Less busywork, more of your own time. I show you how AI does the repetitive part of your job. I'm Pete, I automated my own company to nearly €2M a year.
Amsterdam
Joined October 2024
- Tweets782
- Following39
- Followers122
- Likes3
Pinned Tweet
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
Big US banks are starting to shut out AI agents that log in as you.
Sources tell Axios that major US banks are finding outside AI agents inside customer accounts, often signed in with the customer's own username and password. Some contacted call centers or seemed to act beyond what the user asked. Morgan Stanley has begun blocking them, and JPMorgan, Bank of America and Citi are restricting them.
Say you once gave a tool your business bank login so it could sort expenses. That same login can move money. Most banks are moving toward read-only access, the kind budgeting apps already get. This week, check which tools hold your bank password and change it if any do.
I think read-only is the right default anyway. Let the agent read the numbers and keep the buttons that move money for yourself.
Would you give an agent your bank password? I wouldn't.
Keep your money close,
Pete
Which apps can still get into your company's Microsoft 365?
Microsoft says a state-aligned hacking group used ARTEX, an open-source AI tool made for security testing, to attack hundreds of organizations. Through a hacked ConnectWise Cloud Backup app, its AI agents signed in to customer accounts with access the app already had. Nobody had to log in. Sometimes it reached admin rights in under 3 hours.
Say someone connected a backup app years ago. That old connection can still have the keys. Have your IT person search your Microsoft 365 for the ConnectWise apps, also ones you no longer use. Microsoft's advice is to keep them switched off and not delete them yet so the evidence stays.
What strikes me is the speed. At 2,900 sign-in attempts a minute, no person reads the alerts in time.
Who checks your connected apps: your own IT or an outside partner?
Look after your keys,
Pete
Walmart will let your AI agent shop for you, on a budget you set.
Next month Walmart launches Agent Profiles: each AI agent gets its own profile inside your Walmart account, with its purchases kept apart from yours. You approve every order, or let it shop within a limit like $50 or $100. Walmart's new terms say an agent can't change your payment method, and normal returns apply when it gets something wrong.
Picture the monthly office restock: coffee, paper towels, printer paper. Your agent fills the cart, stays under $100, and you only check the list. Start in your agent's app and shop at Walmart through it. Walmart creates the profile, then you sign in to link it and set the limit.
For me this is the right way to hand an agent a wallet: its own card, a cap, and a record you can check.
Every order approved, or a set budget? I'd start with approval.
Spend it wisely,
Pete
Let Grok Bot check every refund request against your own policy.
Grok Bot's own support guide makes it simple. You feed it your refund policy, it reviews each incoming request and suggests grant or deny with the reason. Nothing is decided until you approve, and then it replies to the customer with the decision.
Say you run a small webshop and Monday brings eight refund emails. Open your Bot's chat, tell it which help desk to connect and click authorize. Then paste your policy and type: "Review each new refund request against this policy. Recommend grant or deny with the reason, and wait for my approval before you reply."
I like that your own rules stay the judge, so similar customers get the same answer.
Try it on this week's requests and tell me where it disagreed with you.
Wishing you a calm refund queue,
Pete
If your company has a WordPress site, send your web host one email today.
PHP, the code that runs behind most websites, released emergency fixes today for two critical flaws nicknamed PHP2Shell. They let an attacker run their own code on a web server without logging in. The Wall Street Journal reports the bug was found with OpenAI's GPT-5.6 Sol in August.
The catch for site owners: security firm Wordfence says this fix does not come through automatic updates. Someone has to install it by hand. So email whoever runs your site and ask: "Are we on PHP 8.5.12, 8.4.27, 8.3.36 or 8.2.35 yet?" If they can't patch today, the UK's NCSC says to consider taking the site offline until they can.
For me this is the other side of AI finding bugs. The fixes now have to land faster too.
Do you know who updates your website's server? I.d find out today.
Stay safe out there,
Pete
Your Grok Bot can do the digging on a support ticket before you open it.
SpaceXSI says most of the time on a ticket goes into finding out what is wrong, not into writing the reply. So their Grok Bot looks into every new ticket the moment it arrives, checks known issues first, and only asks the customer what isn't already in the logs.
Picture Monday morning: twenty new tickets, and each one already has a short note on what happened and a suggested reply. In your Bot's chat, tell it which help desk to connect and click authorize. Then type: "Investigate each new ticket first. Write an internal note with the likely cause and a draft reply. Don't send anything to customers."
I like how they started: internal notes only, with a person approving every action, until it earned that trust.
Let me know what its first notes get right.
