NotYourAIDad Weekly · Issue #001 · October 18, 2026
Issue #001 · Written October 18, 2026. Tools, prices and links may have changed since.
The one thing worth your time this week
AI saves you time. Then checking eats it.
The biggest AI complaint this month isn't that it's dumb.
It's that checking it eats the time it saved.
A Workday study estimated about 37% of the time AI saves gets lost to correcting, clarifying or rewriting what it produced. Nearly 4 hours of every 10. (Accounting Today)
An Adaptavist survey of 2,500 knowledge workers found 42% now spend longer verifying AI output than it saves them. (HR Review)
And a Foxit study found executives think AI saves them 4.6 hours a week… then they spend 4 hours 20 minutes validating it. (Same Accounting Today piece.)
Vendor surveys, so… grain of salt. But they all point the same way.
The problem isn't checking. Checking is the job.
The problem is checking everything, the slow way, every time. This week's lesson fixes that.
3 tools or updates
I saw these three this week. I haven't tried them. Here's what they claim… and the catch.
1. MyLens: AI that gives your presentation for you, one person at a time
What it is: Upload your docs or deck. It walks each invited person through it privately, answers questions from your sources, then tells you who got it and who's stuck.
Who it's for: Anyone who gives the same onboarding call, sales walkthrough or training session over and over.
The catch: It's presenting as you, in conversations you'll never see. If you try it, budget time to read those reports. Product Hunt lists free options; we couldn't find clear pricing.
2. SpeechFast: dictation for Mac that never leaves your Mac
What it is: Hold fn, talk, and cleaned-up text (fillers and false starts removed) lands at your cursor in any app. It all runs on your Mac, not a server.
Who it's for: People who live in email and Slack and talk faster than they type.
The catch: Mac only: Apple Silicon (M1 or later) on macOS 26. Clean-up needs Apple Intelligence turned on. It's paid after the free trial.
Source: https://speechfast.com
3. Tomo: turns any topic into a Duolingo-style course
What it is: Name a topic and AI builds a short course with hearts, streaks and quizzes. On iPhone, Android, or inside ChatGPT via the Tomo plugin. Free to start.
Who it's for: The "I should really learn X" crowd. Five minutes a day.
The catch: Nobody fact-checks an AI-written course but you. A streak will happily drill a wrong answer into your head. Worth a look only on topics where you'd spot a mistake.
Source: https://www.producthunt.com/products/tomo-for-chatgpt
The NotYourAIDad Lesson
Make AI show its homework
By the end of this you'll check an AI answer by testing the few parts that matter, instead of re-reading every word. About five minutes the first time.
1. Ask for the receipts. Right after any answer you're actually going to use, paste this:
Before I use this, list every fact in your answer that someone could check: names, numbers, dates, prices, quotes, links, and anything you assumed about me or my situation. Number them. Next to each one, say where it came from (something I gave you, a source you can name, or general knowledge) and mark it SURE or NOT SURE.
2. Circle what hurts if it's wrong. Money, dates, names, anything legal or compliance-related, anything going to a client. Usually that's a handful. Everything else gets a skim.
3. Check the circled ones somewhere else. The real source. Your own records. The contract. A calculator.
Not the same chat. Asking the same AI "are you sure?" isn't a check. An MIT paper cited by Accounting Today warns that AI checking AI can reinforce the same errors and give you false confidence.
4. Fix it, then re-check only the fix. Tell it exactly what was wrong and ask for a corrected version. Check that one item again. Don't start over.
5. Keep score for a week. Time with AI (checking included) versus doing it yourself. If checking takes longer than doing, that task isn't an AI task for you. Yet.
How you know it worked: you can point at every number and name in the final version and say where you checked it.
When not to use this: when a wrong answer is expensive and you can't check it yourself (medical, legal, tax, moving money). That's not a checking problem. That's a "call a human who knows" problem.
What everyone's getting wrong
The hype check: "AI agents will run your business while you sleep"
The news: Wix's Symphony agents can now pay supplier invoices and reorder stock for small businesses. Every purchase needs the owner's OK, by WhatsApp or text, within a 10-minute window. (Wix via GlobeNewswire) Meta's Muse for Small Business lets owners require approval before it makes purchases or publishes anything. (Retail Dive)
My take: Look at what the agent makers built in. A human checkpoint. Even they aren't saying "hands off."
The approve button isn't the safety net. You reading the request is. Tapping "yes" between appointments is just the checking tax… paid badly.
The reality check: the OECD's 2026 small-business survey found 61% of small and mid-size businesses use AI, but 76% of those are "AI novices" using simple off-the-shelf tools for one-off tasks. Only about one in five reported significant impact. (OECD)
Most people aren't behind on agents. They're still getting good at the basics. That's where the time is.
Your move (credit to Applied AI Insider for this one): before any tool gets to spend, write two lists. "Yes from my phone": repeat orders from known vendors, under a set dollar amount. "Wait till I'm at my desk": new vendors, odd amounts, anything you didn't expect.
One repo worth a look
Indra OCR
I saw this on GitHub. Haven't run it.
What it does: Free OCR (optical character recognition: reading text out of an image). It turns scanned PDFs and photos into clean text you can search or paste into ChatGPT, on an ordinary computer. No graphics card, no uploading your files.
Who'd use it: Anyone with a pile of scanned invoices, permits or receipts… who doesn't mind typing two commands.
The catch: Command-line only (you type commands in Terminal; no app window). Brand new (created Oct 13, one contributor), though it's from Datalab, a company that sells OCR. Its speed and accuracy claims are the maker's own tests. Free for personal use and startups under $20M in funding or revenue; bigger businesses need a license.
Link: https://github.com/datalab-to/indra
That's it
One lesson. Three tools worth a look. One repo.
Pick one AI answer you used this week and run step 1 on it.
Hit reply and tell me what it got wrong.
Try this, or don't… I'm NotYourAIDad.
Not financial, legal, or investment advice. Tools mentioned aren't sponsors.