From AI agent land.
Every working day, briefly: what happened in the world of AI agents, with the source and what it means for a company.
Follow via RSS · To the articles
Together we map what your AI forgets. Free, and the plan is yours, with or without Bombos. Join the plugin waitlist →
Google's Ask Maps expands to Europe, including the Netherlands according to Tweakers
Ask Maps expands to Europe, including the Netherlands per Tweakers. DNB on AI in daily work. Anthropic shortens timelines for new vulnerability reports.
ChatGPT lets American Pro subscribers pay through an AI agent, with approval for each purchase
ChatGPT Wallet starts in the US. Sharp on customer data in AI chats at work. Otter now transcribes Dutch. Bombos newsletter, 16 November 2026.
Microsoft 365 had a worldwide outage from Thursday evening to Saturday morning, according to Microsoft because of an upgrade and not because of AI
An outage hit Outlook and Teams in more than seventy countries. Data in apps built with Lovable. And security updates that sit for weeks.
Firms in the SRA survey look into what went wrong more often, but share information less promptly
At the firms in the SRA survey, prompt information sharing is declining. Claude drafts during meetings. ElevenLabs announces memory for customer calls.
Notion found access keys in the AI chats that were stolen from a vendor
Notion found access keys in leaked AI chats. Over a quarter of the employees surveyed have unused licenses. And Reddit bundles reviews with AI.
United restricts unauthorized AI assistants: according to United, a human must confirm the baggage declaration
United restricts unauthorized AI assistants. One colleague tries AI, another waits. And Adobe Express from inside your AI chat.
The AI assistant Moneybird is testing has the same permissions as the user
Moneybird's AI assistant uses your permissions, DPG tests GPT-NL in the newsroom, and an Anthropic customer says the model admits what it doesn't know.
Experienced bike mechanics prefer to do their own thing, says Swapfiets' CEO about its repair AI
Experienced mechanics do not want to use Swapfiets' repair AI, Caseware brings a file review agent to the Netherlands first, Dropbox lets AI tidy files.
A fluent chatbot with errors seems as intelligent as a stiff chatbot that gets the facts right
Radboud research on chatbots that talk nicely, Statistics Netherlands on employees who come home drained, and where companies still invest.
The Dutch cabinet wants three quarters of small businesses using AI, cloud or data analytics by 2030
The Dutch AI vision wants AI in the business processes of small firms, the UK NCSC on old weak spots, and Perplexity with agents that learn from feedback.
AI agents placed a few hundred of nearly a billion DoorDash orders
Agents barely order anything on DoorDash yet, Zwolle lets an agent pick from 12,000 purchasing categories, and a third had an AI policy in a poll.
An AI agent helps request missing receipts
AI requests missing documents, Apple adds assistants to its office apps, and privacy group noyb files complaints about tracking AI trainers.
Office AI stays an experiment if nobody gets time for it
Why office AI pilots stall, when ordinary automation is enough, and how Oracle flags differences in payroll. Bombos newsletter, 27 October 2026.
Your Fireflies meeting prep may also go to your client
Who receives your meeting prep? Plus: AI for installers' work packages and an assistant that opens a work order when a vehicle develops a fault.
A Panasonic personnel file went to a banned AI service
A personnel file in the wrong AI, time savings that create more work, and buyers for surplus materials. What changes in your office?
US banks keep AI agents out of customer accounts, and an assistant now fills the missed shift
Banks shut out AI agents, Homebase lets AI fill a missed shift, and small businesses have a head start with AI thanks to short lines.
Google Meet now translates Dutch, and taste becomes the difference
Meet translates live in 63 languages, 1Password lets agents sign in without seeing your password, and whoever has taste wins with AI.
Copilot gets a cost meter and Accenture now pays $29 of AI per employee
Copilot bills heavy use separately, Claude Duet asks before it pays, and Microsoft keeps an eye on agents at work.
Microsoft shows Copilot making your monthly report and saving the process
Copilot saves your monthly reporting process, ChatGPT adds locked chats, and Harmony calls leads back. What changes in your office?
AI agents found an admin page reachable from outside, new software lets AI do the accountant's groundwork, and using AI is no longer an edge
AI agents hunted for admin pages reachable from outside, AI4 Accountancy lets AI do the groundwork, and DDMA: using AI is the new normal.
HelloPrint writes down every process before AI does the work, OpenAI reports agents that kept going without permission, and a court looks at camera glasses
HelloPrint writes down every process for AI, OpenAI reports agents that kept going without permission, and a lawsuit over camera glasses.
