A few days ago, I put a simple question out on Threads: “What’s the biggest hurdle you’re facing with AI right now? Can be anything from basic to advanced.”

I expected a few technical gripes or maybe some complaints about hands having six fingers. But the responses that rolled in were much more revealing. They highlighted the real, everyday problems that creative people are working around as they try to integrate these tools into their actual workflows.

It turns out, we’re all kind of struggling with the same things. I read through the thread and categorized the responses. Here is what is actually slowing us down right now.

1. The Compute and Token Grind

For tools that are supposed to make us limitless, we spend an awful lot of time worrying about limits.

(Quick sidebar: If you’re new to this, think of “tokens” as the currency or cognitive fuel that AI models run on. A token is roughly equal to a piece of a word. Every time you feed the AI a prompt or an image, you spend tokens. Every time the AI answers, it costs tokens. And every model has a limit to how many tokens it can hold in its “memory” at once before it cuts you off or forgets the beginning of your conversation.)

• Cost and Limits: @raytray4 simply put it as, "Compute cost." And @exodyne213 expanded on that: "Compute limits. I’m constantly shepherding my sessions, switching between models to maximize tokens, etc."
• The Mid-Process Cutoff: There is nothing worse than being deep in a creative flow and getting a timeout. As @adamisasadamdoes noted, it’s "Hard to really know how far and long tokens will last. Ending mid process and random reset times."

Instead of just creating, we’re being forced to act as resource managers, constantly doing math in the background to ensure we don't hit a wall before the idea is fully baked. This issue caused me to leave Warp.dev which at one time was my favorite tool for coding.

2. The Context and Control Gap

AI is incredibly smart, but it lacks the lived, messy human experience that actually drives our creative choices. Getting the machine to understand why we are making a choice, or getting it to allow us to make that choice at all, is a massive roadblock.

• Missing the Human Nuance: @rosspeili pointed out a fundamental flaw: "AI misses background context. Humans don't do choices only based on skills and knowledge... Cause what matters is irrelevant data points from your past experiences. Eg. Childhood memories, smell of a scent you like..." System prompts just can't capture that easily.
• The Guardrails: Then there is the issue of safety filters becoming creative roadblocks. @sforkofficial called out the heavy censorship, specifically in video models like Runway, Sora, and Veo: "No matter how many times I sink it, they just dredge it back up and put patches in its hull."
• The Glitches: And sometimes, the tools just fail us. As @prime_of_the_day put it: "cursor/claude randomly stops saving its edits 💀."

3. The Overwhelm and Choice Paralysis

This is the category that resonated with me the most. The sheer volume of new tools and updates is paralyzing.

• The Embarrassment of Riches: @henrihelvetica called it exactly that: "The sheer volume of models we know, the hords more we don't that people talk about."
• Finding the Starting Line: @buildcuriosity mentioned the struggle of "identifying where it could actually make a difference so that I can spend more time on the work I enjoy."
• Lack of Energy: @pixolomew summed it up perfectly: "Lack of time and energy - there's so much being developed that I find it difficult to try out everything I want to."

My Perspective: The Pace of Change is Exhausting

If I’m being honest, that last category is my biggest hurdle right now.

I’ll spend weeks dialing in a specific workflow. I’ll figure out exactly how to chain a couple of tools together, get my prompts right, and finally get the output consistent. It feels like a massive win. And then, a week later, an update drops. Suddenly, that complex workflow I just mastered is a native, one-click standard feature in a new model.

Which is great, right? Progress! But then the workflow expands. What's actually possible... changes! There is this constant pressure to keep up, to adapt to the new standard, to learn the new interface.

It makes me wonder: Should I just pick one workflow and stick with it longer? Maybe the real competitive advantage isn't jumping to the newest model every Tuesday. Maybe it’s finding a tool that works well enough for your specific voice, planting your flag there, and ignoring the noise until you absolutely have to move. The tools are going to keep changing, but our ability to orchestrate them shouldn't require us to completely rebuild our foundation every month.

We’re all just trying to figure out how to make this work without losing our minds (or our tokens) in the process.