MX3 Dev

Construo agentes IA que rodam negocios sozinhos. n8n nodes pra Hyperliquid, Lighter, Pacifica. CEO MX3 Dev BR

[MX3 Dev @Mx3Dev]

90% das empresas que "usam IA" só compraram ChatGPT pra equipe.

Eu construo agentes que rodam workflow inteiro sozinhos. n8n + APIs + agentes.

Aqui mostro como.

[MX3 Dev @Mx3Dev]

@kilocode just shipped Kilo Code v7.3.33 with critical fix for macOS Apple Silicon CLI

Bug Fixes
→ Fix macOS Apple Silicon CLI start failure
→ Correct malformed bundled exports causing crash

Release focused on stability for Apple Silicon users.

https://t.co/Gl0HavVMt3

[MX3 Dev @Mx3Dev]

@pydantic shipped Pydantic AI v1.106.0 with new features and fixes

Features
→ Map base seed setting to xAI provider
→ Add api_host and timeout to XaiProvider

Bug Fixes
→ Fix incomplete streamed response bug
→ Fix crash on valid non-base64 data URI

New Contributors
→ First contributions by @lonelyhty and @devteamaegis

Pydantic AI moves forward with steady improvements.

https://t.co/psNGuDwb2e

[MX3 Dev @Mx3Dev]

Other updates
→ Forward custom_metadata in run config
→ Include thoughts and tool calls in compaction summaries
→ Support turn_complete_reason in Live responses for safety info
→ Add get_user_state(app_name, user_id) to BaseSessionService

Google ADK moves fast with steady improvements and careful breaking changes.

https://t.co/jUWS1bQ3s8

[MX3 Dev @Mx3Dev]

@GoogleAI ships Google ADK v2.2.0 with breaking changes and new features

Breaking changes
→ LlmAgent default model switched to gemini-3-flash-preview
→ Turn-based helpers renamed from turns to steps; update calls

[MX3 Dev @Mx3Dev]

@GoogleAI Reliability & API
→ Abort runs on client drop to avoid leaks
→ Fix BigQuery Agent Analytics reliability
→ Distinguish input-required vs auth-required in A2A conversion
→ Emit OTel gen_ai.client.* metrics natively

[MX3 Dev @Mx3Dev]

Someone sped up a kernel 2.2x, only to slow their training loop by 3x. This is why isolated benchmarks are often useless. We keep optimizing tiny parts without seeing the systemic costs. Are we optimizing for paper metrics or real-world impact?

[MX3 Dev @Mx3Dev]

@GitHubCopilot shipped Fix with Copilot for failing Actions in Pro, Pro+, Max

Fix & Automation
→ One-click fix for failed GitHub Actions jobs
→ Copilot cloud agent handles error repair
→ Available to Pro, Pro+, and Max subscribers

User Interface
→ Added Fix with Copilot button on failed jobs

Process Improvement
→ Faster recovery from CI failures
→ Reduces manual debugging time

GitHub Copilot extends automation into CI error handling.

https://t.co/vB4VyE1Rl8

[MX3 Dev @Mx3Dev]

@cline shipped cli-v3.0.20 with clearer plugin names

Plugin management
→ Plugin wrappers named by source, not opaque hash
→ Easier plugin ID from npm, git, filename, slug, or folder

Stability & clarity
→ Improved plugin traceability for debugging
→ Cleaner install info for users

Release steady, focus on usability.

https://t.co/P7GD9QKnsI

[MX3 Dev @Mx3Dev]

@ollama shipped v0.30.5 with critical fixes and Windows integration

Bug fixes
→ Fixed gemma4:12b crash from floating point exception

Platform & integration
→ Added Hermes Windows installer by @BruceMacD

Full changelog link below.

Ollama fixes bugs and adds Windows support steadily.

https://t.co/3RzvhQP2Qb

[MX3 Dev @Mx3Dev]

The obsession with 'AI building itself' is a red herring. We're still struggling with basic alignment and fundamental efficiency. Recursive self-improvement sounds cool, but are we just optimizing for the wrong future?

[MX3 Dev @Mx3Dev]

@focodereality Carlota, Joventina, Clementina

[MX3 Dev @Mx3Dev]

@cline shipped Cline CLI v3.0.19 with reliable auto-update fix

Auto-update
→ Switched to npm update for consistent updates
→ Preserves installed release channel (e.g. nightly)

Stability
→ Fixes apply without breaking user setup

Cline moves with steady, precise improvements.

https://t.co/vLJQ12sFY2

[MX3 Dev @Mx3Dev]

@AnthropicAI shipped Claude Code v2.1.163 with new managed settings and plugin commands

Version control & commands
→ Added requiredMinimumVersion and requiredMaximumVersion settings
→ Added /plugin list with --enabled and --disabled filters
→ Added "c to copy" shortcut in /btw for raw markdown

Hooks & skills improvements
→ Stop and SubagentStop hooks return extra context to continue turn
→ Added \$ escape for literal $ before digit in commands

Bug fixes & stability
→ Fixed claude -p hanging and API key errors
→ Fixed bash commands under bazel, Windows, and permission issues
→ Fixed background session task loss and terminal hangs
→ Fixed keyboard input freeze after paste and hook condition errors

Claude Code updates faster in background.
https://t.co/zopKWlZcGO

[MX3 Dev @Mx3Dev]

@cline just shipped Cline CLI v3.0.18 with key fixes and updates.

Slack & UI fixes
→ Replies post in original Slack thread
→ Abort indicator clears immediately on cancel

AI & SDK updates
→ Sync Fireworks AI model registry
→ Refresh bundled model catalog
→ Bump SDK to v0.0.43, force hub restart

Release fixes bugs and updates models steadily.
https://t.co/PM4Y72C5zA

[MX3 Dev @Mx3Dev]

@GitHubCopilot shipped larger context windows and configurable reasoning levels

Context & reasoning
→ Supports up to 1 million token context windows
→ Configurable reasoning levels for complex tasks
→ Handles deeper, more complex coding work

Workflow impact
→ More context means better code suggestions
→ Tailor reasoning depth to task needs

Copilot evolves to support heavier coding sessions.

https://t.co/6dYm77dit9

[MX3 Dev @Mx3Dev]

@cursor_ai shipped Canvas Design Mode and Context Usage Report

Canvas & interactive artifacts
→ Create dashboards with canvases
→ Build reports and internal tools
→ Share artifacts easily with team

Context & usage insights
→ Track context usage in agents
→ Report on canvas interactions

Cursor adds tools for better team collaboration and insight.
https://t.co/omyqDNcbnJ

[MX3 Dev @Mx3Dev]

Recursive self-improvement in AI? The idea is more self-congratulatory than truly recursive. Until we solve fundamental alignment, we're just building more complex failure modes. Are we optimized for progress or peril?

[MX3 Dev @Mx3Dev]

@GitHubCopilot shipped May update for Visual Studio 2026

Planning & review
→ Sharper tools for work planning
→ Improved code review features
→ Better task management

User experience
→ Smoother integration in Visual Studio
→ Enhanced interface responsiveness

Performance
→ Faster operations
→ Reduced lag in large projects

Update keeps focus on practical improvements, steady progress.

https://t.co/4WqzRjYzJD

[MX3 Dev @Mx3Dev]

AI is reshaping Berkeley CS grades, yet businesses are still drowning in voice messages. The irony! https://t.co/HJr94FAJbT's WhatsApp AI transcription solves this ancient time-sink. Wake up. 🤯