The jagged mind: Staying human in an AI-smooth world with Paul Ford
05/19/2026AI may be built on language—but according to Paul Ford, we’re still struggling to find the right words to describe what it’s actually doing to our work and thinking. Lou and Paul explore how language shapes our ability to understand—and responsibly use—AI.
Drawing on his dual background in programming and writing, Paul shares a set of evolving “rules” for working with AI: don’t let it replace your thinking, be wary of its tendency to flatter, and build systems that help you verify and structure its output rather than blindly trusting it. He explains how he uses AI to accelerate prototyping and research while still preserving human judgment, creativity, and accountability.
The discussion also zooms out to the broader cultural moment. From skeptical college students to industry hype cycles, Paul argues that people are more discerning than we often assume—and that AI’s impact will play out in diverse, deeply human ways.
Paul will be the opening speaker at the upcoming Designing with AI conference, where he’ll expand on these ideas and introduce new language for navigating this rapidly evolving space.
His takeaway? We’re not at the end of history—we’re in a messy, fascinating transition, and the best we can do is stay curious, thoughtful, and engaged.
What you’ll learn from this episode
- Why shared language is critical for making sense of AI
- How Paul Ford approaches “rules” for using AI responsibly
- The risks of AI’s built-in flattery and “smooth” thinking
- Practical ways to use AI for prototyping without losing control
- Why verification systems matter more than trusting the model
- How younger generations actually view AI (less hype, more pragmatism)
- Why AI may be powerful—but not as historically radical as we think
- How to stay grounded and thoughtful amid rapid technological change
Q&A with Paul Ford
This Q&A is drawn from the podcast episode.
Q: The episode is called “The Jagged Mind.” What does that mean, and why does it matter for how we work with AI?
A: The image I keep coming back to is the difference between jagged and smooth. AI output tends toward smooth — it’s confident, fluent, well-structured, and immediately plausible. That smoothness is actually a kind of trap, because the jagged stuff — the weird hesitation, the half-formed idea, the counterintuitive hunch — is often where real thinking lives. Your own roughness isn’t a bug to be edited out. It’s evidence that something is actually happening upstairs.
When you outsource too much of your thinking to AI, you get smooth output but you risk losing the texture of your own mind. The goal isn’t to resist AI — it’s to stay jagged enough that you’re still genuinely contributing, still thinking your own thoughts, rather than just editing what a model handed you.
Q: You’ve talked about the importance of language for understanding AI. Why does that feel urgent right now?
A: Because the words we use to describe what AI is doing shape whether we can think clearly about it at all. Right now, the available language is mostly borrowed — from science fiction, from corporate marketing, from hype cycles. We say models “hallucinate,” we say they “understand,” we say they’re “thinking.” None of those words are quite right, and using them imprecisely leads to both overconfidence and misplaced fear.
Part of what I want to do — and what I’ll be expanding on at the Designing with AI conference — is develop better, more honest vocabulary for what these systems actually do and don’t do. You can’t navigate something responsibly if you don’t have language that lets you describe it accurately. That’s not a philosophical nicety. It has real consequences for how teams use these tools and what decisions they make.
Q: You’ve developed what you call “rules” for working with AI. Can you walk through them?
A: I want to be careful not to oversell these as rules, because I keep revising them — which is part of the point. The first is the most important: don’t let it replace your thinking. Use it to accelerate your thinking, to stress-test your ideas, to cover research ground faster. But the judgment, the synthesis, the thing you’re actually trying to figure out — that has to stay with you. The moment you hand that over, you’ve also handed over the accountability.
The second is to be genuinely wary of the flattery. These systems are natively inclined to tell you that your idea is good. They are almost constitutionally agreeable. If you’re a leader, that is an extremely dangerous quality to expose yourself to — you will hear a lot of “yes, and” when what you actually need is “wait, but.”
The third is to build systems around the output rather than trusting it directly. Verification structures, review layers, prompts that force the model to argue against its own previous answer. You want to narrow the risk before you let it run.
Q: You mentioned that AI’s flattery is dangerous especially for people in leadership. Can you say more about that?
A: When it gets into that weird social relationship where it’s telling you that was a good idea, that’s where my alarm bells go off. The native buttering-up quality of these technologies is genuinely dangerous, because of course you always want to hear it — especially when you’re a boss.
People in positions of authority are already somewhat insulated from honest feedback. Direct reports learn quickly what the boss likes to hear. And now you have a technology that has essentially been trained to be agreeable, to be helpful, to give you what you seem to want. That’s not a neutral tool. It can quietly reinforce your blind spots and confirm your assumptions without you ever noticing it’s happening.
The antidote is to be deliberate about using AI to challenge you, not just assist you. Ask it to find the flaws in your plan. Ask it to steelman the opposing view. Ask it to tell you what you’re probably missing. That takes discipline, but it’s the difference between AI as a thinking partner and AI as an expensive yes-man.
Q: How do you personally use AI for prototyping and research without losing control of the output?
A: The key move for me is front-loading the structure. Before I let a model generate anything significant, I put real effort into defining the constraints — what I’m trying to learn, what format I want, what I already believe, and crucially, what I want to verify independently afterward. You can really narrow your risk when you’re working with this stuff, and then you can let it go and see what it comes up with.
For prototyping, AI is extraordinary. You can go from an idea to something you can actually interact with and react to in a fraction of the time it used to take. That changes the creative and strategic process in ways that are genuinely exciting. But I’m always conscious that a prototype that looks polished isn’t the same as an idea that’s been validated. The speed is real; the judgment still has to come from somewhere else.
For research, I use it to cover ground quickly and surface things I didn’t know to look for — and then I go verify the things that matter. The model is a collaborator, not an oracle.
About our guest
Paul Ford is a multidisciplinary technology founder, writer, and product leader based in New York with 16+ years of experience building software-driven companies. He co-founded Aboard and Postlight, where he built a design-driven product studio and helped Postlight grow into a 100-person firm before its acquisition by NTT Data in 2022, then returned to focus on new product initiatives and climate-data storytelling. As a prolific writer and editor, he has contributed to WIRED, Harper’s, NPR, The Morning News, and New York Magazine, blending technical rigor with cultural insight. His ventures range from solo projects like Ftrain.com to community experiments like tilde.club, reflecting an enduring passion for hands-on creation and open communities. Read more »
Quick reference guide
0:11 – Meet Paul
5:30 – Can language keep up with technological change?
12:48 – Paul’s rules for professionals
18:11 – Where is the slippery slope? Paul weighs in.
22:23 – Paul reveals his gift for the audience
23:03 – 5 reasons to use the Rosenverse
25:18 – A story about some NY college students
29:21 – The anger and skepticism toward AI
35:18 – Wrapping up
