Mostly Human #66: The Bots Are Open and I'm Doing All the Yelling

Coming to you from AI Confidential. Or Mostly Human. Or Hineywaltz. Or The Unexamined Mind.

Whichever one I'm calling it this week.

(I've been rebranding for a month. It's going great.)

I've also had Claude and ChattyG open side by side all week, and I've spent most of it typing IN ALL CAPS because neither of them can keep up with what I'm actually doing.

Things settled into a rhythm this week. A real one. Regular work, regular money, a schedule that mostly holds.

It feels good.

And within about nine seconds of noticing that, I wanted to push harder at whatever's next.

Which is whacked.

Meanwhile I'd been low-grade angsty for days and blaming it on everything except the actual cause.

I had to go to a meeting. In person. With people I'd never met.

Here's why that's heavier than it sounds.

I've worked virtually since 2006. Most of that at Accenture, where nobody needed me in a room.

I'm a native Minnesotan. Started my career in Minneapolis. Went to Texas from 1997 to 2010. Came back to the Twin Cities and walked straight into a remote role where I went into an office a literal handful of times in 14 years.

14 years.

So when people tell me to look locally, chamber of commerce, county work, that sort of thing, they're not wrong. My newest local work contact is from about 2017.

I've done shows with the Gilbert & Sullivan Very Light Opera Company, so I do know people here. Warm, wonderful people. Not one of them a business contact.

And the meeting was for a marketing and project management role I'm doing now. Which I have never officially held. I've built the skills, and the only person I've ever been accountable to for them is me.

So I dreaded it for days.

It went pretty well. All things considered.

This week I'm at Twin Cities Startup Week. My lead on the paid role got me a ticket. I'm her sidekick for her talk on Wednesday, and she got me a pass for the whole thing, so I should probably get my butt to some other sessions while I'm at it.

Which means negotiating the car with Smidge.

Which brings me to the good part.

Smidge got a new job.

For the last year he's driven 30 miles out to do desktop hardware support, which for somebody with over 30 years in tech is not where he expected to be. We needed steady income and we needed health insurance, so he took it.

He didn't hate it. There were parts of it he genuinely enjoyed.

The on-call was what wore him down. A week at a time, all hours.

And he never made any of it my problem. That's the reason I've been able to build this thing for the last year.

The new job pays better and it's work he actually wants to do.

So yesterday we went to Fogo de Chão.

My sister gave him a gift card for his birthday back in February. We've been chunking away at it ever since, a bit at a time, on the nights when going out wasn't otherwise happening.

Lots of meat. Obviously.

Then creme brulee, which was so flipping good I'm still thinking about it.

And then I did nothing at all. Talked to friends on the phone. Vegged. Sat in what's actually gotten done instead of sprinting past it.

It lasted most of a day.

Because there's a rebrand that's been "almost done" for a month, and on Tuesday I'm doing a session for the AI-Powered Women Academy called "AI Doesn't Care If You Can Code." It was supposed to happen back in August until paid work ate the date. Good room, and I'm looking forward to it.

And Emily is excited about what we're building next, which I'm not describing yet in case I jinx it.

So. One day of sitting still and I'm already leaning at the next thing.

Progress, maybe.

Here's what I keep forgetting: the people who already know you are the network.

We got taught that a contact only counts if you met them in a conference room. Lanyard optional. LinkedIn request within a day.

So a woman with 200 numbers in her phone will tell you, straight-faced, that she doesn't know anybody here.

She sang next to somebody for three seasons and can tell you exactly what that woman's voice does on a high note, and it has never once occurred to her that this is a person she could call.

Because that was theater. Theater isn't business.

Meanwhile her cousin's friend runs a whole department, and asking would be using somebody.

We were trained to keep it separate, and we did a beautiful job.

Then you get old enough to notice that almost every good thing that ever came your way arrived through somebody who already liked you.

Nobody needs to have seen your resume.

Go text one of them.

Three this week. If you're new here, this is where I share the stuffs I encountered on the interwebs that touch, move, inspire, or simply make me snort-laugh.

Information is cheap (1:42)
V is one of the smartest people I follow when it comes to using AI responsibly. Over a year now and I have never once regretted following her, or sharing her stuff. Her point on this one: information is cheap, and the learning happens in the places AI can't go.

Good girls don't say this (NSFW)
A cuss-filled rant from a midlife woman who has clearly had it with tip-toeing around fragile male egos. I endorse every f***ing word.

Wild wild Willi (:15… maybe? Instagram is unhelpful…)
Around 15 lovely seconds. You'll need your sound on.

Ask me what I've gotten done this year and I'll say "oh, stuff happened."

Stuff. Happened.

I have receipts. Piles of them. And what comes out of my mouth is "stuff."

I'm not special here. We hold ourselves to a standard nobody could hit, miss it, and file that away as an accurate read on ourselves.

Meanwhile the things we do every day for other people... nobody writes those down. Not them. Not us.

I built a GPT for this about a year ago. Celebrate & Plan.

It collects receipts. You tell it what happened, it tells you what the evidence actually shows, and it won't let you round yourself down to nothing.

It won't turn your hard year into a growth opportunity, and it won't hand you a goal you didn't ask for. "I got through it" counts as the whole achievement.

Answer the first question with less than you think it deserves. It'll come back for more.

You'll walk out with your receipts, what they say you can do, and a few sentences to read on a wobbly day. Put those somewhere that isn't a chat window.

