Right, I want to tell you why I'm building a silent-speech app, because it's personal. About 14 years ago I had a stroke. It affected my right side. I walk with a stick now, and speaking is hard for me. I'm also dyslexic, so typing isn't always easy either. Think about it this way: most people talk to AI by typing or by voice. If both of those are hard work, you miss out on a lot of what AI can do. So I started wondering. What if the computer could just read your lips, with no sound at all? I'll be honest, I don't know yet if it'll work for everyone. Everyone's mouth moves a bit differently, so I'm training it on my own face first. So far it's recognised short phrases like "What's the time" and sent them straight to Claude. It's early days. But for the first time in a long time, I'm building something instead of watching other people do it. If you know someone who finds speaking or typing hard, I'd love to hear what would actually help them. If not, no worries at all. Thanks for reading.

A no-code exercise: submit a fictional receipt twice, once as a photo and once as a PDF. Filename matching misses the duplicate. When evaluating OpenClaw, I would compare merchant, date, amount and reference, then leave ambiguous matches for review. Similar totals alone would not justify discarding a legitimate expense. Present the sample as a customer request, then as an owner instruction. Discuss whether the source changes what the operator may do. Which receipt field would you trust most for a duplicate check? What would that additional check change in your decision?

I’m thinking through a simple inquiry tracker for a small coaching business. The workflow can look healthy while a new inquiry never reaches the list. Before relying on it, I’d want to compare the original inquiries with the records created, then check that each one has a next action. A successful run message alone would not tell me whether something was missed. For those running automations in real businesses: what is your lightest useful check for silent failures? A daily count, a test record, an exception list, or something else? Interested in what you actually use and who checks it—especially when the owner is busy delivering the service. Please keep customer details out of examples.

Name: Jorge Martinez What you do: Own an expediting cargo business. What you want to build: I would like to build an all-tracking system that tracks, creates invoices, emails letters, responses to calls & emails replies and more. Who you want to collab with: ANYONE! - Would love to learn from others and grow.

In June 2026, expense software company Emburse published a survey of 2,000 workers in the US and UK. Four in ten US employees admitted using AI to generate a fake receipt for a business expense. Of those who did: - 19% used AI to fabricate a purchase entirely - 15% used it to inflate the value of a real one - 40% used AI tools their employer was paying for It is a vendor survey, so treat the exact numbers with care. But the direction is hard to argue with. A convincing receipt now takes seconds, and the old tells, such as odd fonts or a blurry photocopy, are disappearing. I trained in audit at PwC and KPMG. One principle from that world matters more than ever now: evidence from an independent source is more reliable than evidence produced by the person making the claim. A receipt is evidence from the claimant. A card transaction comes from the bank. That difference is now your best control. Five changes any business can make this month, no new software required: 1. Move to company cards where you can. The transaction record is created by the card provider, not the employee. 2. Match receipts to bank or card data, not to how real they look. If you cannot trace it to a transaction, treat it as unverified. 3. Update your policy wording. If it only mentions "altered" receipts, add "AI-generated or digitally fabricated". 4. Never change a supplier's bank details from an email. Call a number you already had on file, not one in the request. 5. Split the jobs. The person who adds a supplier should not be the person who approves the payment. Notice that none of these try to spot the fake by eye. That fight is already lost. The answer is to stop relying on the document and start relying on the source. AI makes forgery cheap. Good controls make it pointless. Which of these five does your business not have yet?