robotshelp

Every day there's new AI news. Most of it doesn't matter for your work. I pick the few things that do, and tell you in plain words what they mean for you. The useful part, without the noise. Best, Pete

Start a fresh chat for every task and your agent stays on your own AI account. OpenClaw now keeps your linked model account through new chats, resets and the daily rollover.

Choose once and stay focused on the work: OpenRouter Auto Code picks the AI model for each coding task, shows you why, and in OpenRouter's own tests kept pace with Claude Opus 5.5 at 74% less per task.

Claude Fable 5.2 is live: up to 30% faster, cheaper to build with, far better at reading charts. In Oracle's test it got the business totals right while two other models were 5 to 8 times off.

Agents that do hundreds of small tasks just got more for the cost. Meta's smallest model, Muse Ember, carries work across apps, calls tools in order, reads huge document sets plus images, audio and video, and fixes code.

Review how your team checks payments. US federal prosecutors say AI voice clones of bosses have cost businesses more than $129 million. They say scammers phone a company with an AI-generated voice that sounds like one of its own executives, and ask for money. That figure covers the period since July 2024. Think of someone in finance who gets a call from "the CEO" late on a Friday about an urgent supplier payment. The voice sounds right, so they pay. Here is my own suggestion, not the prosecutors': in your team chat, write: "No payment goes out on a phone call alone. We hang up and call back on the number in our contacts." For me the hard part is that a voice you know no longer proves who is calling. For approving a payment, which do you trust more: a call-back or a code word? I'd pick the call-back. Look out for each other, Pete

Good news for your business: Google DeepMind AI, guided by people, found two weak spots in old RSA encryption. Modern 2048 bit systems stay safe from these attacks. Ask your team one question: are we on modern encryption?

Nvidia's board is probing a possible spy plot. Here's what's actually confirmed. According to Bloomberg, the board has started an investigation into whether Jensen Huang was the target of a foreign spy plot. The board says it has no reason to believe any Nvidia secrets were compromised. The story is still developing, so we don't know the full picture yet. You'll probably hear about this at work this week, and that is all that's on the record so far. My own suggestion, not advice from Nvidia or Bloomberg: post this in your team's work chat, "If someone contacts us with an odd request, who do we tell?" For me the honest point is that nobody has said yet whether anyone was actually targeted. Does your team have one clear person to tell about an odd contact: yes or no? Take care, Pete

Your images now show up right in the terminal with Hermes Agent. You look, you decide, you keep moving. Less jumping around, more done.

Anthropic scientists used Claude to make the first 3D map of a giant bacterial protein barrel called a vault. Work that takes weeks to months by hand took days, the first model in under two days. Claude made mistakes, the people made the calls.

Stop opening every banking app and doing the math. Claude can now connect your bank and cards and tell you where your money went, what's left before payday, and flag charges that hit twice. Read only, rolling out in the US.

Artex is a warning for any team that hosts its own AI tools. Artex is free, open-source AI software that tests systems for weak spots on its own, and anyone can host it themselves. It has been linked to several cyberattacks since July. Now Censys has found more than 800 copies open to the web, with no password at all. Picture a colleague setting up a free AI tool on a company server, just to try it out. This week, try my own suggested security check, which is not an Artex setup step. Email your IT lead or post in your team's work chat: "Do any of our self-hosted AI tools open from the internet without a password?" While you're at it, ask whether multi-factor login is on everywhere. For me the lesson isn't one hacking tool, it's how easily a free tool ends up open to anyone. Does your company keep a list of the AI tools people host themselves: yes or no? Stay safe out there, Pete

Your next product idea could move a lot faster. Two OpenAI people and Astra designed a real device in under a week: the look, the case, the circuit board, the software. A factory built prototypes in under a month.

ChatGPT can now talk about your money. Connect your accounts in Finances and ask out loud what you spend and where you can save. It finds subscriptions and recurring charges, and it can't move your money. US only for now.

OpenAI launched GPT-6 Cyber, an AI that finds security holes in software before attackers do. It found 81 unknown serious holes in Firefox, and Mozilla already fixed many. It stays inside its allowed scope in 99.93% of tests.

OpenAI launched Images 2.6 in ChatGPT, Codex and the API: images come 50% faster, you can animate a product shot or brand mascot into a GIF, and characters, products and brand details stay consistent while you edit. People make 4 billion images a week with it.

Explain the big AI picture at work tomorrow: Elon Musk says future super intelligence compute will run almost entirely in space. It is a prediction, not a product.

Grok Bot has a voice chat: you talk, your Bot answers out loud and keeps listening until you hang up. Afterwards the whole call is in your chat to read, and the Bot can follow up with next steps.

Apple Watch beta: read a transcript and summary cards on your wrist. The watchOS 27.3 beta adds two tools, Live Rewind and Siri Recap. In Apple's screens, Live Rewind shows a written transcript on your wrist with an Ask Siri button under it. Siri Recap shows short summary cards stacked on the watch screen. Picture a busy day with your phone in your bag. You can read a short summary card on your wrist with a quick look instead of unlocking your phone. For now it is only a developer beta, so try it on a spare device and keep it off the phone and watch you rely on every day. I couldn't verify the setup steps, so I won't guess them. I like AI that puts a short summary on your wrist more than AI that asks you to type more. Where would you rather read a quick summary: wrist or phone? I'd pick the wrist. Listen well, Pete

Your OpenClaw work can now start by itself when a connected app reports something new, even mid meeting. Switch on the new Advisor and it can suggest a fix when your agent takes a clear wrong turn. Less waiting, more done.

DeepSeek Harness now builds your own work tools from a description. DeepSeek calls it creator mode. You describe a tool in normal words, and Harness builds it as a plugin inside the app, like a document reader or an editor. Other examples include a 3D model renderer and a video editing workbench. Say you check the same weekly numbers every Monday. You could describe a stats dashboard and use it inside the app you already work in. I couldn't verify where to start or what to type, so I won't guess the steps. I like this more than another chatbot upgrade, because the tool fits your work. What would you build first: a document reader or a numbers dashboard? Have fun making something, Pete