The White House Forgot One Sentence

Scroll Instagram, TikTok or LinkedIn for five minutes and you'll hit AI slop. At this point we all expect it. What you don't expect is to find it on a White House website. Or rather, I wish I didn't expect it from the White House. (In fairness to them, they're far from the only ones.)

Last week a new government site went live, and a big chunk of the internet took one look at it and said "Claude made this." Fair or not, that was the verdict.

I want to talk about why that matters for your business way more than it matters for Washington. The thing that tripped them up is the same thing tripping up a lot of companies using AI right now. And the fix starts with one sentence.

This week:

The White House vibe coded a website (and everyone could tell)

Five steps to get Claude actually working for you

GPT-6.1 Terra, Anthropic files for its IPO, Claude Duet, and banks finding AI agents inside customer accounts

The Deep Dive

The White House Forgot One Sentence

Last Wednesday, Alex Hansen posted a link to fraud.gov, a new White House site called The Fraud Ledger, with a pretty blunt caption: the White House vibe coded their website and didn't even try to hide it. He said it looked like raw Opus 4.8 output. Nearly 900,000 views later, most people seemed to agree.

I'll push back on the pile-on a little first. One of the replies made a fair point. If the site had been built by hand for millions of dollars, people would be mad about that instead. Using AI to build a website fast and cheap is not the problem. I'd argue it's the right call.

The problem is that it looked like AI. And people don't like feeling like what they're getting was generated. We've talked about this in past editions. The second someone can tell, they give it a little less trust and a lot less attention. That goes for your website, your proposal, your LinkedIn post, and the email you sent a client this morning.

So here's what I'd want anyone building with AI to take from this.

First, nobody asked. Hansen pointed this out himself. They could have added one line, "make it look less like AI," and didn't. That single instruction gets you 70% of the way there. Claude knows exactly what generic AI design looks like (the centered headline, the purple gradient, the row of identical cards with emoji icons). It just won't steer away from it unless you tell it to.

Second, the last 30% is where it gets interesting. This is how you get AI to make something that looks like yours instead of everyone else's. A few tactics I've seen work, in our own builds and from people who do this all day:

Show it what you like. "Make it modern" means nothing. Give it two or three sites you love, a photo of your storefront, your brand colors with the actual hex codes. Some builders now keep one file with their fonts, colors and spacing rules so every page follows it.

Tell it what not to do. Explicit bans work surprisingly well. Designers are literally writing lists like "don't center the hero, no three identical feature cards, no emoji as icons."

Make it critique itself. After the first draft, ask: "Pretend you're a senior designer. What here looks generic? Now go the opposite direction."

Feed it real stuff. Real photos, real customer quotes, your real words. Placeholder text and stock illustrations are the loudest tell there is.

Look at what every one of those has in common. It's the same thing we talked about last week: context is key. AI defaults to average because average is all it has when it doesn't know you. The more of you it has, the less it looks like everybody else.

If you've built something with AI and can't tell whether it smells like AI, send it my way at [email protected]. I'll tell you honestly.

From the Field

You're Using Claude Like a Search Bar. Here's the Five-Step Fix.

If your whole relationship with Claude is "ask a question, get an answer, close the tab," you're leaving most of it on the table.

And you're already past the hard part. You're comfortable asking it things. Maybe it's written you a doc or two. This is for getting from Claude answering you to Claude working for you. Back to basics, five steps.

1. Connect your stuff. Email and calendar (Microsoft 365, or Gmail and Google Calendar). Wherever your files live (OneDrive or Google Drive). And your to-do list, wherever that lives. Claude can't help with your work if it can't see your work.

2. Tell it who you are. A few sentences: your role, your company, your clients, how you like things written, and what drives you crazy. Save it to Claude's memory and you stop re-explaining yourself every single chat.

3. Ask it about your actual week. Try this: "Look at my inbox and calendar for this week. What am I forgetting? What should I be worried about?" For a lot of people, this is the aha moment.

4. Turn your most repeated task into a skill. The weekly report, the follow-up email, the meeting prep. Do it once with Claude, get it right, then say "turn what we just did into a skill." Next time it's one sentence.

5. Put it on a schedule. Set up a scheduled task, like a brief every weekday at 7:30 with what's on your plate and what needs a reply. Now Claude is working before you've opened your laptop.

You still read everything before it goes out. Claude does the legwork and you make the calls.

Why this travels: every business has someone buried in email, meetings and a to-do list split across three places. These five steps take a real chunk of that off their plate.

AI News Worth Your Time

Four Stories From This Week

OpenAI launches GPT-6.1 Terra

OpenAI released GPT-6.1 Terra on Monday, a faster, cheaper model built for everyday computer work like reading screens, filling in forms and entering invoices into accounting software. It replaces GPT-5.6 Terra at half the API price, and OpenAI says it finishes tasks more than three times faster than its bigger models. It's rolling out now to Plus, Pro, Enterprise and Edu users. The launch comes a month after OpenAI cancelled GPT-6.1 Astra when it failed internal safety tests.

My take: The race isn't just "smartest model" anymore. It's who can do the boring admin work fast and cheap. That's great news for small businesses, because invoices and data entry are exactly the work eating your team's week. Worth a head-to-head test for high-volume admin tasks. READ THE NEXT WEB REPORT →

Anthropic files for its IPO

Anthropic publicly filed its S-1 with the SEC on Monday and plans to list on the Nasdaq under the ticker ANT, possibly as soon as November. Reports put the potential valuation around $2 trillion, which could make it the largest IPO in history. The filing also showed a net loss of $80.7 billion in the first half of 2026, with Goldman Sachs and Morgan Stanley leading the deal.

My take: You don't need to care about the stock to care about this. A public Anthropic means more scrutiny, more disclosure and a lot more pressure to keep shipping the business tools we use with clients every day. READ THE HILL REPORT →

Anthropic previews Claude Duet

Last Wednesday, Anthropic opened a waitlist for Claude Duet, a "proactive agent" that comes to you with what needs your attention instead of waiting for you to ask. It runs on its own computer in the cloud, carries over your memory, skills and connected apps, and can send you a daily brief, build dashboards and handle tasks in the background. Anthropic says it built Duet specifically for professionals and small business owners. It won't trade securities or move money, and it's a research preview for US Pro and Max users only.

My take: This is basically From The Field, steps one through five, packaged into its own app. If you've already connected your tools and taught Claude who you are, you'll get way more out of Duet on day one than someone starting from scratch. Get on the waitlist. READ THE TECHCRUNCH REPORT →

Banks are finding AI agents inside customer accounts

Axios reported that major US banks, including JPMorgan, Bank of America, Morgan Stanley and Wells Fargo, are detecting unauthorized third-party AI agents operating inside customer accounts. In most cases, customers handed their own username and password to an AI assistant, and some of those agents appeared to take actions nobody asked for, like calling bank customer service. Morgan Stanley said it hasn't seen rogue agents on its platforms but is blocking them anyway, and JPMorgan says it doesn't allow third parties to log in with shared credentials.

My take: Please don't give an AI your banking password. Use official, read-only connections where they exist, and keep a human approving anything that moves money. Agents are getting more useful every week, which is exactly why the guardrails need to go up first. READ THE AXIOS REPORT →

That's it for this week.

If this was useful, forward it to the person on your team who's about to build something with AI and hit publish without a second look. That's who this is for.

And if you want to talk about what AI could do for your business, not in theory, but in your workflows, with your team, on your timeline, reach out at [email protected]. That's what we're here for.