I've been building with AI daily for almost a year. Here's what I believe that most people don't want to hear.
1. A lot of people try AI once and give up. That's the product's fault, not theirs.
I was in this boat for a long time. Tried it, didn't get it, only used it sparingly for email, writing, and low value tasks. I wasn't impressed with the enterprise solutions roled out to me in work settings and I couldn't see the value in my personal life. It took a lot of learning and exploring to unlock real value. The learning curve is real and nobody's solving it well yet. Your friends who "don't like AI" aren't wrong—they just haven't found their way in.
2. AI needs to learn conversation, not just prompts.
Right now AI is trained to answer sophisticated questions. But that's not how humans actually communicate. We use tone, context, half-finished thoughts, sarcasm. When AI moves into robots and voice interfaces, this gap becomes a real problem. It needs to understand how humans talk, not just what they ask, and how to probe for more information. This is a real barrier for adoption today, because if you aren't writing good prompts or know how to work with AI, you may be unimpressed with the results. (see my prompt tools to help!) The next layer of usefulness comes from AI and humans better collaborating.
3. Different AIs are good at different things—and nobody talks about this.
Gemini is like a search engine. Really good when you ask one specific, well-crafted question. ChatGPT is chatty—great for writing, but can go in circles and never get anything done. Claude is my favorite, but it takes real investment to train it with context. There's no "best AI." There's the right AI for the right job.
4. Companies are implementing AI completely wrong.
A generic chatbot slapped on your website isn't innovation. It's lazy. Same with deploying one AI to your entire company without training. That's not helpful—that's a search bar with extra steps.
5. Everyone should have their own AI, trained to them.
Not one company AI. YOUR AI. Trained on your role, your context, your communication style. The productivity gains don't come from AI existing—they come from AI knowing you.
6. Different functions need different AIs.
Marketing AI should think differently than Engineering AI. Product AI should think differently than Finance AI. We're treating AI like one tool when it should be a specialized toolkit.
7. AI companies are rushing to monetize before earning trust.
Most people still don't understand what AI can do. Jumping straight to ads, subscriptions, and commerce is going to confuse people and scare them away. You can't monetize confusion. Get people on board first.
8. Ads and shopping in AI chat are going to be terrible for a long time.
Commerce bolted onto conversation is clunky. AI-powered shopping isn't going to work until websites build proper metadata for AI to actually read. We're a long way away from this being good.
9. AI making you efficient shouldn't mean you still work 40 hours.
If AI makes me 3x more productive, I don't want to do 3x more work. I want to do the same work in a third of the time. The goal is less work, not more output.
10. Two-week sprints and quarterly planning might be obsolete.
When you can prototype in an afternoon, rigid planning cycles feel absurd. The PM frameworks we learned are built for a slower world. I've been experimenting with what comes next—more on that here.
The real takeaway
Right now, AI takes investment to make it work for you. That's friction. That's a learning curve. But it won't always be this way.
The tools will get better. The conversation will get smoother. The personalization will come built-in instead of hacked together.
This is the messy middle. The people building through it now are going to have a massive head start when it clicks for everyone else.
Agree? Disagree? I'm always curious what other builders are noticing.