I use AI for 90% of my code. Here's what it still can't fix.
I've been building solo for 2.5 years. In the last year, my stack basically became: describe -> AI generates -> I review -> ship. Cursor, GitHub Copilot, Gemini, a few others. I'm faster than I've ever been.
But I keep running into the same walls. And I think it's worth being honest about them, because most "AI changed everything" posts skip this part.
What AI genuinely changed for me:
Boilerplate is dead. Setting up auth, DB schemas, component skeletons - I don't think about this anymore.
I can work in unfamiliar territory without being blocked. Need a Cloud Run deployment config? Done in minutes.
Iteration speed on UI is absurd. I describe a layout, get 80% there, tweak the rest.
What it still can't fix:
-
Knowing what to build. AI is a brilliant executor. It cannot tell you if the thing you're building matters. That judgment gap is entirely yours.
-
Debugging novel errors. When something breaks in an unusual way - like a specific ADK + BigQuery interaction at deploy time - AI gives you confident answers that are wrong 3 times before it gets there. You still need to understand what's actually happening.
-
Product intuition. "Make this feel more trustworthy" or "this onboarding feels off" - AI will generate 5 versions. Knowing which one is right is a skill it doesn't have.
-
Momentum when you're stuck. When I'm genuinely blocked on a decision (pivot? kill it? keep going?), AI is useless. It'll validate whatever I lean toward. You still need outside opinions, real users, or just time.
-
Distribution. Zero. AI writes decent copy but it cannot build an audience, create relationships, or make people care. This is still 100% a human problem.
I'm currently building Built By - a platform where solo builders log daily work and AI verifies it with proof-of-work quizzes. The irony of using AI to fix a problem AI partially created (fake productivity, unverifiable output) isn't lost on me.
Curious what walls other AI-first builders keep hitting. What's the thing that still doesn't get easier?
AI removed a lot of execution pain, which makes strategy mistakes easier to scale.
The bottleneck now isn’t building fast, it’s choosing the right thing fast.
This is the part most people miss.
AI didn’t remove difficulty — it just shifted it from execution to judgment.
Which honestly feels harder, because there’s no clear feedback loop for “taste” or “does this feel right”.
Curious — have you felt this more during onboarding/first impression, where small decisions have outsized impact?
True that. Yes I have experienced and it was real learning and useful insights on how to think with AI not be dependent on it.
Exactly — that’s the shift.
AI removed execution friction, but amplified decision friction.
Which is why things like:
– what to build
– how it’s perceived
– whether it clicks instantly
matter way more now than before.
Speed isn’t the bottleneck anymore — judgment is.