, 2026
OpenAI ships GPT-6.1 Sol at near-Astra coding quality and one-fifth Astra list token prices
via OpenAIDetails
Tools September 25, 2026
GitHub Copilot adds Claude Opus 5.5, GPT-6 Sol, GPT-6 Luna, and Grok 4.7
via GitHubDetails
Claude September 22, 2026
Anthropic releases Claude Opus 5.5 as a cheaper Fable-class coding and knowledge-work model
via AnthropicDetails
OpenAI September 22, 2026
OpenAI expands GPT-6 with Sol and Luna at 50% lower list prices than GPT-5.6 promo rates
via OpenAIDetails
Tools September 21, 2026
xAI and Cursor ship Grok 4.7 for long-running coding agents at Grok 4.6 prices
via xAIDetails
Open Source September 20, 2026
Alibaba launches Qwen3.8-Omni-Flash for native omnimodal agent workflows
via AlibabaDetails
Tools September 17, 2026
Anthropic redesigns Claude Code Projects around a coordinator and parallel cloud threads
via AnthropicDetails
Live first
Get these updates as they break
The fastest takes land in the WhatsApp community before they reach this archive.