To be on the right side of change, we need to remain one step ahead.
During the previous weeks' newsletters, I focused on agentic AI tools because it was evident that most hyperscalers focused on agentic AI:
Let's focus on ChatGPT Agent that was just released (July 17, 2025). Now, ChatGPT can think and act independently on its virtual computer to complete tasks from start to finish.
This feature is available now (or very soon) for OpenAI Pro, Plus, and Team users. You can turn on "agent mode" from the tools dropdown in any chat.
The agent can browse both visually and textually, run terminal commands, call APIs, and use connectors like Gmail and GitHub to gather data for multi-step workflows.
It will ask before taking important actions, and you can pause, take over, or stop it at any time.
How good is it?
Humanity's Last Exam (HLE) is a ~2,500-question, expert-level benchmark covering dozens of advanced subjects. It was built to measure advanced PhD-level knowledge in not one but multiple disciplines.
Humans are extremely bad at this test (think: <5%) because HLE hops from, say, advanced organic reaction mechanisms to abstract algebra to medical imaging physics to regulatory law. Even brilliant humans rarely train deeply across all those.
Now look at those numbers achieved by state-of-the-art AI agents:
Side note for NVidia investors: Grok 4 scaled reinforcement-learning training on Colossus (∼200K GPUs) with about 10 times (!!!) the prior compute and expanded verifiable data beyond math and coding to many domains to improve reasoning.
Native tool use allows Grok to run code, browse the web, and search deeply across X content to enhance its thinking in real time.
The SuperGrok Heavy tier costs $300 per month and gives access to Grok 4 Heavy and early features. The high cost reflects the heavy multi-agent compute requirements. ChatGPT Plus users (~$20/m) will soon have access to super PhD-level AI everywhere.
Takeaway: Agentic AI is transitioning from demos to deployable systems.
The agent costs have come down by ~10x in 1-2 weeks since the xAI launch - expect a few more orders of magnitude cost improvements during the next couple of months.
In other words, be ready to go ALL IN agentic very soon - Finxters ought to be the ones deploying an army of one billion superhuman PhD-level AI agents to solve cancer!
Be on the right side of change!