- The Atomic Builder
- Posts
- The Hard Truth: AI Building is Still More Art Than Science
The Hard Truth: AI Building is Still More Art Than Science
An Honest Guide: What’s Improving, What’s Not, What You Can Do About It
Faisal Shariff
March 14, 2025
Hi, and welcome to The Atomic Builder!
This is where product managers, founders, and non-tech creators learn to master AI-powered software building - you could be lost in a Discord server debating AGI timelines, but instead, you chose AI, product strategy, and making things happen. Thank you for being here!
Today’s issue aims to cut through the influencer noise that saturates social media and get down to facts. Yes, I’m optimistic about AI powered software creation, but let’s get serious, where are the struggles today when building with these tools?
Got this from a friend? Join other product managers, founders, and creators staying ahead of AI-powered product building. Subscribe to get The Atomic Builder every week.
AI-Powered Product Creation: A Silver Bullet?
AI is everywhere. Founders pitch "AI-powered" everything, investors chase it, and creators (like you and I) scramble to integrate it. But if you're actually building with AI tools (Replit, Cursor, Bolt et al), you quickly realize, it’s still quite early and it's not always the magic bullet it’s cracked up to be…yet. |  Two silver bullets… |
Yes, AI speeds things up, sometimes dramatically. There are numerous claims that you can build an app while brewing your coffee, what are the struggles though?
Hard Truths from Real Experiments
TL;DR:
AI-generated apps are fantastic for prototypes but often leave manual cleanup.
Costs can quickly spiral as usage scales.
Integrations, while improving, still often require debugging and manual effort.
UX matters more than ever—"powered by AI" isn't a user benefit unless it's seamless.
So, what does this all mean in practice? Enter the Atomic AI Scorecard - tracking the real progress of AI-powered tools.
🚀 Atomic AI Quarterly Scorecard (Q1 2025)
Each quarter, I’ll track what's improving, what's still challenging, and how these areas have shifted from previous quarters.
You can also access this resource from The Atomic Builder’s product hub. Whether you’re a seasoned product manager or just starting to explore AI-driven tools, this site is packed with insights and hands-on resources to help you build smarter, faster, and with less code.
This scorecard is just the beginning. Over time, I want to expand it beyond AI-powered software building to include other game-changing AI tools that help with the software/build process. Future editions may explore areas like AI-powered video creation for product managers—helping you create product demos, onboarding materials, and explainers without big budgets.
AI isn’t just reshaping how we build software, it’s changing how we communicate about our products, too.
Before we dive into the scorecard, here’s the scale we’ll use.
Rating | Definition |
|---|---|
🟢 4-5 (Strong Performance) | Effective for most non-technical users, requires minimal intervention. |
🟡 3 (Moderate Performance) | Works but has limitations—requires adjustments, extra learning, or manual fixes. |
🟠 2 (Limited Use) | Has potential, but gaps in functionality make it unreliable for real-world application. |
🔴 1 (Needs Major Improvement) | Struggles to deliver on AI-powered claims; requires too much manual work to be useful. |
Now, let’s see how the space in general is actually performing. This scorecard breaks down strengths and weaknesses across key areas - where AI-powered building is thriving, where it needs work, and where it’s still more hype than reality.
This scorecard isn’t just about rating AI tools—it’s about understanding how they will improve your experience building products, over time. Let’s break it down.
🛠 Core AI Capabilities: How Well Are They Evolving?
AI tools have come a long way, but are they actually making it easier to build products?
Category | Q1 25 Status | Comments |
|---|---|---|
API Integration (Ease of Connecting Services) | 🟢 4/5 (Strong Performance) | AI tools now simplify integrating API calls. |
Natural Language Building (Text-to-Build Platforms) | 🟡 3/5 (Moderate Performance) | Great for generating basic structures, can break with complex logic. |
Debugging & Fixes | 🟡 3/5 (Moderate Performance) | Debugging is improving, but AI still struggles with advanced issues and can lack ‘common sense.’ |
🎯 Ease of Use & Learning Curve
How Quickly Can You Get Up and Running?
Category | Q1 25 Status | Comments |
|---|---|---|
Ease of Use & Learning Curve | 🟡 3/5 (Moderate Performance) | Lovable and Bolt offer simple onboarding for beginners. Replit offers customization but requires more familiarity. Cursor and Windsurf give maximum control but have steep learning curves. |
🎨 AI in UI/UX: Can It Design Well?
