Calvin's Updates

The Latchkey Club Daily Draft — August 6, 2026

Teleprompter / Blog Script

I was connecting another tool to my AI assistant recently, and the setup screen asked for permission to read my email, view my calendar, open files, and act on my behalf.

I caught myself doing what most of us do with those screens. Scroll down. Look for the blue button. Click accept.

Then I stopped and thought, that is a lot of keys to hand over in about eight seconds.

Welcome back to the channel, guys.

Today I wanted to talk about the part of AI assistants that is becoming more important as they get more useful: access.

The big technology companies are all moving toward some version of an assistant that can work across everything. It can read the message, find the appointment, pull the document, research the trip, order the item, and possibly send the reply.

I understand why that sounds good. I use an AI assistant every day, and the useful part is not that it can answer a trivia question. The useful part is that it can help me keep track of real life. Calendars, reminders, research, family logistics, expenses, unfinished projects, and all the things that tend to fall between the cracks when work gets busy.

But the assistant can only help with those things if it can see some of those things.

That is where this gets uncomfortable, because convenience keeps asking for another key.

If I want it to summarize email, it needs email access. If I want it to coordinate the calendar, it needs the calendar. If I want it to organize documents, it needs the files. If I want it to buy something, now we are talking about payment information and permission to make a decision that costs money.

Each request makes sense by itself.

The problem is that, six months later, one system may have access to nearly everything, and I may not remember why I approved half of it.

I think people our age may actually have an advantage in thinking about this, because most of us have spent years working inside systems where access matters.

At work, being trusted does not mean you get every password in the building. A good employee still has a role. A manager may approve a purchase, but somebody else verifies the invoice. A technician may change a device, but there is a backup before the change. We learned those habits because something eventually went wrong.

Sometimes the lesson was expensive. Sometimes it was just embarrassing. Either way, you stopped giving everybody administrator access.

So why would an AI assistant be different?

There was a good MIT Technology Review piece earlier this year about the security problem with connected AI assistants. A chatbot that gives a bad answer is one kind of problem. An assistant that can read email, browse the web, open files, and take action is a different kind of problem because a mistake can move beyond the chat window.

One issue is called prompt injection. In plain language, the assistant reads something from outside, maybe an email, a website, or a document, and hidden or misleading instructions inside that content affect what the assistant does.

The danger is not only that the AI says something wrong. The danger is that it has enough permission to do something wrong.

OWASP, which does security guidance, recommends a pretty familiar answer: give the system only the access it needs, and require a person to approve high-risk actions.

That is not very futuristic. It is the same basic idea we have used around tools, money, and responsibility for a long time.

Trust the helper. Limit the keys.

I am still working out what that means in my own use, but I think there are a few questions worth asking before connecting another account.

What does the assistant need to see to do this one job?

Does it need to take action, or would a draft be enough?

If it makes a mistake, can I undo it?

And would I be comfortable if somebody else got the same access the assistant has?

That last question changes things.

I might be comfortable letting an assistant read a folder of notes I created for research. I would think differently about giving it every document I have ever saved.

I may want it to draft an email, but not send the email without showing me first.

I may want it to remind me that a bill is due, but not move money on its own.

I may want it to suggest a purchase, but I definitely want to see the total cost before anything gets ordered.

That is not refusing the technology. That is deciding where judgment still belongs.

The other thing I want is a way back.

We tend to talk about AI in terms of what it can do when everything works. I want to know what happens when it does not work.

Can I disconnect the account? Can I see what actions it took? Is there a backup? Can I recover a deleted file? Does the family know how to reach the calendar or the important documents if this assistant disappears tomorrow?

That matters more as we get older.

A lot of us are slowly becoming the keeper of the family’s digital life. We know where the tax documents are, which account pays the bill, where the photos live, which calendar has the real appointment, and why there are three different passwords for something that should only have one.

If we put an AI layer over all of that, it may make our lives easier. But it should not make the rest of the family more dependent on a system only we understand.

A useful assistant should reduce confusion, not become the new single point of failure.

