I'm mostly using Gemini 3.7 Flash, GPT-5.6, Fable 5 in that order. Gemini 3.7 flash for pretty much everything
GPT-5.6 for coding, review and loops
Fable 5 for most ambitious ideas (/goal)
@saboo_shubham_
-
Workflow Assigns Gemini, GPT and Claude Fable to Different Task Types
By
–
-

GPT-5.6 Writes and Claude Reviews an Overnight Hermes Integration PR
By
–
This is the WAY. Long running AI Agent harness where GPT 5.6 writes and Claude Opus 5 reviews. Asked it to add Hermes Agent support to my memory bank plugin. Woke up to a tested PR.
-

Grok Bot Workflow Migrates Six Agents and Seven Scheduled Jobs
By
–
This is what a one-person Grok bot run company looks like in 2026. Just asked it to move my FRIENDS and THE Office squad from OpenClaw and Hermes. 6 Grok bots. 7 cron jobs. 0 human employees. It figured out everything on its OWN.
-
Open Agent Stack Runs in Docker With PostgreSQL and User-Chosen Models
By
–
Yup. Great job by @ataiiam and @CopilotKit team. shouldn't be very heavy – so you can run it in your existing infra with Docker "Docker Compose brings up every part of it, the data sits in your PostgreSQL, and the model is yours to choose: no model ships in the box, and an
-
Open Desktop Tool Lets Coding Agents Train Local Models
By
–
WILD…you can now train AI models locally from a desktop app.
— Shubham Saboo (@Saboo_Shubham_) 19 août 2026
Point Claude Code or Codex at a local model with one command.
100% free and Open Source.pic.twitter.com/l2iAsvOSRaWILD…you can now train AI models locally from a desktop app. Point Claude Code or Codex at a local model with one command. 100% free and Open Source.
-

Hermes Agent Adds Context-Preserving Recurring Task Loops
By
–
Hermes Agent shipped Recurring Loops. Give it a task and it repeats. Leave the interval off and it picks its own pace. Like cron, except it runs inside your session with full context.
-

Agent Harness Uses Configurable Plugins and Built-In Model Traces
By
–

Yes. Two thing that I find very interesting in this harness that comes by default: 1. Everything is a plugin, even the harness UI. And it is editable by making changes in one config file. 2. Observability and visual traces built-in. whatever the model sees is recorded a
-
Public System Prompts Aid Debugging and May Expose Rules to Attackers
By
–
I guess this should help debug model behavior change came from a prompt update. May help attackers a little bit as they know the exact rules, so they can target one by name.
-

Anthropic Publishes Dated System Prompt Updates for Claude Models
By
–
Anthropic NOW publishes system prompts for Claude Fable 5, Opus 5, other Claude models. Available as dated release notes. You can read what was added to and removed from a production prompt.
-

DeepSeek Harness Makes Models, Tools, and Sandboxes Interchangeable
By
–
Crazy…DeepSeek Harness is already at 114k+ stars on GitHub. Everything is a plugin. Model, tools, skills, sessions, sandboxes and schedules all swap at the config layer. 100% Opensource.