AI-assisted design tools promise fast UI/UX creation, but how well do they actually work?
Category | Q1 25 Status | Comments |
|---|---|---|
AI-Assisted UI/UX Generation | 🟡 3/5 (Moderate Performance) | Tools like Bolt / Uizard are leading design, but design intuition is still hit-or-miss unless you provide a reference. |
🤝 Knowledge & Collaboration: Can AI Tools Work for Teams?
Collaboration and learning remain AI's weak spots—where do we stand?
Category | Q1 25 Status | Comments |
|---|---|---|
Community Knowledge & Resource Sharing | 🟠 2/5 (Limited Use) | Transparency is growing, but AI tool marketing is full of hype (watch out for those influencers!). |
Collaboration & Multi-User AI Projects | 🟠 2/5 (Limited Use) | Emerging improvements (Tempo, Replit), but still limited team-management capabilities. |
🤖 AI's Logic & Reliability: Where Does It Still Struggle?
These are the areas where AI still falls short and demands the most improvement.
Category | Q1 25 Status | Comments |
|---|---|---|
AI-Generated Logic (Beyond UI Generation) | 🟠 2/5 (Limited Use) | Struggles to maintain logical flows and decision-making. Prompt crafting is critical to guide what you want. |
Reliability & Consistency of AI Outputs | 🔴 1/5 (Needs Major Improvement) | AI tools produce unpredictable results—identical prompts yield different outputs. The nature of probabilistic systems. |
⚠️ Common AI Frustrations & How to Overcome Them
Even with improvements, AI tools still introduce friction. Let’s break down the most common struggles—and how to work around them.
Frustration | Q1 25 Status | Recommended Strategy |
|---|---|---|
Unexpected Costs | 🟡 3 (Moderate Performance) | Monitor token usage carefully; AI usage costs can spiral quickly. |
UX Clunkiness in AI Features | 🟡 3 (Moderate Performance) | If possible, use reference visuals to guide design. |
“Last 10%” Debugging | 🟠 2/5 (Limited Use) | AI helps with scaffolding, but final code needs human cleanup. Production-scale work should involve IT. |
Opaque AI-Generated Code | 🟠 2/5 (Limited Use) | Prompt AI for step-by-step explanations; request structured, readable outputs. |
AI Hallucinations & Overconfidence | 🟠 2/5 (Limited Use) | Validate outputs thoroughly; request structured explanations and build in error-checking safeguards. |
📈 Strategy Spotlight: Why Measuring Matters. Don’t Just Build, Track Impact
AI lets you build faster, but are you building smarter? Too often, we (may…) launch features, workflows, or content without knowing if they make a difference. Great creators and teams track impact, not just completion.
As you create things, ask yourself:
|  Measuring Matters |
With that in mind, we’ll continue to track the above over the course of this year to see how things improve.
If what you build isn’t improving at least one of these, should it even exist?
Final Thoughts
AI-powered building is exciting—but it’s not a silver bullet yet. It’s still unpredictable, requires trial and error, and forces us to balance efficiency with control. The best builders don’t just experiment; they track, adapt, and refine their approach.
Your turn: What’s been your biggest surprise—good or bad—building with AI tools lately? Hit reply and let me know—I’ll feature the most interesting insights.
With this issue wrapped up, I’m off to grab a half-skim, double-shot, extra-foam, vanilla oat, matcha latte. Or maybe I’ll just grab a flat white...
Until next time, keep experimenting, keep building, and as always - stay atomic. 👊
Faisal
This Week’s Build Beats 🎵
Each issue, we pair the newsletter with a track to keep you inspired while you build.
This week, because AI building is still more art than science—what better choice than:
🎧 “The Scientist” – Coldplay
Grab the playlist on Spotify - I add to it each week!
Enter at your own risk
 | Thanks for Joining! I’m excited to help usher in this new wave of AI-empowered product builders. If you have any questions or want to share your own AI-building experiences (the successes and the failures), feel free to reply to this email or connect with me on socials. Until next time… Faisal |
P.S. Know someone who could benefit from AI-powered product building? Forward them this newsletter!