MTB

Exploring the edges of AI-assisted dev workflows • Prototyping faster than I should • Building bots, dashboards, and weird little systems that work

[MTB @mtallentb]

@NESpower I know y’all are working hard and this storm just blew through. Any idea on timelines for Green Hills/Lipscomb area?

[MTB @mtallentb]

@itsnoahd Hey Noah! Appreciate the work y'all are doing to get things back up. I know there's no ETA but do you anticipate any data loss from this outage for postgres DBs and things of that nature?

[MTB @mtallentb]

@om_patel5 it shows: https://t.co/uUaXHvAEb5

[Eyal Toledano @EyalToledano]

we've shipped taskmaster v0.23 🚀

→ official @taskmasterai @code extension
→ manage your taskmaster workflow visually
→ introducing scope-up & scope-down tools
→ dozens of QOL improvements & fixes
→ we're hiring 🐹

follow + retweet + bookmark
👀👇

[Eyal Toledano @EyalToledano]

we've shipped taskmaster v0.22 🚀

→ @kirodotdev hooks like:
→ hook for task tracker
→ hook to link tasks to commits
→ set task to complete when tests pass
→ gemini cli upgrade
→ better tag handling
→ token limit improvements

follow + retweet + bookmark
👀👇

[MTB @mtallentb]

@jackfriks build* the rocking chair

[MTB @mtallentb]

@EyalToledano @geminicli @claude_code @grok @GroqInc 🔥

[Eyal Toledano @EyalToledano]

we've shipped taskmaster v0.20 🚀

→ mcp sampling support (zero api keys)
→ @geminicli integration & provider
→ advanced @claude_code rules
→ language override
→ @grok 4 support
→ @GroqInc support
→ parse-prd auto-selects task number
& more

follow + bookmark + vibe
👀👇

[Alex Albert @alexalbert__]

We’ve rolled out another update to Claude Code to help customize your workflows: Hooks.

[Claude Code Community @claude_code]

Treasure Trove of Actionable Information for Multi-agents.

Tip for @ claude_code users: Launch subagents by requesting it.

"Explore codbase with 3 parallel tasks. Each agent handles a different stack ( frontend, services, database ) and updates the readme README."

---

Key points from blog:

1. Multi-agent systems => more token usage => improving performance.
a. Claude Opus 4 as lead + Sonnet subagents outperformed single-agent Claude Opus 4 by 90.2% on internal research eval.
b. Token usage explains 80% of performance variance in BrowseComp evaluation, with tool calls and model choice as other factors.

2. Evaluations: ~20 samples with LLM-as-a-judge (0.0-1.0 scoring, pass-fail grade) are scalable.
a. Start with ~20 queries reflecting real usage patterns for small-scale testing.
b. Single LLM call with one prompt, outputting 0.0-1.0 scores and pass-fail grade, aligns best with human judgments.

[MTB @mtallentb]

one of the under-appreciated aspects of all these AI dev tools is the ability to write a quality commit message

[MTB @mtallentb]

@wesbos https://t.co/4yUPQoY37h

[MTB @mtallentb]

I built a dev workflow that feels like I'm managing a team of tireless engineers powered by AI.
No backtracking. No confusion. Just smooth planning → execution → shipping.
Here's how I use @cursor_ai + @taskmaster + @allhands_ai to get it done: 🧵

[Tom @tomcrawshaw01]

This is the holy grail for building n8n workflows with Claude...

Just discovered n8n-MCP and it is literally game-changing...

It gives Claude deep knowledge of all 525+ n8n nodes instantly.

Your AI goes from guessing to deeply understanding how n8n works.

Video setup guide + git repo below 👇

[Kevin Kern @kevinkern]

Cursor is now available on your phone.

- Create many tasks in parallel
- Choose your project and model
- View and manage your tasklist
- Create follow-ups
- Create PR and let Bugbot Review them

When you get home, continue working directly in Cursor

[Visual Studio Code @code]

VS Code: the open source AI editor. Starting today.

[MTB @mtallentb]

@happybydefault @claude_code @OpenAI @allhands_ai All that's to say I was just surprised to see such a difference using the same tool with different API keys.

[MTB @mtallentb]

The difference between using @claude_code and @OpenAI to review PRs with @allhands_ai is... noticeable 😅

[MTB @mtallentb]

@happybydefault @claude_code @OpenAI @allhands_ai I agree with that for the formatting. I would still still point to the depth of the review. The 2 Claude screenshots are slices of a much longer comment, while the OpenAI review is basically "did this -- this ensures best practices".

[MTB @mtallentb]

claude: