Alright, let’s talk AI.

This newsletter is my lab notes….the rants, the experiments, and the tools that actually matter. Some weeks it’s spicy, some weeks it’s tactical, but always stuff you can steal and use.

This round? A roast of creepy AI video, why 95% of AI projects are failing (hint: it’s not the tech), and the vanity move that’s tanking companies’ traffic. Plus a few tools that are actually worth the clicks.

Let’s get into it.

AI video is creeping in. And it’s not the cool kind.

I’m not talking about VEO 3 or gorilla clips on TikTok. I’m talking about those AI videos on websites that turn into a staring contest...unblinking eyes boring into your soul until you finally click “Talk to me.”

It’s not working. It can’t be.

Every time I muster the courage to engage one, it feels like a Chucky reboot. Laggy, dead eyes, and the mouth doesn’t even sync with the voice track.

Look, I’m an AI guy. I believe in the tech. But this isn’t “cutting edge,” it’s the sales equivalent of garlic breath. Slap this on your homepage and you’re not innovating, you’re scaring people away. Kill it.

Maybe wait a few versions until the tech actually works… or better yet, find a different use case… like scaring kids at Halloween.

Because right now, your creepy avatar isn’t helping. It’s costing you deals.

MIT says 95% of AI projects are failing. McKinsey says 80% of companies using GenAI can’t find a single dollar of impact on their P&L.

That’s not because AI is broken. That’s because we are.

Here’s how most companies “do AI”:

▸ We drool over shiny demos while the real problems rot in the corner.
▸ We duct-tape AI to broken workflows like it’s going to magically fix stupid.
▸ We let middle managers stall adoption because they’re scared.
▸ We run “pilots” because blowing up the old system sounds scary.
▸ We hand AI to teams so far from revenue they need GPS to find a customer.
▸ We feed AI garbage data and act shocked when it spits garbage back.
▸ We measure adoption in logins, not outcomes, and high-five ourselves for it.
▸ We build in-house even though external partners succeed twice as often.
▸ We let one exec “own it” and kill adoption everywhere else.
▸ We sprinkle AI on top of broken workflows instead of rebuilding the workflows with AI at the center.

Then we sit in board meetings looking surprised, as if we didn’t design the mess ourselves.

MIT calls this the GenAI Divide. On one side, the 5% of companies pulling millions in ROI. On the other, the 95% stuck in pilot purgatory. The difference isn’t the tech. It’s the choices.

The winners are playing a different game:

1. Buy, don’t build.
MIT found internal builds fail twice as often. But we still throw armies of engineers at “AI projects” that never escape pilot phase. The winners skip the ego trip, partner with vendors, and get to ROI while everyone else is still arguing about architecture diagrams.

2. Custom beats generic.
A $20 ChatGPT subscription outperforms a $200K enterprise “solution” because at least it listens. The companies crossing the divide aren’t buying chatbot wrappers, they’re embedding AI deep into workflows, contract review, AP/AR automation, support routing, stuff that actually moves the P&L.

3. The real ROI is boring.
McKinsey shows 50–70% of AI budgets go to sales and marketing because it looks sexy on slides. But the actual wins? Killing BPO contracts, slashing agency fees, automating back-office sludge. Quiet, unsexy, million-dollar outcomes while everyone else is chasing vanity pilots.

4. Speed is the moat.
Mid-market companies are getting pilots to production in 90 days. Enterprises take nine months or more. Every extra month you stall, switching costs go up and your competitors are already locking in vendor relationships you won’t unwind for years.

5. Distributed, not centralized.
Letting one exec or an “AI Center of Excellence” own everything is a death sentence. The companies crossing the divide push adoption to line managers and power users who actually know the workflows, while still holding them accountable to business outcomes.

And here’s the kicker: employees already know what good AI feels like. MIT found 90% of workers are using ChatGPT or Claude on their own, while only 40% of companies even bother to pay for it. This “shadow AI economy” is delivering more ROI than the official programs. Workers are crossing the divide. Leadership is stuck behind.

So let’s stop blaming the tech. AI isn’t failing us. We’re failing AI.

The companies that get this right won’t just save time. They’ll change the scoreboard. They’ll slash costs in places you never looked, unlock revenue streams you didn’t think possible, and quietly take your customers while you’re still running pilots and high-fiving “login numbers.”

The GenAI Divide is real. You can either cross it, or get crossed off.

Should you torch your .com for .ai? Gladly did. Bold? Yes. Smart? No. This isnt a rebrand. It is a vanity play.

You give up prime digital real estate, .com, for what? To trade Park Place for Baltic Ave and impress the AI cool kids at lunch. Let’s call it what it is, a Hail Mary to look modern without doing the work.

Here’s the lie companies are telling themselves:
“If we switch to .ai, buyers will think we’re more AI-native.”

No they won’t.

They’ll still judge you based on product depth, proof points, and whether your AI actually works, not your URL.

Meanwhile, the real cost?
▸ 40% drop in organic traffic
▸ SEO juice flushed overnight
▸ Months of technical lift across dev, ops, and QA
▸ No measurable lift in buyer perception

And good luck reversing it if things go sideways We’ve seen this before. WooCommerce did the same stunt. It tanked them.

Switching to .ai doesn’t make you an AI company. Just like wearing scrubs doesn’t make you a surgeon.

The upside’s here is feeling. The downside’s on the balance sheet. Show me one buyer who decided you were AI-native because of a URL. You can’t. Because they don’t exist.

Want to reposition around AI? Cool.

Then earn it.

▸ Ship real AI
▸ Train your team
▸ Prove it with outcomes
▸ Back it with customer wins
▸ And for the love of everything that is holy, nail the damn narrative.

But don’t trade real traffic for imagined credibility. I don't think this was bold. I think it was reckless. Disagree? Fine. The numbers will decide but so far, it’s not looking good.

In the meantime, enjoy the vanity tattoo. That traffic ain’t coming back.

Props to Gaetano for breaking this down, been waiting for this one to drop. Worth the read:https://lnkd.in/gGmnsuXK

Willow Voice
Talk instead of type. Dictates emails, Slack notes, or docs about 3x faster, and even cleans up the formatting for you.

Command A Reasoning
A powerhouse model that can write, summarize, and analyze right on your own servers. Open weights, multilingual, and built to handle heavy lifting.

Mentionlytics
Handy for keeping tabs on your brand, track sentiment, mentions, and chatter across social and news.

Grammarly
Still the go-to for fixing tone, grammar, and clarity. Reliable, if not flashy.

Fyxer.ai
Your AI personal assistant for scheduling, inbox triage, and the admin junk that usually eats your mornings.