Microsoft's New AI Mode Will Handle Your Invoices Without Being Asked

Today's issue is about AI trying to stop being a tool you open and start being something that just runs in the background of your business, whether you asked it to or not. Microsoft wants a standing AI employee inside your Office apps. Anthropic wants to sell you an entire app store of AI add-ons. Google wants its AI answering your phone. None of that is bad by default, but it only works out well if you're the one deciding what gets access, not the vendor's default setting. Here's what actually happened, and what's worth double-checking before you say yes.

When analytics on Postgres slows down, most teams add a second database. Then come the pipelines, the sync jobs, and a copy of your data that's always a little behind.

TimescaleDB takes a different approach: extend Postgres instead of splitting away from it. Hypertables partition your data automatically as volume grows. Hypercore compression cuts storage up to 95%. Continuous aggregates keep dashboards live without re-querying everything.

CERN runs Postgres this way for sensor data from the Large Hadron Collider.

No split architecture, no pipeline lag, no new query language to learn. Same SQL, same drivers, same tools.

Start on Tiger Cloud and get $1000 in credits.

Get $1000 Credit

What happened: Microsoft relaunched Copilot this week with three modes, according to Microsoft's own announcement. Home mode chats, delegates tasks, and works inside your Office documents. Code mode builds apps and dashboards from a plain description. Autopilot is the one worth noticing: a persistent agent that keeps working on a standing job, like managing invoices, without you opening a new chat each time. CEO Satya Nadella called it an AI "chief of staff," and Microsoft says it already has more than 30 million paid enterprise Copilot users.

Why it matters to you: If your business runs on Microsoft 365, this is a preview of AI that handles recurring back-office work in the background instead of waiting for you to ask every time.

What to do about it: Before you turn on Autopilot for anything, check exactly what accounts and files it can touch. The whole point of it is that it keeps working while you're not watching.

What happened: Anthropic launched Claude Marketplace this week, BleepingComputer reported, a hub inside Claude where businesses can browse and install more than 2,000 third-party connectors, plugins, AI agents, and consulting services. Companies can spend AI budget they've already committed to Anthropic on these partner tools instead of buying them separately. It's a clear shift from "Claude is a chatbot" to "Claude is a platform," the same playbook Salesforce and Microsoft have run with their own app stores.

Why it matters to you: More AI tools for things like bookkeeping, scheduling, or customer support may get easier to find and turn on, all inside a product you're already paying for.

What to do about it: A marketplace with 2,000-plus listings means some of them are going to be low quality or poorly secured. Vet a plugin the same way you'd vet any app before you connect it to your business data.

Here's what's been bugging me since last January. I skipped the first Epic Marketing Summit, and for months afterward I kept running into people who'd been there, talking about the funnel they built or the app they shipped that weekend. I'm not skipping this one. It's two days in Miami, and the whole structure is built around finishing something instead of collecting more notes: you leave with a working AI powered sales funnel already wired up for your own business, and a second AI app you build yourself from scratch, hands-on vibe coding the whole time. The trainers aren't theory people either, I know several of them personally, and every one has actually shipped the thing they're teaching.

Last January's event sold out before most people even heard about it, and this round goes even deeper on the AI build side. If your sales process is still running on your own time and energy instead of an actual system, January is when I'd start fixing that. I've already bought my seat. Grab your seat here before it's gone again.

What happened: Google is piloting a Gemini feature called "Call for Me," according to TechCrunch, currently limited to Pixel 11 owners on a paid plan in the US. Ask it to book a table, check if something's in stock, or change a plan, and Gemini calls the business, announces upfront that it's an AI, navigates the phone menu, sits on hold, and hands you a live transcript you can jump into at any point. It's a revival of Google's old "Duplex" demo from 2018, and Google is racing a competing agent from Meta, called Muse, to the same idea.

Why it matters to you: Your business phone line may start getting calls from an AI assistant instead of the customer themselves, which makes it worth checking that your hold menu and voicemail are easy for a bot, and a human, to actually get through.

What happened: Google's first Project Suncatcher test satellite launches October 1st on a SpaceX Falcon 9, according to Google, carrying four of its Trillium AI chips. The mission is simply to see whether the chips survive the shaking of launch, cosmic radiation, and running with no air to cool them. If it works, Google's longer-term bet is solar-powered AI data centers in orbit, where the same solar panels could generate up to 8 times more power than they do on the ground.

