Everything published on Agent Upstart, newest first
This page lists everything published on Agent Upstart, newest first. The newest work sits at the top, and every entry names its section.
- GitHub Agentic Workflows are now GARepo owner · GA setup · GitHub Agentic Workflows
- Codex 0.167: run parallel sessions in one placeDeveloper · supervision · Codex command centre
- Agent-desktop drives apps via the access treeDeveloper · computer use · agent-desktop
- OpenAI's Agents API: the harness is now hostedBuilder · platform read · OpenAI Agents API
- Gemini Computer Use is generally availableBuilder · API read · Gemini Computer Use
- Cursor builds live apps from cloud agentsDeveloper · launch read · Cursor Build Live Apps
- November shutdowns: Runloop, Conductor, MoondreamBuilder · shutdown moves · Runloop Conductor Moondream
- Check every package name your agent inventsDeveloper · install check · hallucinated packages
- Kilo Enterprise free for Cursor teamsTeam lead · offer read · Kilo Enterprise
- DHS DeepSeek order: inventory your stackTeam lead · compliance read · DHS DeepSeek order
- Google Jules ends Nov 11: export your workDeveloper · shutdown moves · Google Jules
- MAI-Code-1.1-Flash: local coding on RTX SparkDeveloper · hardware read · MAI-Code-1.1-Flash
- Antigravity 1.5.2 tightens headless permissionsDeveloper · changelog read · Antigravity CLI 1.5.2
- Baseten Sandboxes keep agents running for daysBuilder · launch read · Baseten Sandboxes
- Letterpress-1: a growth agent that files PRsGrowth team · product read · Letterpress-1
- Kiro CLI 3.0: V3 is the default, Classic is goneDeveloper · migration · Kiro CLI 3.0
- OpenAI: agents tried to escape their sandboxesBuilder · security read · OpenAI sandbox escapes
- Copilot Automations learn from real outcomesTeam lead · changelog read · Copilot Automations
- Pick a memory server for your coding agentDeveloper · comparison · memory MCP servers
- GitAgent: gated commit messages and PRs, locallyDeveloper · tool read · GitAgent
- GPT-6 Cyber: OpenAI's restricted security modelSecurity team · restricted access · GPT-6 Cyber
- 948 Linux exploits: what the pipeline actually didSecurity builder · pipeline read · Linux LPE exploits
- Plan allowances: what each agent plan lets you doReference list · plan allowances · agent plans
- Claude Code subagent updates now cost 99% lessClaude Code user · usage saving · subagent status
- Message Threads cut Claude Code waits by a thirdClaude Code user · faster first token · Message Threads
- METR: Anthropic's AI R&D sped up about 1.9 timesBuilder · audit read · METR R&D acceleration
- Claude Code now refuses commissioned game workGame developer · refusal boundary · Claude Code
- Snorkel: agents peeked at the reward code insteadBuilder · experiment read · Snorkel ping-or-peek
- Model specs: context, tools, and cutoffs in one tableReference list · model specs · context and cutoffs
- Review Copilot's work in a nested Claude sessionVS Code user · nested review · mixed harnesses