Delivery Manager @ Redberry
Tbilisi, Georgia
Joined February 2022
- Tweets325
- Following97
- Followers46
- Likes540
How do three stars move around each other? 🤯
I saw a post about a real triple-star system, found the research papers, and used ChatGPT to visualize it in 3D. You can explore it here 👇
pulsar-triple-orbit-lab.maes…
P.S. The inner orbit is enlarged for visibility.
MaestroError retweeted
Things move fast with Synapse - two more releases landed before we could publish the blog. ⚡
@RevazGh, our AI Lead, walks through v0.3.1 of our development dashboard for testing and debugging Laravel AI agents.
In full 👉 redberry.international/synap…
MaestroError retweeted
Introducing Synapse 🚀
Our open-source dashboard for Laravel AI SDK agents, created by @RevazGh. Chat with your agents, inspect tool calls, and track token usage in your browser.
Explore Synapse 👉 redberry.international/open-…
AI output needs evidence we can check.
Happy Laravel AI added source citations, and we’ve brought them to Synapse too!
Developers building agents with Laravel AI can now easily inspect sources and debug their agents’ responses.
#synapse #laravelai #laravelagents #devtool
🤖 Made with AI
What did your AI agent actually do before answering? 🤔
We built Synapse to help Laravel AI SDK developers test agents and inspect tool calls, token usage and errors.
See it in Laravel News’ walkthrough:
laravel-news.com/synapse-lar…
Automation isn't trustworthy just because it runs. I design it so a person can see what work it selected, what state changed, and who owns the next decision.
That visibility is part of automation design, not a reporting layer.
Human control should be a workflow state, not an exception.
In one session, a disclosure decision paused work. The brief, question, and source stayed intact until an operator answered.
It did not automate judgment away. It made the decision visible, inspectable, and resumable.
An AI workflow should be allowed to do nothing.
Before granting it authority, define which cases route to a person and which stop. Otherwise, it can take a confident wrong action when it meets an unsupported case.
I accidentally met an AI agent on Discord.
Mid-conversation, it dropped a suspiciously polished block about its developer experience. So I asked:
“Are you an AI agent?”
“yeah”
My guess: someone built it to find freelance leads.
Not sure whether to block it or hire it.
I’m building coworkers now 😂
I gave Jeff one post idea. He wrote the copy, designed the visual, handled feedback with me in Trello, then published it after approval.
This setup could run in Notion, Slack, Jira, or ClickUp.
The most useful AI agent is not the one that can do everything.
It’s the one that knows:
- what it owns
- when to stop
- when to ask a human
Autonomy without boundaries is just fast, confident chaos.
Building CommandsCenter has made this impossible to ignore.
The useful question is not "can AI do this?"
It is: "where is it allowed to be wrong?"
That decision defines the workflow more than the model does.
MaestroError retweeted
LarAgent is an AI agent development framework for Laravel by @RevazGh 🤖 - madewithlaravel.com/laragent
Product documentation is not a final report.
It is part of the product system.
Strong docs reduce support load, speed onboarding, and make decisions traceable.
If a feature is important enough to build, it is important enough to document so a new person can use it in 10 minutes.
Documentation is operational leverage.
#ProductManagement #Documentation #SoftwareEngineering #AIAgents