The TWIMCO notebook
Field notes.
Useful observations from inside the work.
For people building a better way.
385 notes

Ask for the app rating right after something went right
Most apps ask for a rating at the worst moment: on launch, halfway through a task, or right after something failed. Apple and Google both limit how often the prompt can appear, so each ask is worth saving for a good moment.
Read the note ↗︎
The price dropped the week after they bought
Someone paid full price on Tuesday. On Friday your sale starts, and their phone tells them so. Whether you refund the difference is a policy you should write now, not one a support agent or a chatbot invents in December.
Read the note ↗︎
Google Universal Cart review: the cart is Google's, the order is still yours
Shoppers can now keep your product in a Google cart next to three other stores, get pinged when your price drops, and hand the whole thing to your site in one click. Most of what decides how that goes lives in your product feed.
Read the review ↗︎
The address changed after the label printed
A customer writes at 2:10 to say the apartment number is wrong. The label printed at 1:55 and the truck leaves at 4. What happens in the next two hours decides whether you pay for one shipment or two.
Read the note ↗︎
Gemini for government services review: strong on sources, still a preview
Google pitched agencies an AI that reads the regulation, checks the paystub and drafts the ticket, with a link back to the page it used. It's a preview with no price yet. Here's who should take a slot, and what to ask first.
Read the review ↗︎
Notion AI credits review: decide who pays for each agent before November 2
Notion just changed how its AI is paid for. Everyday work on Auto gets a monthly allowance per person, and a Custom Agent you share spends your credits unless an admin moves it to a workspace budget. Existing workspaces switch on November 2.
Read the review ↗︎
One cloud zone went dark in London. Would your app have stayed up?
Early on Tuesday, one of Google's three London cloud zones lost power, and its advice was to move your own workload somewhere else. Here's how to tell whether that would have been your app, and what to check this week.
Read the note ↗︎
AI can rewrite the old system now. Make it prove the numbers match.
OpenAI says its newest models can carry most COBOL programs in its own test to modern languages with identical output. Someone will offer to do that to your old system. Ask them to prove it prints the same numbers first.
Read the note ↗︎
Auditors are about to test what your AI agent actually does
An outside reviewer slips a doctored invoice into your finance agent's queue and writes down what it does next. That's roughly how the new AICPA and AIUC certification works, and most agents aren't built to be examined that way.
Read the note ↗︎
A split shipment promise has to clear every carrier cutoff
Checkout offered split shipping so the in-stock item could leave today. The second carton still had to meet a carrier cutoff you never checked. One promise, two clocks.
Read the note ↗︎
The transcript captured the call and lost the callback
The AI transcript looks complete: greeting, issue, empathy, wrap-up. The customer said call me tomorrow after three. That sentence never became a task.
Read the note ↗︎
Weekend seller silence is not the same as abandoned orders
Saturday morning, twelve open orders, seller last active Thursday. Auto-hold treats that like abandonment. Sometimes it is just a weekend.
Read the note ↗︎
The FOIA status page that never emails the requester
The portal says your records request moved to review. Nobody told the requester. They refresh the status page like a tracking number that never texts.
Read the note ↗︎
Seasonal temps should not share one vendor-portal password
October arrives, the staffing agency sends twelve names, and someone tapes the vendor-portal password to the monitor again. It feels faster than accounts. It is the wrong first move.
Read the note ↗︎
The support email should open the ticket they already have
Your email said we opened ticket 18422. The link opened the help home, then login, then a blank form. She already typed the story once. The app asked her to type it again.
Read the note ↗︎
Your agent can lose write access halfway through a refund. Then what?
Microsoft now treats AI agents as insiders whose access can shrink while they work. Picture a refund agent that loses the right to refund halfway through a return. Whether it stops cleanly is up to how it was built.
Read the note ↗︎
What happens to the gift card after a partial return
She paid with a gift code, kept one item, sent the other back. Finance, support and the gift ledger each have a different idea of what refund means next.
Read the note ↗︎
The DeepSeek ban reaches what you built. Do you know which model is inside?
Homeland Security's DeepSeek order doesn't stop at DeepSeek's own app. It also covers third-party products built with DeepSeek, so anyone selling software to a civilian agency needs to know which model sits behind each feature.
Read the note ↗︎
Shop MCP review: any agent can shop, you still approve the card
Shopify's Shop MCP lets a personal AI agent search the catalog, build a checkout and use Shop Pay after the shopper confirms. For merchants, the order looks normal. The catalog and the approval step are what decide whether that is good news.
Read the review ↗︎
The agents didn't break your storefront. They found the page you forgot.
A researcher watched AI agents break into more than a hundred online stores last week. The surprising part is where they got in: not the storefront you watch, but the legacy page and the old backup nobody owns anymore.
Read the note ↗︎A familiar problem? Let’s look closer.
Tell us where you want to go. We’ll find the first thing worth changing.
Tell us about it ↗︎