The AI Times
EN
Volume I · Daily editionDaily roundup of AI newsMCPMD
EN
THE AI TIMES
Amazon · OpenAI · Meta · EU

A quiet weekend, so here are Sunday’s top stories again

Official sources at AI companies brought no news worth covering over the weekend. So here are the seven strongest stories from Sunday’s edition again. You will probably be most interested in Amazon’s program, where people with no coding skills built their own AI agents in six weeks: programs that carry out tasks on their own, step by step. OpenAI made GPT-6.1 Sol faster, but only on expensive plans. Meta is donating a thousand smart glasses to people with disabilities in Singapore, and an EU scientific panel is examining cases where AI went out of control. You will also find a business assistant that checks what you are allowed to see, simulations from NVIDIA, and Postman’s lesson on why an AI assistant with fewer choices makes fewer mistakes. At the end is a fact-check of a fake video of Noah’s ark.

Amazon · Working with AI7 October

People with no coding skills built their own AI agents in six weeks

Amazon let its sales and operations employees build AI agents: programs that carry out tasks on their own, step by step. The winning team, without a single engineer, built a financial adviser made up of five collaborating agents.

Amazon described on its blog how it asked employees who work with AI every day but do not code what was holding them back. According to the company, 90 % wanted to try building an agent themselves. They could talk about the Amazon Bedrock AgentCore platform, and 80 % had explored it, but fewer than one in five had used tools for actually building agents. They did not sign up for hackathons, competitions to build programs, because they lacked confidence alongside engineers.

The program therefore ran for six weeks at roughly four hours a week. Participants formed teams of three to four people with different levels of experience, and each team had a mentor who helped with technical issues but did not do the work for them. They wrote programs in Kiro, a tool where you describe what you want in everyday language and it prepares the application’s skeleton. According to Amazon, people gained three times as many practical skills from the program spread over weeks as from an intensive two-⁠day format.

The winning team consisted of four people in customer-facing roles who had never worked together before. Their WealthWise prototype divides financial advice among five agents: portfolio analysis, risk assessment, financial planning, market overview and investment recommendations.

After the program, 82 % of participants rated their understanding of agents as strong or expert, compared with 27 % at the start. The share ready to look for ways to use AI rose to 85 % from 41 %. These are participants’ self-assessments, however, and Amazon itself published the figures; this is no independent measurement. For everyday users, the takeaway is that tools that write code from everyday language can help build a working prototype without coding skills, given time and someone to offer advice. According to Amazon, 100 % of participants recommended the program to others.

OpenAI · ChatGPT · 8 October

GPT-6.1 Sol gets a fast Ultrafast mode, only for pricier plans

According to the Codex changelog, OpenAI added an Ultrafast mode to GPT-6.1 Sol that makes the model write responses faster. It works in Codex, OpenAI’s coding assistant, and in ChatGPT Work.

It is available only on the Pro $500 plan and selected Enterprise and Edu plans.

Meta · Smart glasses8 October

Meta to donate a thousand AI glasses to people with disabilities in Singapore

According to its announcement, Meta will donate 1,000 Ray-Ban Meta AI glasses to four Singaporean organizations, including the association for the blind SAVH and Guide Dogs Singapore. According to Meta, the glasses read text aloud, recognize objects and describe surroundings for blind and visually impaired people.

Another 30,000 dollars will fund a free course teaching people to use the glasses safely.

EU · Regulation9 October

EU scientific panel examines cases where AI went out of control

According to its announcement, the European Commission convened a special meeting of its scientific panel on AI. The panel is examining recent cases of loss of control over AI and, together with the EU AI Office, has prepared questions for the companies that developed the models involved. The Commission does not name the companies.

For users, this means the EU is investigating how safe the models behind the applications they use are. The panel consists of 60 independent experts who advise the office on issues including risks and model evaluation. Commission Vice-President Henna Virkkunen is attending the meeting. According to her, the EU has the world’s first law addressing society-wide AI risks.

Amazon · Business data7 October

Amazon’s business AI assistant checks what you can see with every query

The Amazon Quick business assistant answers questions using documents in SharePoint, Google Drive or Confluence. According to the AWS blog, it now checks directly with the source on every query whether you have access to a document, keeping anything you are not allowed to see out of the answer.

The previous approach only copied permissions periodically, so someone whose access had been revoked could still receive answers from documents they were no longer allowed to see for some time. The assistant first selects relevant passages based on stored permissions, then verifies them directly with Google Drive. Mondelēz deployed Amazon Quick for more than 35,000 employees across four regions.

NVIDIA · Simulations8 October

NVIDIA shows simulations built by an AI agent from everyday instructions

According to the company’s blog, NVIDIA staff gave AI agents tasks in everyday language, and the agents used GPT-6 Astra to build simulations of robots, cars and a space station from those instructions. This has no direct impact on the applications you use, but shows how complex the software AI can now build from verbal instructions is.