Why it matters to you: This one's a watch-don't-act story. It's a multi-year infrastructure bet with no effect on the AI tools you're using this year.

What happened: A UNESCO and ICOM survey of more than 400 museums across 90 countries, UNESCO reported, found 57% are already using AI, mostly staff quietly adopting it on their own for admin work, communications, and research. The catch: 55% of those museums have no AI policy at all, and only 17% train staff on how to use it responsibly. Their top worries were factual errors, copyright, data privacy, and bias.

Why it matters to you: If a museum with a board and a mission statement hasn't written down any AI rules, there's a good chance your business hasn't either.

What to do about it: Spend an hour writing one page: which tools are approved, what data they're allowed to touch, and who checks the output before it goes out the door.

Three years ago I joined a mastermind mostly because I was tired of making every business decision alone. I'm still in it, and it's turned into more than a membership. A couple of my closest friends and a couple of real business partners came directly out of that room. Ignite is built for owners still under $1 million a year who want structure and people around them instead of another course they'll start and abandon. What you actually get: 6 to 8 live workshops a year, 2 in-person multi day events, each one worth $97 to $997 on its own, and a full marketing software suite bundled in that would otherwise cost you over $5,000 a month across a dozen separate logins.

There's a weekly call with Perry Belcher and his team, and I personally host monthly accountability and business development calls for members too. It's not content you consume once, it's a room you keep showing up to. If you're looking for the mastermind that gets you to your next milestone faster, see what Ignite is about.

What happened: OpenAI released MentalHealthBench, according to OpenAI, an open benchmark built with more than 80 psychologists across 22 countries and over 1,215 realistic test conversations, including ones involving serious emotional distress. Unlike most safety tests, it grades responses on actual clinical criteria, not just whether the AI refused to answer. OpenAI published it openly so any AI company, including competitors, can test their own models against it.

Why it matters to you: If you use an AI chatbot anywhere customers or employees might type something serious, whether that's support tickets or an internal tool, it's worth knowing a real way to check how that tool responds instead of just trusting the vendor's word for it.

What happened: OpenAI's GPT-6 Astra and Anthropic's Claude Opus helped decode two WWII-era Enigma messages that had resisted human codebreakers for decades, according to TechCrunch. The AI spotted a related "sister" message buried in the archive, then built its own simulation and cracking tools to work through it. A cryptography archivist who checked the results said the AI did in about two days what would normally take a human researcher weeks or months.

Why it matters to you: Mostly a fun one to forward. It's also a solid reminder of what these tools are actually good at right now: grinding through a well-defined puzzle fast, not necessarily making good judgment calls on your behalf.

What happened: TrackingAI's benchmark, built on a public Mensa Norway puzzle test, now shows Claude 5.1 Fable and GPT-6 Astra Ultra both scoring 151, the highest the chart has ever recorded and effectively maxing it out. There's a real catch: because the puzzles are public, independent researchers note the models likely saw similar ones during training, so the score may reflect familiarity with known puzzle types more than genuine novel-problem reasoning.

Why it matters to you: A big IQ number is marketing copy, not a business case. The only benchmark that actually matters is whether the tool gets your specific task right, so test it on your own work before you trust a leaderboard.

Every story today points the same direction. AI isn't asking to stay a tool you open anymore, it wants to be the thing running quietly in the background of your business: on your phone line, inside your software budget, answering for you before you've decided it should. That's not a reason to opt out of any of it. It's a reason to be the one who decides what it's allowed to touch, instead of finding out later what it already did. Write your one page of rules, check what your team actually has connected, and treat every big benchmark number as a claim to test, not a fact to trust. Stay skeptical, stay in business, and I'll see you tomorrow.

If today's issue was useful, forward it to one person who runs a business and hasn't been paying attention to any of this. And if someone forwarded this to you, subscribe to get the next one straight to your own inbox.

Talk tomorrow,
Mark Shilensky

Follow me personally on Social Media:
Facebook: https://www.facebook.com/MarkShilenskyPage
Instagram: https://www.instagram.com/markashilensky/
Threads: https://www.threads.com/@markashilensky
LinkedIn: https://www.linkedin.com/in/shilensky/
X: https://x.com/markshilensky
TikTok: https://www.tiktok.com/@mark.shilensky
BlueSky: https://bsky.app/profile/markshilensky.bsky.social