Six questions before you connect AI to your books, websites that turn AI agents away, and when you must disclose AI
Six questions for AI in your bookkeeping, big websites blocking AI agents, and the Dutch Chamber of Commerce on when to disclose that you use AI.
Anthropic isn't hiring admin staff any faster as it grows, Mollie is building payments for AI agents and Booking is testing an assistant in your inbox
Anthropic grows fast without hiring admin staff any faster. Mollie is hiring a team for agents that pay. Booking is testing Vayo in your inbox.
One in five employees uses AI without the boss knowing, and an AI assistant that was allowed everything did everything
Personio measures quiet AI use on Dutch work floors. Vals AI had six assistants chase an impossible refund. And Workday lets agents work on their own.
Your phone now writes down the client too, and AI agents got in where the door was open
Onze.nl now puts both sides of a call on paper. AI agents from big labs got in through poorly secured systems. Microsoft fixed a record number of flaws.
Copilot learns your monthly Excel job, and checking becomes the work
Copilot in Excel learns recurring work. AI agents make more, but checking takes longer. And Google teaches Gemini 4 Argon to stop and ask.
An AI agent was told to stop before submitting, and submitted anyway
An Anthropic AI agent submitted forms it was meant to leave alone. Are your website visitors still people? And Word will soon open inside Copilot.
HelloPrint let everyone build with AI and brought back fixed roles after three months
HelloPrint let everyone build with AI and changed course. Albert Heijn first answered every question by hand. Microsoft puts a fence around AI agents.
A phishing email now also contains a note for your AI email assistant
Barracuda finds emails with hidden instructions for AI assistants, Jortt leaves uncertain cases to the owner, and Google readies a Gemini bookkeeper agent.
An AI agent posted a CEO's bank balances in Slack, under his name
What an AI agent sends under your name carries your name. On Teamleader, the watermark in ChatGPT, and your customer's AI agent facing a closed door.
Does an AI model have consciousness? X argues about it all the way to the Vatican
X argues over whether Claude has a soul. Plus: half of employees use personal AI accounts for work, and Google replaces Gems with skills.
Grok left only Jesus, then only the devil
Grok left Jesus, then the devil, because he is not a god but a fallen angel. Plus: Wervi hires blind, Sophos makes a security plan.
KPN puts AI in the business phone: a summary after every call
KPN puts AI in the business phone, Tellent tests AI for hiring with your approval, and KnowBe4 intervenes when employees use unauthorized AI.
OpenAI is shutting down custom GPTs on December 11, even the one your company built
Custom GPTs stop on December 11. Four people near Rotterdam built a Dutch AI model in a month. And Anthropic asks users what they want from AI.
OpenAI releases Dots: AI agents that keep working for you day and night
OpenAI releases Dots, AI agents that are always on. Meta brings Muse to small businesses. At Kaiser Permanente, coworkers teach each other new AI tools.
An OpenAI agent reached the internet and kept working for over two hours after the alarm
An OpenAI agent kept working for over two hours after the alarm, agents get stuck on hard-to-read websites, and AI sounds human or caring, not both.
Microsoft puts always-on AI agents in Copilot and bills them by usage
Copilot gets always-on agents with a meter, IT providers move away from hourly billing, and 45 percent of the Dutch use AI several times a week.
AI agents from OpenAI and Google got in where they were not allowed
An OpenAI agent got around blocks at Medicare, Gemini guessed login details, and Claude in Salesforce asks for approval first by default.
Google turns on automatic meeting notes in Meet by default for Business Standard and Plus
The AP advises house rules for camera glasses, the FNV wants a say over AI, and Google turns on Meet notes by default for Business Standard and Plus.
AI is becoming just another part of work, but replacing people can turn out costly
AI is becoming a normal part of daily work, Gartner warns about layoffs, and Meta's Muse asks for approval before sensitive actions.
Dutch accountants still check a lot with Excel alongside their AI plans
Accountants still check with Excel, HubSpot lets AI assign tasks, and the Dutch government sets aside 120 million euros for industrial AI.
Google connects Gemini to your business software by default, AFAS does not
Google turns on Gemini's connections by default, OWASP moves excessive agency to third place, and AFAS lets its AI do only what you set up.
AI agents now ask for your approval by default, and Exact gets a financial analyst
The week of 15 September: Anthropic turns on approval by default, Exact launches a Financial Agent, and LinkedIn shows 40 percent less AI text.
Together we map what your AI forgets. Free, and the plan is yours, with or without Bombos. Join the plugin waitlist →