As a PM, working with stakeholders and teammates is natural. It’s most of the job. So when AI became genuinely useful as a thinking tool, something I could bounce ideas against at 9 a.m. or midnight, it slipped into my workflow faster than I expected. That became especially valuable in design, which is the part of the work I’ve often had to reason through without a dedicated team around me.
A year in, I’ve tried a lot of AI tools. Some stayed. Some didn’t. Some are still on probation. Over time, I noticed something interesting: the tools that lasted weren’t always the ones with the biggest feature lists. They were the ones that matched how I naturally think and work.
That’s probably why I stopped treating AI tools as interchangeable.
This post isn’t really about which model is “best.” It’s about how different systems shape different kinds of thinking, and how that changes the way I work.
Not a verdict. Just observations from someone spending a lot of time inside these systems.
The one I keep coming back to
Claude Adaptive is where I reason through problems
If I had to keep one tool, this would probably be it.
Claude Adaptive maps unusually well to how I think. I can ramble at it. Change directions halfway through a sentence. Drop an idea and come back to it later without rebuilding the entire context. It adapts to the shape of the conversation instead of forcing me into rigid inputs.
That matters more than I expected. I’m not a designer drawing pixels or an engineer writing code. I’m a PM thinking through user movement, implementation friction, edge cases, and whether something that sounds good today will survive contact with reality six weeks from now.
That kind of reasoning is messy.
Claude Adaptive handles messy thinking well.
Different systems
ChatGPT and Claude work differently, but together
I see a lot of “Claude vs. ChatGPT” content out there, but that framing has never fully matched how I actually use them. The longer I worked with both, the more I realized they create different kinds of thinking pressure for me.
That difference became useful.
When Claude gets locked into a particular line of reasoning, and sometimes it does, I’ll take the same problem over to ChatGPT. Sometimes ChatGPT confirms what Claude already surfaced and I move forward with more confidence. Other times it exposes a blind spot I hadn’t noticed.
Either way, the second perspective usually saves me from endlessly reworking the same conversation.
ChatGPT is also where I tend to go for smaller visual tasks. Icons. Quick image concepts. Lightweight assets. But the moment the work becomes interconnected, flows, systems, navigation, I usually move back into Claude.
Not because one is smarter.
Because they create different kinds of thinking pressure.
Different modes
Claude Design isn’t Claude Adaptive
This one took me a while to understand.
Claude Design and Claude Adaptive may come from the same ecosystem, but they behave very differently. Claude Adaptive is exploratory. Conversational. Good at helping me think through ambiguity.
Claude Design is more structured. More execution-oriented. It responds better when the direction is already mostly formed.
At first, I kept expecting the same experience from both tools, and that mismatch created friction. Once I stopped treating them like interchangeable systems, the workflow made more sense.
I do most of my exploratory thinking in Claude Adaptive. Once the idea is mostly stable, I move into Claude Design for refinement and structure.
That shift changed the experience entirely.
The one I'm still unsure about
Figma AI and me
I was excited when Figma AI launched. I’d already spent months prototyping conversationally with Claude, so the idea of AI built directly into a design tool sounded perfect for my workflow.
In practice, it didn’t click for me.
The reality is that AI-assisted design still assumes a fairly strong understanding of design systems and interaction patterns. I’d steer the tool one direction and it would drift another. I’d course-correct and we’d drift again. Instead of reducing friction, I felt like I was managing it.
Then came the moment I still remember clearly: hitting the token limit mid-iteration while deliverables were piling up behind me.
I went back to Claude that afternoon.
I haven’t seriously returned to Figma AI since.
That doesn’t make it a bad tool. It just means the interaction model doesn’t fit how I naturally work, at least not yet.
How I decide
What I ask before adding a tool to my workflow
AI is evolving too quickly for me to pretend I have a final system figured out. So when a new tool shows up, and lately that feels like every week, I usually run it through three questions.
One: does it help with the actual work sitting in front of me right now? Not the demo. Not the marketing. The real work.
Two: if it doesn’t help, can I let it go quickly? No attachment. No “maybe someday.” If a tool earns a permanent place in my workflow, it has to earn it in practice.
Three: am I being honest about why it does or doesn’t work for me? Some tools fail because they’re weak. Others fail because they were built for a different kind of user. I’m not a full-time designer or SWE. I work through conversation, systems thinking, and iteration. The tools that work best for me tend to support that style of reasoning instead of fighting it.
That’s basically the framework.
Three questions. Honest answers.
What stays open
This list will probably look different in six months
That used to bother me. Now I think that’s part of the point.
The goal isn’t to lock in the perfect AI stack and stop adapting. The goal is to stay honest about which tools genuinely improve the work, which ones create friction, and which ones still haven’t found their role yet.
For now, these are the systems I keep returning to.
And honestly, I’m still trying to understand something underneath all of this: why the same person, asking the same questions, can get such different reasoning patterns from different AI systems.
That’s probably a separate post.