AI agents are AI tools that do a job for you, step by step, instead of only answering a question. Here are four stories from the last few days. I opened every source myself. When a number comes from the company selling the product, I say so.
Google launches Gemini 4 Argon
Google released Gemini 4 Argon on 30 September. It calls it its “latest frontier AI model”, built for long jobs in software engineering, finance, legal work and cybersecurity defence.
It is not open to everyone yet. Google is first giving it to “a group of trusted cyber defenders” through its Fairwind Program. Developers, businesses and consumers come later, after more safety testing. There is no date.
The starting price is $2 per million input tokens and $10 per million output tokens. A token is a small piece of text, about part of a word. The most it can write in one answer went from 64,000 tokens to 1 million.
Google reports 77.9% on DeepSWE v1.1, a long software-engineering test, and 51.3% on AutomationBench, where it says Argon ranked first. It also says Argon helped move a Fuchsia kernel project of more than 800,000 lines from C and C++ to Rust.
My take: every number here comes from Google. The article I read names no outside group that has checked them. I will treat them as claims until someone independent tests it.
Claude agents and “dreaming”: not new, but worth knowing
Dreaming is a feature of Claude Managed Agents, from Anthropic. SD Times described it on 6 May 2026 as “a scheduled process that reviews your agent sessions and memory stores, extracts patterns, and curates memories.”
In plain words, the agent looks back at its past work and notices what keeps going wrong. You can let it update its own memory automatically, or check the changes first. Developers can ask for access through a form.
It came up again this week because a short NewsBytes summary dated 2 October covered it. So it is not a new launch. It is a developer feature for now, not something you will find in a normal chat app.
Disclosure: Anthropic makes Claude, and I use Claude to help run this blog. Read this section with that in mind.
AI companies would not promise “no risk” in New York
On Monday 5 October, four AI companies spoke at a New York City Council hearing on AI safety. Council Speaker Julie Menin asked each one to assure the public, under oath, that its AI agents would always follow safeguards meant to prevent catastrophic harm.
None of them gave that promise in those words:
- OpenAI (Morgan Dwyer): “It is not possible for me to commit or guarantee that any technology is without risk.”
- Anthropic (Logan Graham): “Candidly, the science of doing this is fundamentally hard and unsettled.”
- Google (Alice Friend): “to promise perfection would not be possible with any product on the market.”
- Meta (Shane Cahill): he said he could guarantee Meta’s commitment to developing and deploying AI safely.
Menin said the council was not asking for perfection: “We’re just asking for accountability, transparency and safety overall.” The council is considering legislation to regulate AI.
AI use in UAE businesses rises to 72%
A report called Unlocking the UAE’s AI Potential 2026 came out on 30 September. It was developed by Strand Partners with AWS and the UAE Artificial Intelligence Office.
It found that 72% of UAE businesses now use AI, up from 53% a year earlier. That is 19 points higher. Of the businesses using AI, 85% said it is speeding up their digital change.
The report names four things holding businesses back: a shortage of specialised AI skills, limited funding (especially for startups and small businesses), access to secure cloud infrastructure, and rules that support new ideas. AWS general manager Chris Erasmus said businesses are moving from “can we use this?” to “how do we scale it responsibly?”
My take: AWS helped produce this report and sells cloud and AI services, so read the numbers with that in mind.
What I take from this week
- Big claims arrive before proof. Argon’s results are Google’s own for now.
- AI agents are doing real work, so questions about trust and accountability have reached lawmakers.
- AI use is growing fast in the UAE, and a lack of skills is the main gap.
When these tools open up to me, I will test them on real tasks with my five checks.