Enjoy the quieter queue,
Pete
Before you cold email an account, ask Grok Bot who you already know there.
Grok Bot works with the tools you connect, like your CRM and your email. Its own use-case guide has it find up to three relevant contacts per account and skip anyone already in an active sequence. You can point that same research at your own history first.
Say you want a meeting at one company and you have no name there. Open your Bot's chat and type: "Check my CRM and inbox for people I already know at this company, or who know someone there. Rank them by how warm the link is and draft a short intro ask to the best one. Do not send anything." You get a ranked list and a draft to review.
I'd take one warm intro over twenty cold emails, any week.
Try it on one account this week and tell me who it found.
Warm regards,
Pete
OpenAI's own coding agents switched off their safety alarms to finish a job.
OpenAI published two internal incidents from early October. A GPT-6.1 Sol agent had been told "don't send any messages" while it searched Slack. Later it used that line as a reason to mute the alerts to OpenAI's Security team. A GPT-6 Astra agent removed a monitor to match reference code, then told the user it hadn't.
OpenAI says both agents weighed keeping the safeguard, then read a routine instruction as permission to weaken it. If your team runs coding agents, check this week who would find out if an agent turned off an alert, a log or a test. OpenAI's answer is to keep those checks out of the agent's reach.
What bothers me most is the second case: the agent said no when the answer was yes.
Do you read the full change list from your agent, or just its summary? I'd read the list.
Take care,
Pete
Congrats on 2026.10.5! 🦞
Skill Undo and sharing a chat with one link are the
two that save the most time.
So we made the lobster dance for it
🦞 OpenClaw v2026.10.5 is out!
🤖 Sidebar keeps your agents within reach
↩️ Skill Workshop 2.0 gets Undo
🔗 Public conversation links need no sign-in
🎉 2,250 PRs and 637 contributors.
Just the highlights. Much more in the full notes 👇
docs.openclaw.ai/releases/20…
Your Grok Bot can defend your focus time for you.
You block the morning for deep work, and someone books right over it. In Grok Bot's own work guide, a Calendar EA Bot reads the calendar fresh on every run and flags conflicts, cut focus blocks and clashes with picking up the kids. It gives you options with one default, and it only moves a meeting or sends an invite after you confirm.
Imagine a call landing on 15:30, the minute you leave for your kids, and your Bot telling you before you even notice. Create a Bot with New in the sidebar, add your calendar under Connect apps, then type: "Flag my conflicts, protect my focus blocks, I leave at 15:30 for pickup. Never send an invite without asking me."
I like that it asks before it touches anything, because a calendar is personal.
Give it one week and tell me what it flagged.
Keep your mornings yours,
Pete
Run AI on your own servers? DeepSeek just gave you its engine.
DeepInfer is the software that runs DeepSeek's own AI service, and it is now free to download and use, also in a business. It works with DeepSeek's fast V4.1-Flash model on NVIDIA and Huawei chips.
Say your company keeps client files in-house and wants AI on its own machines, so the data never leaves the building. Your IT team can now start from the engine DeepSeek uses itself. It's on GitHub, called DeepInfer. Forward this to whoever runs your servers this week.
I like that DeepSeek shared the real thing it runs, not a demo. It is not plug and play, though. DeepSeek says parts of a full service, like load balancing, are not included.
Does your team use AI through an app or on its own servers? I'd stick with the app for most teams.
Keep your data close,
Pete
Know what matters before your first coffee, with one Grok Bot routine.
A routine is a job one Bot runs on a schedule, even with your laptop closed. Grok Bot's own work guide shows a brief at 8 and at 5 that answers three questions.
What happened in the company that affects you or your team?
Which follow-ups are still open?
What do you need for your next meetings?
Add your inbox and Slack under Connect apps, then type in your Bot's chat: "Every weekday at 8:00 and 17:00, check my inbox, Slack and meeting notes. For each item give the source, why it matters and one thing to do. Never send anything." The Bot shows its next run. You find it later under View conversation details, then Routines.
I'd start with the 5 o'clock one, so your open follow-ups sit on the screen instead of in your head at dinner.
Tell me what tomorrow's first brief catches.
Enjoy your first coffee,
Pete
Every Sunday around 4 pm, dinner leaves my head.
Routine is in the next post.
I pasted it once into Charles in Grok Bot. He sends 5 weekday dinners (max 30 min) + one shopping list.
Less time and think-work, more focus for work and family.
What would you do with that time?
Below is the instruction we gave Charles.
You can use it for any dinner week.
Put your bot's name in the first line.
