Five minutes into a fresh session, with the reset timer showing 4 hours 55 minutes remaining, 28% of my session allowance had already been consumed. Claude Code sessions had been cleared for the previous 24 hours, Cowork was disabled, Claude in Chrome was off, and I was on a single browser and device. Over the following four idle minutes, with no prompt sent, the figure rose to 32%. This pattern has held for the past 24 hours, with the session allowance reaching 100% within roughly an hour on several occasions without matching use on my side. I have been on the platform daily for more than four months and have not encountered this before. Anthropic published a postmortem in April 2026 concerning a caching fault that produced a similar pattern, so this seems worth reporting rather than absorbing. If you would like to check your own account: open Settings → Usage at the start of a session, note both the percentage and the reset timer, remain idle for five minutes, then check again. I would welcome your figures in the comments.

Quick share. We manage multiple client accounts, and this kept killing us: the AI hits a limit, falls back to some freemium model, and suddenly the output isn't something you'd want to send a client. Happened way too often. Here's what fixed it for us: OmniRoute. What you actually get with it, in simple terms: Your AI never stops working. When one model runs out, it moves to the next one automatically — one that YOU picked, so the quality never drops. Everything sits on one screen so you can see what's running at all times. And it trims wasted usage, so your costs come down too. We've got all our models connected through it, including Kimi from OpenRouter, plus our Hermes agents. Screenshot of our dashboard attached so you can see it in action. It's free and open source. Set it up for your own agents here: https://github.com/diegosouzapw/OmniRoute If you try it, report back here. Already using it? Tell me what you think — would love to compare notes.

Quick update for the community: in my last post I mentioned I was planning something with Fable 5. That plan is now in motion, and honestly, the results are next level. I handed it a seriously complex task — multi-step, the kind of thing that usually breaks other models — and it just… handled it. I genuinely don't think any other AI model could have done this end-to-end. Discussion for the group: 1️⃣ Who here is using Fable 5 in Claude Code? 2️⃣ What's the most valuable thing it's done for you so far — a workflow, an automation, a build? Share below — let's learn from each other's use cases. 👇 Meanwhile, I'm off to power up a client's business with Claude Code with Fable 5 AI Model, and Hermes AI. Full breakdown coming in the next video. 🚀

Quick story: Anthropic released Claude Fable 5 (their most advanced public model) a few days back. I jumped in early and pointed it at my own web app, speka.me — ran a full vulnerability pass on it — and it genuinely crushed it. I burned through about 4–6 hours of usage, hit a 2-day cooldown, and planned to resume Monday. Woke up and the model was completely gone. Turns out the U.S. government issued an export control directive (national security grounds) and Anthropic had to pull Fable 5 + Mythos 5 for ALL customers to comply. So I basically got a tiny window to stress-test speka.me with one of the most capable models ever shipped before it got yanked. Some of my clients were lining up to try it and missed the window entirely. Anthropic says they think it's a misunderstanding and want to restore access — so maybe not the end. 💬 Did anyone here get hands-on with Fable 5 before it went offline? What did you build or test with it? Share below.

Agentic OS was supposed to run multiple AI agents from one dashboard — but it was barely functional. I debugged it, patched it, and got it running cleanly on a single Hostinger VPS. Now three agents live under one Mission Control: → Claude for deep research (Claude Code, full tool use + MCPs) → OpenClaw for outreach and lead gen → Hermes for SEO and content Live status, latency, and heartbeat for every agent. Everything grounded in an Obsidian vault and saved as markdown. No lock-in. My client and I went from juggling three separate tools to managing one — and they've been running it daily for a week or two now. From broken to genuinely useful. The single-chat-window era is ending. The orchestration era is just getting started. What would you wire up first? #AI #AgenticAI #BuildInPublic #Claude #Hostinger