Last week, hundreds of people on X confidently roasted a painting for being obvious AI slop. The brushwork. The composition. The "missing soul." A few even ran it through AI detectors that confirmed their suspicions.

Last Tuesday, an X user named SHL0MS shared one of Monet's Water Lilies paintings and labeled it "Made with AI." He asked people to describe, in as much detail as possible, what made it inferior to a real Monet. Here is the post.

The replies came fast and confident. "No cohesion of elements." "Looks like high school art." "The brushwork is wrong." "The reflection is just noise." "There is no soul." A few accounts even ran the image through "AI detectors" and screenshotted the high "AI-generated" probability scores. Then the reveal. Real Monet. Painted in his garden in Giverny over a century ago. There was an audible flinch on the timeline.

It’s a funny story on the surface. Easy to write off as haters being haters. But sit with it for a minute and it starts to look like one of the most important things that happened in AI this month... and almost nobody is treating it that way. The same painting that read as masterful yesterday read as AI slop today. Not one pixel changed. The label did the work, not the image.

There is a name for what happened here. It is called the toupee fallacy. The classic version goes: "I can always spot a bad toupee." Well, of course you can. The only toupees you ever spot are the bad ones. The good ones, by definition, slip past you without you noticing. So your confidence that you can always tell is built on a sample of just the failures.

It is the same with AI. When people say "I can always tell when something is AI-generated," what they really mean is they can spot the bad AI. The clumsy stuff. The em dashes that show up uninvited. The opening that sounds like ChatGPT cleared its throat. The good AI, the kind that has been edited, shaped, and put through a real human filter, sailed past them. They never knew.

This is what I have started calling the AI Trap. As a company, you absolutely should be using AI to get more done. The output gains are real. We see them on every engagement. But the moment your client, your prospect, or your team starts feeling like the work came from AI instead of was helped by AI, you lose trust. And trust is the entire game.

The two failure modes I see most often. One, people copy and paste raw AI output into emails, decks, and proposals without putting themselves back in. Two, people get so excited about speed that they stop reviewing. They send the thing the second it comes out. Volume goes up. Care goes down. Clients feel it.

A specific version of the second one that I see all the time. Someone asks an AI to "put together a guide to X" and gets back a 30-page document complete with eight headers, four nested sub-lists, and a glossary at the end. They send it. After scrolling through 30 pages they will never read, the same suspicion creeps in that the X timeline had about the Monet. Did you make this, or did the AI? The honest answer is usually obvious from the volume alone. Nobody would have written that much by hand for a simple report. A real human writing on a deadline would have cut it to a tight three pages. The 30 pages are not a sign of effort. They are a sign that AI ran for free and nobody bothered to edit.

The way out is not to use less AI. It is to use AI better, with a human still very visibly in the loop. Use it for the heavy lifting. The research, the first draft, the outline, the analysis. Then put yourself back in for review and touch-ups. Cut what you would not actually say. Tighten what is too long. Add the one specific story or detail nobody else has, and then update your AI to be better next time.

Used right, AI does not make your work look like AI. It makes your work look like more of you, faster.

If you are worried your team is falling into the trap, or you want to figure out where AI fits into your workflow without sacrificing the thing that makes your work feel like yours, just reply. That is the conversation we want to have.

Here is a small Claude/ChatGPT workflow that has quietly saved me hours every week.

If you keep your Outlook (or Google) calendar even halfway up to date with meetings and the deep-work blocks you actually do, you can set up a scheduled task in Claude or ChatGPT that runs every morning or every night. It reads yesterday's calendar, classifies each block by client and type of work, and logs your hours straight into Notion (or whatever doc system you use).

Two reasons this is worth fifteen minutes of setup. One, billing gets dramatically easier. End of month, your hours are already there, broken out by client. Two, you can actually talk to Claude/ChatGPT about where your time is going. "Where did I spend the most hours last week?" "How much time is going to client X vs internal work?" "Which day of the week am I most fragmented?" That kind of conversation is hard to have with a static time-tracking app.

No new software. No extra subscription. Just your calendar, AI, and the doc system you already use. If you want a hand setting one up for your team, reply and I will walk you through it.

Anthropic just passed OpenAI in business adoption for the first time

Ramp published their May AI Index this week, and the headline is the one I have been waiting on for a year. Anthropic now leads OpenAI in paid business adoption. 34.4% of businesses on Ramp's platform are paying for Anthropic. 32.3% are paying for OpenAI. Anthropic grew adoption 3.8 points in April alone. OpenAI dropped 2.9. Over the last year, Anthropic has roughly quadrupled its business share. OpenAI grew by 0.3%. The Ramp dataset covers spend across 50,000+ companies, so this is not a vibes survey. It is actual money changing hands.

My take: This is the data catching up to what we see every day in the field. Business owners do not want a flashy chatbot. They want an AI that does real work, plugs into the tools they already use, and does not creep them out. That is the lane Claude has been in from day one, and the lane Bridge AI sits on top of. Will OpenAI fight back? Yes. Codex is good and getting better. But for the kind of serious workflow build-outs we do for clients, Claude has become the obvious choice, and the spend data now agrees.

READ THE RAMP AI INDEX →

Anthropic just shipped a product aimed directly at small businesses

On May 13, Anthropic launched Claude for Small Business. It is a toggle inside Claude Cowork that ships with 15 prebuilt agentic workflows (payroll planning, month-end close, invoice chasing, campaign management, cash-flow forecasting) and connectors to QuickBooks, PayPal, HubSpot, Canva, DocuSign, Google Workspace, and Microsoft 365. No new SKU. No extra cost. It just turns on inside the Claude Cowork license a business already has. Alongside it, Anthropic kicked off the Claude SMB Tour, a free 10-city workshop run with Tenex.co (Chicago, Tulsa, Dallas, Hamilton Township, Baton Rouge, Birmingham, Salt Lake City, Baltimore, San Jose, Indianapolis), with a one-month Claude Max included for attendees.

My take: This is the announcement Bridge AI has been waiting for. Small and mid-size businesses just got a frontier-lab product built for them, with the workflows they actually run and the apps they already pay for. The catch is that the product alone is not the thing that creates value. Using Claude for Small Business is the easy part. Getting your team to actually save significant time on the work they actually do, with the standards your clients actually expect, is the hard part. That is the implementation layer. That is the work.

READ THE ANTHROPIC ANNOUNCEMENT →

OpenAI brought Codex to your phone

On May 14, OpenAI rolled Codex, their cloud-based AI coding agent, into the ChatGPT mobile app. The pitch is real and useful. You can now monitor a long-running coding task from your phone, approve agent decisions, review diffs, redirect a stalled task, all from wherever you happen to be. It is in preview, rolled out to every ChatGPT plan including the free tier, on iOS and Android. The catch is the host machine has to be a Mac for now, with Windows support "coming soon."

My take: The actually-interesting part of this story is the timeline. Anthropic shipped Remote Control for Claude Code in February. OpenAI shipped this on May 14. Four months later. That is the pattern of the year so far. Anthropic ships the thing first, OpenAI ships it next with more polish on the consumer side. It is genuinely good for the field. Competition pushes both sides to ship faster. But if you are running a business and trying to decide where to put your AI bet, the lab that defines what these agents do first is the lab worth paying attention to.

READ THE TECHCRUNCH STORY →

That's it for this week.

If you found this useful, forward it to someone running a business who is using AI but quietly worried their work is starting to feel like AI.

And if you want to talk about what AI could actually do for your business, not in theory, but in your workflows, with your team, on your timeline, just reply to this email. That's what we're here for.