I also think there is a difference between privacy and secrecy here.

Privacy does not mean I refuse to let any tool know anything about me. If I did that, most useful technology would stop working. Privacy means I make a reasonable choice about what is shared, why it is shared, and who or what can act on it.

I am not going to pretend I read every privacy policy. Most of us would need to retire just to finish the terms and conditions we have already accepted.

But I can slow down long enough to notice when a useful feature is asking for a much larger permission than the job requires.

I can separate low-risk work from high-risk work.

I can let the assistant organize information while keeping final approval over messages, money, medical decisions, and anything that would be difficult to reverse.

And I can review the connections once in a while instead of treating every permission as permanent.

Maybe that is the hidden advantage of being 55 or 60 in this moment.

We have lived long enough to know that capability is not the same as authority. The fact that somebody can do a thing does not automatically mean they should be allowed to do it without review.

That applies to people. It applies to software. It probably applies even more when the software can move faster than we can notice.

I still want the assistant. I want the reminders, the research, the organized schedule, and fewer loose ends.

I just do not want convenience to quietly turn into unlimited permission.

So the next time a setup screen asks for every key in the house, I want to pause before I hit accept.

What is the job? What is the smallest access that will do it? What still needs my approval? And if this goes wrong, how do I get back?

Those questions are not a sign that we are afraid of the future.

They may be the reason we can use it without losing control of the things we are trying to protect.

Anyway, that’s what I’ve been thinking about.

How much access would you give an AI assistant, and which decisions would you never let it make without you?

Leave me a note in the comments. Thanks for listening.

Video Prompt Script — Questions to Answer Without Reading

Use these as prompts. Don’t read them on camera; answer them naturally.

1. Opening: What permission screen made you stop before clicking accept?
   • Follow-up: Why did the amount of access feel different once you named each key?
2. Real usefulness: What practical jobs does an AI assistant help you with now?
   • Follow-up: Why does useful help require some context rather than only a chat window?
3. Permission creep: How can several reasonable permissions quietly become access to almost everything?
   • Follow-up: When was the last time you reviewed which apps can reach your email, calendar, or files?
4. The Gen X advantage: What did decades at work teach you about roles, approvals, backups, and administrator access?
   • Follow-up: Why is trusting a helper different from giving it every permission?
5. The real risk: What changes when an AI can act instead of merely answer?
   • Follow-up: How would you explain prompt injection without turning the episode into a security lecture?
6. The four questions: What does the assistant need to see, does it need to act, can the action be undone, and would you give a stranger the same access?
   • Follow-up: Which question makes you reconsider a permission fastest?
7. Human approval: Which actions are fine to automate, and which should stop for review?
   • Follow-up: Where do messages, money, purchases, medical decisions, and deleted files fall?
8. A way back: What backups, logs, disconnect options, and family instructions should exist before one assistant becomes central?
   • Follow-up: How do you prevent a helpful system from becoming a single point of failure?
9. Close: How can somebody use a capable AI assistant without letting convenience become unlimited permission?

Title Options

1. I Want an AI Assistant, but I Don’t Want to Give It Every Key
2. Before You Connect AI to Everything, Ask These Four Questions
3. The More AI Can Do, the Less It Should Access by Default

Thumbnail / Onscreen Text Options

• DON’T GIVE AI EVERY KEY
• USEFUL DOESN’T MEAN UNLIMITED
• TRUST THE HELPER. LIMIT THE KEYS.

Shorts / Reels Cutdowns

• “That is a lot of keys to hand over in eight seconds.” Use the permission-screen opening and quickly name email, calendar, files, purchases, and the problem of permission creep.
• “The danger is that it can do something wrong.” Explain the difference between a chatbot giving a bad answer and a connected assistant taking an action.
• “What is the smallest access that will do the job?” Cut the four-question test and give concrete examples: draft but do not send, remind but do not transfer money, suggest but do not purchase.

Viewer Question

How much access would you give an AI assistant, and which decisions would you never let it make without you?