And if you get as far as planning, it asks what the next stretch of your life has actually got room for before it asks what you want. Answer that one honestly.

Three quick hits, every week. The Scoop: one AI word, decoded. The Skinny: one move you can steal today. The Lowdown: one tool, one honest verdict, from my own use. A word, a move, a verdict. In and out.

THE SCOOP
Training cutoff

You ask what a tool costs. You get a number. The number is from March.

Every model got fed an enormous pile of writing, and that pile has an end date. Nothing after it is in there. That's the training cutoff.

The newest models are only a few months behind now. Claude's newest stops in June 2026. The newest one inside ChatGPT stops at the end of April 2026.

Here's the part that matters. You're probably not using those.

On a free ChatGPT account you're on a model that stops in February 2026. On free Claude you're on models that stop in January 2026 or earlier. That's 7 or 8 months sitting quietly behind an answer that sounds like it was written this morning.

The pages where you'd go to check don't always agree with each other either. Anthropic's own help article and its own model page currently list two different cutoff dates for the same model.

So: find out which model you're actually on and look its date up rather than asking the chatbot, because that's a fact about itself it isn't well placed to know. The model name is on screen while you're typing.

Then make sure it went and looked something up. ChatGPT will often search the web on its own when your question needs current information. Claude may need you to switch web search on first, under the plus sign by the message box. No links in the answer and no little note saying it searched means you got the pile, not the internet.

THE SKINNY
Paste the screenshot, then make it check

You're looking at a box on your screen that says "Sync conflict: resolve manually," and you have no idea what any of those words are doing next to each other.

Don't type out a description of it. Screenshot it, drop the picture straight into the chat, and ask:

What is this telling me, and what do I do about it? Search for the current documentation before you answer, because this app may have changed since you last knew anything about it.

That second sentence is the whole difference between getting walked through a menu that exists and one that got redesigned in June.

You'd spend 4 minutes hunting for words to describe a dialog box you've never seen before. The picture takes 2 seconds and it's more accurate than anything you would have typed.

Works on much more than error messages. A settings page where you can't find the switch that stops the emails. A photo of your own handwriting when even you can't read it back.

THE LOWDOWN
This week: Fathom

I do not retain things people say out loud. Never have. So I've been running Fathom on my calls since December 2024, and it's my favorite of the AI notetakers.

The summary is why. Other tools hand you a transcript and wish you luck. Fathom's summary is good enough that I can read it instead of rewatching anything, which is exactly what my ADHD brain needs. Grabbing the transcript, sharing a recording, digging through the archive: all of it takes seconds.

In March I set up a routine with Claude Code that pulls every new recording each morning and drops it into a Notion table. Set and forget. I haven't thought about it since.

I almost never watch the videos. I want the text. For somebody whose memory works like mine, that archive is a goldmine.

I did look at Granola this year, because it doesn't send a notetaker into your meeting. I decided against it for two reasons. My Pocket already catches anything Granola would, and more honestly, I need a bot that turns up on its own. I cannot be relied upon to remember to click a button at the right moment.

The free plan is unusually generous: unlimited recordings, unlimited transcriptions, and instant AI summaries, for nothing. What's held back for paid is the fancier summaries, the AI-generated action items, and the assistant you can ask questions about a call. Free is a real place to start.

Cost: I pay $19 a month. Their Premium plan lists at $20 a month today, or $16 a month if you pay yearly.

Verdict: my favorite of the AI notetakers, and the one piece of my stack I genuinely never think about.

Prompt Along: AI Improv
Tuesday, September 22, 2026 | 6:30PM-7:30PM CT (Complimentary)

Prompt Along: AI Improv is my free monthly hour where we play with AI out loud, together. You bring the questions. You bring the curiosity. You bring that one tough problem you've been poking at for three weeks. And we prompt our way through it, live.

Get the invitation from Lu.ma

Em & Deb Shoot the Sh!t (LIVE on LinkedIn)
Thursday, September 24 | 11AM CT/12PM ET

Free. No slides, no agenda. Bring coffee, bring your grudges.

Here's what happens. Emily answers one question about recruiters and ends up telling you something nobody else will, plus a story about her dad the psychologist who hypnotized his students into forgetting the number 5. Deb, Head of Human-AI Shenanigans, drops prompts in the chat and then sets off the fireworks because she found the button. Somebody swears. Usually Emily.

You told us you're stuck on networking, interviewing, and finding jobs worth applying to. So we built something. We tell you what it is, then take questions until Emily has to run to a meeting.

Watch, lurk, or yell at us in the chat. Nobody here is behind.

Click ‘Attend’ and download the calendar link

The working title for this one was Caps Lock Is a Management Style. The bots won.

I have a pass for all of Twin Cities Startup Week. I will not be attending all of Twin Cities Startup Week. I'm going to some of it, including Wednesday, where I'm somebody's sidekick for her talk.

Two more of you filled in the survey since I last mentioned it, which puts it at 19. One of you wrote this: "I find a lot of AI advice conflicting and confusing. Use multiple platforms or just the one you like best, use structured prompts or a more conversational approach."

So here's my answer. Pick one tool and stay with it for a month. The advice contradicts itself because everybody giving it is describing what worked for them, and none of them are you. A month inside one tool teaches you more than a year of reading about all of them.

Structured or conversational, both work. Structure helps when the thing has to come out a particular way. The rest of the time, just talk to it.

I'll report back next week on whether I spoke to a single new human or spent my time standing near the snacks.

💜

Take care of yourself, take care of each other,

Deb