Quick tutorial connecting qbash MCP servers to Grokbot. Use these to share memories across users/sessions, distribute otherwise unavailable connectors or share deterministic AI tasks with a team spread across Grokbot, Muse, Codex, Claude, etc: youtu.be/i4emsHOFraQ
Alex Blom retweeted
Nobody owns which agents keep running or who answers when they break.
Some call that role AI ops. Most of the job is moving agent work into fixed tasks.
qbash.com/blog/who-owns-the-…
This week on qbash : Muse + Grokbot Support, a heavily improved Task Builder AI and build custom MCP servers with any Connector/Task/Memory qbash.com/blog/release-notes…
Alex Blom retweeted
As requested, I built a Jev + Claude loop in @qbashdev to conquer AI slop.
Long story short: Don't try it.
Seeing people solve problems with Grokbot + Muse this week has been fun. But what happens when they need to repeat the solve? I still lean towards agentic being used to discover the answer to a problem but if there is a pattern I move deterministic pretty quickly: qbash.com/blog/grok-bot-muse…
Quick demo using Jev inside of a @qbashdev Task. This is an easy way to use Jev for classification and other models (ie Astra, Fable, Sol) for other parts of the flow - without having to build your own bindings on infra. qbash.com/blog/use-jev-with-…
Alex Blom retweeted
I used one prompt in @qbashdev to build a test suite for our upcoming LinkedIn integration.
Alex Blom retweeted
My @qbashdev tasks work in Codex, too.
In this demo, Codex uses MCP to run my changelog and news tasks, read past newsletters, and put together a draft.
The tasks still run on qbash. Codex calls them.
Apologies for the sound. My neighbour thought it was a great idea to drill into concrete all day :)
We see deterministic tasks as enabling a new type or third rail of the application that is being developed. Thought it would be fun to share one of our internal strategy decks on the pivot and what we've learned having been iterated on what is now called @qbashdev for the last 2.5 years.
qbash.com/blog/the-strategy-…
On the AI Frontier Slowdown: The best systems we built are years old and don't really need better models: qbash.com/blog/our-most-usef…
Alex Blom retweeted
Triaging our busiest Slack channel into Asana tickets we could act on was a Monday morning job for whoever had the time.
So Jack built a Task to do it for us, reading Slack, Asana and GitHub together so status updates come from what actually shipped instead of who remembered to post about it.
Alex Blom retweeted
In my previous video, we used the @qbashdev Task builder to create a complex model testing suite.
I wanted to zoom in on the mechanics of what a Task is by creating a simplified single model version of that.
Huge!
GPT-6 Astra is now available on qbash.
Enable it in Org settings, and it's available in all of your Tasks, Prompts, Chats, and Projects.
Test it with your existing Prompts, and compare the output with Playgrounds.
qbash.com/models
Alex Blom retweeted
GPT-6 Astra is now available on qbash.
Enable it in Org settings, and it's available in all of your Tasks, Prompts, Chats, and Projects.
Test it with your existing Prompts, and compare the output with Playgrounds.
qbash.com/models
Alex Blom retweeted
Fable 5.1 and GPT-6 Astra in two days begs the question: When should you upgrade models?
I built a @qbashdev task that runs 15 business tests across 8 models every Monday, scores them, and diffs the results against last week.
Alex Blom retweeted
Upload your docs and start asking questions. Instant RAG with vector search.
qbash.com/blog/instant-rag-v…