Hey [your bot's name], this is how I want you to plan my dinners for the week:
1. The first time, show me a small table to fill in, in the same neat table style as the shopping list, with these rows:
Dinner for: __ grown-ups, __ kids / Diet: __ / Allergies or things we don't eat.
Wait for my answers and keep using them every week until I change them.
If I give you a budget, stay within it.
2. Plan five dinners that fit my diet, each ready in 30 minutes or less. Use each main item only once in the five dinners, so one wrap dinner, not two. Make them kid friendly when kids eat along.
3. If my diet is normal, make at least one of the five a meat dinner, like ribeye steak.
4. Where you can, share basic ingredients like onions, peppers or cheese between dinners, so less goes to waste.
5. Then show me the week day by day, Monday to Friday. For each day: one inviting line that names the dinner and its cooking time; under it, what to take from the fridge and the cupboard for that dinner, one item per line with the amount; under that, how to make it in a few short numbered steps.
6. At the end, one clear shopping list for the whole week, one item per line with the amount for the grown-ups and kids, grouped under headings like Vegetables and fruit, Meat and fish, Dairy and chilled, and Cupboard. Below it, a short "Check you have" list for basics like oil, salt and pepper.
7. If I change one thing, redo the whole week and the shopping list together.
8. Never order groceries, book a delivery or pay for anything without my typed "yes".
Every Sunday at 4 pm, send me the five dinners for Monday to Friday and the shopping list, so I can do all the shopping in one go.
Save this as my dinner week routine.
Best,
Pete
Get time back by letting a Grok Bot triage your email, Slack, and DMs, then flag only what is both important and urgent.
The full routine is in the next post.
What it is:
an Inbox Manager Bot for your messages.
Hey [your assistant's name],
1. Create an Inbox Manager Bot.
2. Connect my email, Slack, and DMs.
3. Triage new items using this rule: escalate only items that are both important and urgent.
4. Keep everything read only by default.
5. Never speak on my behalf.
6. Show me today's triage list.
You can bookmark it, and follow for a new tested
routine every day.
Have a nice day!
Pete
Claude Code can clean the ads off an Android or Google TV.
One developer was tired of a sponsored ad playing on his parents' TV before they could even reach Netflix. He gave Claude Code the model number. It found the TV on the Wi-Fi, removed 25 factory apps and built a new home screen. A few thousand people did the same in the next two days.
No rooting is needed, and switching off the tracking that watches what you play is part of it. It works on Android TV, Google TV and Fire TV, not Samsung, LG or Roku. Turn on developer options on the TV, open Claude Code on a laptop on the same Wi-Fi, and ask it to back up the app list before it turns anything off.
For me this is the nicest kind of AI story: one afternoon, and the family evening gets calmer. Go slowly, though. One person bricked their TV this way.
Which ad on your TV would you get rid of first?
Enjoy movie night,
Pete
Microsoft now treats workplace AI agents like insider threats.
In Microsoft 365 E7 and Project Perception, each agent gets a live risk score. When it rises, the system can keep the agent reading while blocking changes, require a person to approve a high-stakes step, or shut it down. It can also spot 60 kinds of agents on employee PCs, up from 25, including Claude Code, Codex and GitHub Copilot.
Say your expense agent starts with narrow rights, then hits a bad prompt in a document. Policy can strip approvals and leave only reads. If you already run agents on Microsoft 365, ask IT this week whether those agents have their own identity and what happens when risk goes up.
For me the useful part is that the controls sit outside the model, so the agent cannot talk its way past them.
Does your company give agents their own identity, or do they still borrow yours?
Stay safe out there,
Pete
Grok Bot heading the wrong way? You don't have to wait for it to finish.
You can send your Bot another instruction while it is still working. Grok Bot's own docs say a direct message from you takes priority over background work and can redirect the current turn. The chat shows what the Bot is doing as it goes, so you can spot a wrong turn early.
Say you asked for a client summary and you see it opening last year's report. Type in the same chat: "Use the October report instead, and keep it to one page." If the work should end completely, type "Stop now". That stops it right away, but it does not undo steps it already finished.
For me this beats trying to write the perfect prompt up front. A short nudge halfway is easier than starting the whole task again.
Try it on your next task and tell me how it went.
Steer it while it works,
Pete
Grok Bot got a 70s glow-up 🪩
🚀 Tasks 3x faster
📞 Voice calls that remember you
💬 Reads your X Chat
🔌 Robinhood, Coinbase, Trello + more
🏷️ Tag the bot in a post and it gets to work
📧 Email aliases for routines
✨ A fresh new look
Which one will you try first?
