I achieved 6x productivity boost by leveraging a cutting-edge AI Stack (mileage may vary)
Over Christmas, I decided to run a 3-day personal hackathon. The goal? Build a native iPhone app using a cutting-edge AI stack (Cursor AI + ChatGPT o1). Here’s what I learned.
1. Inspiration
The idea sparked during a call with IBM 's Mohamad Ali about digital workers and AI assistants, combined with discussions with IBM’s Matias Karvinen , Finnish MP Jarno Limnell , academic experts Pekka Toivanen & Virpi Hotti ( University of Eastern Finland ), and 微软 Finland’s MD Mervi Airaksinen & COO Teemu Vidgren .
Us Finns ???? kept asking:
“How do we boost productivity in a country (Finland) that’s experienced near-zero growth for 17 years?”
I wanted to see what could be done in the IT space, especially around software development. Could I accomplish in three days with AI what might otherwise take me a month by using traditional development tools? Could I achieve dramatic improvements in my own workflow? The Christmas ?? break provided the perfect opportunity to conduct this experiment.
2. Software Stack & AI Tools
I chose 苹果 's Xcode (v16.2) and Swift—tech I only knew the basics of—to avoid my usual Python, PHP, or Java. Then, I used Cursor AI (v0.44.9) and ChatGPT o1 to go from concept to App Store for a “medium complexity” app. The app—Road Radar (Finland)—leverages the Fintraffic Digitraffic APIs (licensed under CC 4.0) to show real-time weather and traffic camera data across Finnish roads. For visuals, I relied on DALL-E and Midjourney .
Why is This Stack Cutting-Edge ??
? Cursor AI (Claude 3.5 Sonnet by Anthropic ) & ChatGPT o1 both demonstrate top-tier reasoning metrics (e.g., GPQA ~59.4%, MMLU ~88.7%, Humaneval ~92.0%, MGSM ~91.6%). These place them among the most advanced large language models (LLMs) available.
? By combining Cursor AI for rapid code generation and ChatGPT o1 for complex troubleshooting, you tap into extremely high-level problem-solving.
? I intentionally didn’t use AI “agents,” focusing on direct AI-driven development for speed and control.
3. Observations ??
Cursor felt like an “overcaffeinated lumberjack”
Cursor AI running Claude 3.5 Sonnet by Anthropic proved extremely productive—much faster and more stable than my previous experimentations with ChatGPT-4o + Canvas. Once I grounded Cursor AI with Xcode docs, Apple Swift basics, and the relevant Digitraffic endpoints, I could assemble the app’s core in under two days.
However, Cursor also felt like an “overcaffeinated lumberjack”: it could rapidly cut down a forest but might occasionally saw off crucial parts if not carefully supervised.
Additionally, certain Xcode/Swift “known issues” (like avoiding NavigationLink inside an Annotation) weren’t in official docs—this is where ChatGPT o1 really shined. Meanwhile, DALL-E and Midjourney handled all graphical assets.
4. Results
I observed a 6x productivity boost in my workflow
Blending Cursor AI + Claude 3.5 Sonnet for the main structure, ChatGPT o1 for tough debugging, and DALL-E/Midjourney for graphics yielded remarkable efficiency—provided I accepted that these digital workers can’t do everything. About half the time, I was fixing bugs and polishing the final product. Even so, in just over three days, I went from idea to App Store submission with a fully functional iOS app. Normally, that might have taken me around 20 working days—so I observed about a 6× productivity boost.
5. The 500%+ Productivity Stack (aka Future Workflow)
My whole experiment got me thinking about what a future software development stack would look like that would take full benefit of the latest AI tools. Here’s my recommendation for the most advanced AI dev setup for early 2025.
Will it provide you with 5x productivity? I’d like to emphasize that results vary with each project’s team and scope, but it should yield significant gains if you aren’t actively using AI tools in your stack yet:
领英推荐
1. Copilot for Microsoft 365 : Transcribe and summarize meeting notes about the app’s requirements.
2. Copilot for Microsoft 365: Generate Markdown (MD) docs primarily for development (plus any stakeholder or presentation needs) based on transcriptions.
3. ChatGPT o1: Feed MD docs into ChatGPT to decide on the best technical architecture with your system architects.
4. Cursor AI (running Claude 3.5 Sonnet): Build the app, refine visuals with DALL-E + Midjourney + Vercel 's v0. As Github Copilot just started supporting Claude 3.5 Sonnet , it is a very solid choice as well.
5. ChatGPT o1: Solve tricky technical issues and finalize the code manually.
6. ChatGPT o1: Export everything from Cursor AI to ChatGPT for marketing copy and internal documentation.
7. Microsoft 365 Copilot: Convert that marketing copy or internal docs into PPT slides and other formats.
Side note: If you have deeper pockets, you could replace Cursor AI ($20–$40/mo) with Cognition 's DevinAI (~$500/mo), depending on your budget.
6. Road Radar (Finland)
Road Radar—made mostly by AI ??, tweaked by a human ??????
Curious to see what can be built in around three days with AI? Check out Road Radar (Finland) on the Apple App Store—made mostly by AI ??, tweaked by a human ??????. It features:
? Over 600 cameras and 400 weather stations
? Multilingual support (English, Finnish, and Swedish)
? Works on both iPhone & iPad
? Dark Mode and a clean, intuitive design
? Completely free (free to download, ad-free, no data collection)
Link to the app: https://apps.apple.com/us/app/road-radar-finland/id6740081315
Enjoy—and let me know what you think!
#AI #SoftwareDevelopment #Productivity #Innovation #CursorAI #Anthropic #Claude #ChatGPT #Microsoft #Copilot
Turning tech insights into engaging content that drives engagement
1 个月It’s amazing to see how AI copilots like ChatGPT and tools like Cursor AI can unlock such levels of productivity. Your hackathon story is a perfect example of what’s possible when AI-driven development meets creativity. Speaking of AI copilots, we’re hosting a webinar on leveraging copilots and AI agents. Curious to see what’s possible? Don’t miss out—reserve your spot here:?https://www.dhirubhai.net/events/copilotandaiagentsforsales-serv7282757944104890368/
Managing Director @ mExcelle | Corporate Strategy, Digital Transformation CIO, AI Scribbler, B2C Product Management "The mind is the ultimate architect; master it, and you command the backbone of innovation."
1 个月I add a simple line to my prompts. “Avoid overengineering and removing working code unnecessarily. Take your time, think a lot and fix the errors.” It works most of the time without deleting code.
Digital Transformation, Operational Excellence, Fundamental believer in AIA (AI and Automation)
1 个月A fascinating forecast of the future of work. Thank you for sharing!
Great work! I agree with the 3–6x boost in development, especially in unfamiliar areas. These AI tools foster a more T-shaped skill set (though design and creativity still need a human touch). They’re basically a “context-aware StackOverflow,” taking away tedious tasks and replacing basic Googling. I use them heavily, but they’re not great at complex software design or truly creative problem solving—humans still orchestrate the bigger picture. PS. The above message is AI-generated, I just gave the key points and "high level design" :)
|Technology and AI News | Contributor | Blog | Connect and Chat Tech
1 个月ChatGPT integration with MacOS and apps is inspiring, and helps with some of this. I hope app integration for desktop app for windows comes soon too, I guess just too many variables at the moment. Impressive results and thanks for the step by step guide.