@gregisenberg I bookmark posts for my Hermes bot. It checks all bookmarks daily and creates actionable summary, extract knowledge and proactively recommends next steps
@jankeesvw@robzolkos 1. Claude decides for you what is in your context. Pi gives you full control
2. You can run multiple models with Pi.
3. You can use a headless Pi
4. You can create Custom Plugins
Most Ruby teams shipping LLM features lack a robust evaluation layer. They eyeball a few prompts and ship.
I built RubricLLM to fix that. It's a lightweight eval framework for Ruby. This gem is provider-agnostic (via RubyLLM by @paolino ) and can be part of your CI.
https://t.co/8DwEZCL650
@bradgessler Hermes with GLM 5.1 or similar open source models. You can run those locally or via Fireworks.
If you need a SOTA orchestrator, use GPT latest model as brain and validator and cheap models as execution
Every codebase you inherit is someone else’s Horse Farm
@JohnAthayde talks about Sustainable Software Development and lessons from Permaculture
at @blueridgeruby
I kept seeing the same problem with companies trying to adopt AI:
They usually do not need another random tool. They need to know where AI would actually help first.
So I built a quick AI Maturity Audit for Crux.
It scores a company across 6 areas: strategy, adoption, governance, workflow integration, augmentation, and feedback loops. Then it gives you the strongest area, the biggest gap, and a few next actions.
Here’s a sample report I ran:
https://t.co/ejE9Km4dpb
And if you want to run it for your own team, the audit is here:
https://t.co/6b3whaG8aL
The goal is simple: move from “we should be using AI more” to “this is the workflow we should improve first.”
10 unpopular opinions about AI:
1. AI made code cheap. But it made good code matter more.
2. Most AI products are demos with a billing page.
3. Wrapping GPT is easy. Building trust around it is the hard part.
4. The real AI moat is workflow, not model choice.
5. If your AI needs perfect prompts, it doesn’t have product-market fit.
6. Most “AI automation” is just moving the human to QA.
7. Better models won’t fix a bad product thesis.
8. AI features fail when they save clicks but add doubt.
9. The best AI UX often looks boring.
10. If users have to babysit the AI, you didn’t automate anything.