James Kimbley

Sent a link to someone to a 19-minute video. They replied less than five minutes later that it was a great video.

Ever trusted AI to help you code, only to discover it led you straight into a trap you never saw coming? That’s the hidden danger of vibe coding: you don’t know what you don’t know.

Over the weekend, a client contacted us. They are vibe coding a new internal app for their business, which will save them a lot of time. They wanted certain Google Workspace settings changed because the AI told them that was required. All good and well, but the problem was that changing the settings would have opened a vulnerability.

The client wouldn’t know this; they’re simply following the AI’s advice because they’re unaware of what they don’t know, so they can’t recognise when the AI is cutting corners. All the AI wants is to give them the results they request; it doesn’t care how it gets them.

AI and vibe coding are so easy, but it removes the essential element of understanding how something should work. If you go back to the days of Googling, you had to know something about the topic you were researching so you could tell whether the page you landed on was accurate.

This highlights the problem of vibe coding: you don’t know what you don’t know, so AI can lead you down the wrong path.

While major platforms can security‑check the code provided, AI will only act on the prompt you give it, and if the prompt lacks vital information, the AI won’t point out what’s missing - unless you specifically ask.

And that’s why partnering with a leading AI IT support partner like Kimbley IT is vital, so we can identify these risks before you vibe-code them into production.

So how did I help the client at the weekend? I didn’t deploy the changes they asked for; instead, I guided them down the correct path and produced the prompt they should have asked their AI for if they had known what they didn’t know. They are now on their way to creating a really handy app for their business. 👍

Here’s a photo of a friendly juvenile robin that came right up to me in the garden over the weekend.

Our first Google Workspace Marketplace app has been submitted to Google.

Hopefully it gets approved 🤞

It will save our Google Workspace by Kimbley IT clients a lot of the time when dealing with particular file types that aren’t natively supported in Google Drive.

We’re also making it free to the world so anyone can deploy it on their Google Workspace stack.

I’ll reveal more and the Marketplace link once the app gets approved!

If you’ve lost control of the model because it’s actively escaped, how can you guarantee the kill switch would even work?

PSU went bang this morning. It’s the first time I’ve had a PSU fail on one of my machines.

For the past 18 months, I have been building Kai, testing, spotting problems, testing again, and fine-tuning.

During testing, Kimbley IT clients interacted with Kai during short live test runs, which I constantly monitored; in that time, only one person guessed they were interacting with an AI agent rather than a living member of the team.

This week, Kai has been released and is autonomously interacting with clients without the overwatch I have provided in the past. So far, Kai has done brilliantly, with no need for the team at Kimbley IT to get involved or deploy the kill switch.

With these increased real-life interactions, I’m able to fine-tune Kai even more, especially when scenarios play out that you hadn’t expected or seen during the testing phase.

My vision for Kai is that within the next 12 to 24 months, Kai will be running at least 90% (hopefully more) of Kimbley IT, and the already-large proprietary instruction set and system prompt will be highly valuable.

To get there, I plan to focus on a few key next steps: continue scaling Kai’s involvement across more client interactions, integrate feedback from real-life scenarios to enhance its performance, and continually review the system prompt to ensure it captures our evolving client needs.

The value, though, for our clients is that Kai is enabling us to provide increasingly more services and features at no additional cost, and the feedback so far has been brilliant.

One client recently shared that a new email security system powered by Kai, which analyses dodgy-looking emails and guides team members on how to deal with them, was a “Bloody good system in my book.” When Kai advised them not to engage with a lead that would turn into a scam later down the line. This was based on Kai’s knowledge of past emails and current scam trends, allowing Kai to make an informed decision and provide excellent guidance.

Kai’s definitely going to win the team member of the month award.

The hatchling has emerged. Young pigeons develop quite rapidly; I estimate this one hatched a week ago, but this is the first occasion I’ve been able to observe it.

(The body to the left is the parent) I have only been able to get a photo of the hatchling’s head.

Heads up to business owners, sales teams, and procurement departments: the Net 30 Purchase Order/Request for Quote (RFQ) scam is making a comeback.

Years ago, I used to see this scam email style often, but I haven’t seen it in years, and my research shows it’s making a comeback. The screenshot is of one that was just submitted via our website contact form.

While it frequently targets the IT sector with high-end tech (like the high-value processors shown in the screenshot), scammers are rapidly expanding across all industries. Whether you sell construction materials, medical supplies, laboratory equipment, specialised tools, or wholesale goods, if your inventory is valuable and easy to flip on the secondary market, your business is a target for these fake bulk orders.

The playbook is always identical. A scammer (or a bot filling out your website’s contact form) will impersonate a real buyer from a legitimate organisation, like a university, hospital, or major corporation, often using a lookalike email domain. They place a large, unsolicited order and casually slip in “Our payment terms are Net 30” to bypass immediate payment. If you ship the goods on credit, they vanish, the real organisation has no record of the order, and you are left with a devastating financial loss.

Always independently verify new bulk inquiries, scrutinise the sender’s actual email domain, and never grant credit terms without a rigorous, verified background check.

Just announced, Google is renaming NotebookLM to Gemini Notebook. So if you’re a fan of NotebookLM, be prepared for the name change when you go looking for it in the future.

Gemini in Chrome is now rolling out in the UK! Google has taken its time with this release, that’s for sure!

• Chat with Gemini in Chrome via a new side panel to summarise articles or compare data across tabs.

• Work in one window. Schedule meetings in Google Calendar, look up locations on Google Maps, draft messages in Gmail, and ask questions about YouTube videos without leaving your page.

• Gemini remembers past conversations to give you personalised, tailored answers.

• Instantly edit or transform online images with simple text prompts in your Chrome side panel.

• Google’s models are trained to recognise threats like prompt injection and include safeguards that request your confirmation before sensitive actions.