Aaaand we’re back …

Good morning.

Predicting something almost a year out and seeing it become reality must be one of the best feelings in the world.

In this very newsletter I predicted a couple of things :

Coming back from a very nice holiday and opening up the laptop it looks like I stepped into the future as I saw the news of the release of Grok Bot.

Grok Bot is xAI’s new “always‑on” AI agent product, built on top of the Grok models and launched in beta on August 11, 2026. Instead of being just another chat window, it is positioned as a team of “AI teammates” that run on a persistent cloud computer, log into the tools you already use and keep working even after you close your laptop .

Crucially, Grok Bot is a separate product from the regular Grok chatbot that lives on X and on grok.com; you can use one without automatically having the other. The classic Grok assistant is mostly conversational-answering questions, searching the web, generating code, images and text-while Grok Bot is designed to actually complete work inside inboxes, CRMs, dashboards and other web apps, only coming back to you when a decision or approval is needed

Under the hood, each Bot is a persistent, named agent that runs on a user‑scoped cloud VM with its own browser, filesystem and terminal. All Bots on your account share that same computer, including its files and logged‑in sessions, which makes hand‑offs easy and allows multiple Bots to collaborate without constantly re‑authenticating. You interact with a Bot by messaging it like a colleague, giving it a role and a task; it can then learn multi‑step workflows from live demonstrations, store them as routines and re‑run them on a schedule or on demand, continuing to execute even when your local device is offline.

xAI says Grok Bot grew out of internal prototypes that teams used for sales outbound, marketing campaigns, office operations and bug‑fixes, and external coverage adds examples like inbox triage, CRM hygiene, research and scheduling. The design goal is to avoid complex workflow builders and API wiring: instead, the Bot operates like a human RPA worker, clicking through real web interfaces to get to a finished outcome in the target tool rather than leaving work as drafts in a chat.

If you want to see the future of work - this is it. Because this is the first step towards fully AI-automated companies. If you look at how this works, then WE ARE NOT THAT FAR OFF.

The only role that needs replacing now is the one of the coordinator sitting in the middle and telling the agents WHAT to do .

At launch, Grok Bot is in early beta and only available on paid tiers (of course) : SuperGrok Heavy (around 300 dollars per month), Cursor Ultra (around 200 dollars per month) and Cursor Premium Teams (around 120 dollars per seat per month).

So there you have it. It’s not the the end of the road yet for the knowledge worker, but I can spot the edge of the ravine from where I’m standing already.

Welcome to the Blacklynx Brief

Autonomous AI agents breach Taiwan's government and nuclear networks
Suspected China-linked operators turned two open-source AI agents into a near-autonomous hacking tool that broke into Taiwanese government systems over four days, then spread to the island's nuclear safety agency and seven energy firms. The intrusion compromised at least 85 accounts and lifted more than 2,500 personnel records, and researchers call it the first publicly documented autonomous AI attack on a government target. (CyberScoop)

OpenAI locks its frontier cyber models behind vetted access
OpenAI launched Daybreak, restricting its most capable cyber models to approved defenders inside AWS environments for authorized vulnerability research and security testing. The move concedes that the same tools that speed up defense also lower the cost of offensive hacking. (OpenAI)

EU AI Act transparency rules become enforceable
As of August 2, the European Commission's AI Office began enforcing transparency obligations requiring chatbots to disclose they are AI, deepfakes to be labeled, and AI-generated content to carry machine-readable marks. Non-compliance can trigger fines up to €15 million or 3% of global turnover. (European Commission)

Alibaba open-weights its flagship Qwen3.8-Max
Alibaba is releasing open weights for Qwen3.8-Max, a 2.4-trillion-parameter mixture-of-experts model (95B active per token) — the first time it has open-sourced a top-tier "Max" model. Alibaba's own benchmarks put it ahead of GPT-5.6 and Claude Fable 5 on agentic tasks while matching US closed-model pricing. (South China Morning Post)

OpenAI starts putting ads in ChatGPT
OpenAI began testing advertising for logged-in adult users on ChatGPT's Free and Go tiers across the US, UK, Mexico, Brazil, Japan, and South Korea, with paid tiers staying ad-free. It is the clearest sign yet that consumer AI is shifting toward an ad-supported business model. (OpenAI)

That’s it for us this week. Please like and subscribe 🙂