In the Robo Olympics project, a simulated Unitree G1 humanoid robot cleared an obstacle in 64 out of 100 attempts. In another project, agents using the Astra and Claude Fable 5 models spent roughly three days comparing simulated cameras and laser sensors with real recordings.

What experts say

Simon Willison

developer, creator of the LLM and Datasette tools

Anti-Patterns in Software Blogging

Willison agrees with Michael Lynch that technical writing should make sense even to a reader who does not click any links, and that people should write as they speak. He admits that he often uses links instead of explaining terms himself. He sees Lynch’s warning as crucial: when developers hand writing over to AI, their blogs become increasingly bland and alike, while readers are looking for writing with personality.

simonwillison.net
Postman · AI agents9 October

Postman finds that an AI agent with fewer tools makes fewer mistakes

Postman, an application for testing connections between programs used by 40 million developers according to the AWS blog, added Agent Mode, in which AI works independently within the application. The team expected model quality to be the biggest challenge. But a greater problem was that the agent had too many tools to choose from: individual actions it is allowed to perform in the application. For everyday users, the takeaway is that an AI assistant with fewer actions to choose from tends to be more reliable.

When the agent could see more than roughly 40 tools, it more often chose the wrong one, called tools that did not exist or sent them incorrect data. Today, from more than 170 tools, the main agent selects roughly 15 that fit the query and passes the task to an assistant that can see only those tools.

Before the agent changes anything in the application, the user must approve it. Postman can also remove personal data before it reaches the model, if the company’s administrator enables this. Depending on the type of work, the agent can switch between faster and more capable Claude models running on Amazon Bedrock.

VerifiedMyth of the day

“A video filmed from a helicopter shows excavations of Noah’s ark on Mount Ararat.”
False

An account on X posted the video on 5 October, and according to Lead Stories, it was created by AI. The Hive Moderation detector identified it as AI-generated with a probability of 99.9 %, and OpenAI Verify found an invisible SynthID watermark in the footage indicating that the video was made using OpenAI tools. The location in the video also does not resemble the Durupınar site that ark hunters are actually investigating.

Verified by: leadstories.com

Events in CzechiaNext 14 days · until 25 October

  1. Computer Science Evening: Language models – the foundation of today’s AI

    A lecture in the Computer Science Evenings series on the neural language models behind today’s generative AI.

    LecturePrahaprice on website
  2. Courseonlinefree
  3. ConferencePrahaprice on website
  4. AI HotSpot

    AI HotSpot 2026 – an overview and networking for AI-⁠led meetings.

    ConferencePrahaprice on website
  5. Rakathon 2026

    Help change the future of cancer care with AI!

    Česká asociace AI

    Prahaprice on website
  6. AI in culture: Why and at what cost?

    A conference on practical uses of artificial intelligence in museums and cultural institutions.

    ConferencePrahaprice on website
  7. Sharing experience for CMOs: AI in B2B marketing – smart ways to fill your pipeline with quality leads

    MeetupBrnofree
  8. Automation & AI Clinic

    An event for companies that want to discuss a specific automation, digitalization or AI problem directly with specialists in small expert zones.

    WorkshopPrahafree

Model leaderboardIntelligence Index · top 10

Which models are the smartest today

Artificial Analysis combines several independent benchmarks into one number. A higher score means the model handles harder tasks.

  1. 58Claude Opus 5.5
  2. 56Claude Sonnet 5.5
  3. 53Claude Fable 5.1
  4. 53GPT-⁠6 Astra
  5. 53Gemini 4 Argon
  6. 52GPT-⁠6.1 Sol
  7. 51Claude Opus 5
  8. 50Claude Fable 5
  9. 48Muse Spark 1.3
  10. 48GPT-⁠6 Sol

Cost per completed task

  1. Claude Opus 5.55,98 $
  2. Claude Sonnet 5.55,46 $
  3. Claude Fable 5.17,63 $
  4. GPT-⁠6 Astra3,26 $
  5. Gemini 4 Argon1,99 $
  6. GPT-⁠6.1 Sol0,72 $
  7. Muse Spark 1.31,60 $
  8. Grok 4.73,74 $
  9. MiMo-⁠V2.6-⁠Pro0,13 $
  10. Qwen3.8 Max5,41 $

Best image models

  1. GPT Image 2.5 Sunburst (max)1198
  2. GPT Image 2.5 Flare (max)1191
  3. GPT Image 2 (high)1172
  4. Nano Banana 2.11160
  5. Grok Imagine Image 2.01155
  6. MAI-⁠Image-⁠2.61151
  7. Nano Banana 2 (Gemini 3.1 Flash Image)1125
  8. FLUX 3 Image1116
  9. Muse Image1114
  10. GPT Image 1.5 (high)1108

Data: Artificial Analysis · Intelligence Index, higher is better, cost per completed task (Intelligence Index task average) and Text to Image Arena · as of Oct 12, 2026