For 3 years I've been the only non-engineering hire at Friday AI. So I built a marketing stack with our own product that runs most of it for me. It's still a work in progress, but here's iteration one: https://t.co/UNof82v6Op
The AI conversation has been obsessed with models. The other half of the system is the harness.
I wrote up where that conversation stands in 2026, what good harness design actually requires, and where it's headed.
https://t.co/9gzeswH9pn
First-class support for human in the loop elicitations landed in the latest Friday Studio! https://t.co/zk9F6nBZIw
Get started with a simple prompt like: "Create a workspace that reviews my email inbox with human-in-the-loop approval. Show me each email, ask what to do with it (reply, archive, delete, snooze, skip), then execute the approved actions."
Work through a few rounds so it learns your preferences, then set it up for autopilot and finally get to inbox zero.
Join us on Discord if you have a HITL use case you want to setup! https://t.co/0edlyX1GPo
#hitl #agents #ai #openclaw
@upen946 https://t.co/Ks1U3nMozH
We're a more reliable, consistent, and easier to set up alternative to OpenClaw. It's config driven and observable for technical users, but has chat-based features for ease of use.
@Hesamation My read is that it promised a lot but lack of consistent, reliable results, difficulty in setup, and major security holes made it lose its luster.
Shameless plug, we built https://t.co/gzMwrynASf to help address these issues. We launched 11 days ago.
@DanielSmidstrup Our team is building @HelloFridayAI. It's an AI agentic runtime that we think is a more reliable, consistent, and easy to set up than OpenClaw. https://t.co/gzMwrynASf
@mreflow Give @HelloFridayAI a try (Disclaimer: I work there). We built this with reliability as our #1 goal, its config driven which ensures consistent results. Locally hosted and source available.
What makes Friday Studio different from OpenClaw:
- Workflows as versioned config
- Isolated creds never written to config or logs
- Skills are .md, not executable code
- Every agent run is inspectable step by step
- Shareable and importable
https://t.co/SEAk7pvxJW