Stephan Schlecht
sw7com.bsky.social
Stephan Schlecht
@sw7com.bsky.social
Software developer, mainly iOS and macOS.
Claude Sonnet 4.5 behaved identically to 4.0 on my initial interactions, giving same incorrect answers to tricky UIKit/SwiftUI hit test questions and following the same flawed test case behavior, though with a really nice generated UI as always. Too early for conclusions, though. Let's see.
#AI #dev
October 1, 2025 at 5:53 AM
The AI predicted a coffee price of $404/lb, gave incorrect API advice, and created faulty DSP code: Learn why AI confidently hallucinates. Find out how to check results, verify sources, and proceed step by step. Use AI wisely, but trust your tests more.

www.software7.com/blog/why-you...

#AI #DEV
Why Your AI Assistant Might Be Wrong: Keep Healthy Skepticism | Software 7
AI assistants can hallucinate; recognize errors, verify sources, use the right context/model, and consider a step‑by‑step approach.
www.software7.com
September 21, 2025 at 9:43 PM
At first, I suspected an error in my code, in the request-response process, or in the feedback from tool calls to Anthropic. I then added a log and now know why my process is not continuing. The reason for this is HTTP status code 529, `overloaded_error`. OK.
#AI #DEV
August 21, 2025 at 10:17 AM
Just blogged: How Moonshot AI Kimi K2 Performs in Real Coding Work: www.software7.com/blog/how-moo...
For someone like me, who is used to the massive dominance of the US in the field of computer science, this is quite interesting.
#AI #Dev
How Moonshot AI Kimi K2 Performs in Real Coding Work | Software 7
www.software7.com
July 20, 2025 at 7:02 PM
Seems like everyone's trying to get a chat with Kimi-K2 right now. 🤔 #AI #Kimi #Kimi-k2 #moonshotAI
July 13, 2025 at 5:03 PM
Speechless
July 12, 2025 at 10:21 AM
Reposted by Stephan Schlecht
We ran a randomized controlled trial to see how much AI coding tools speed up experienced open-source developers.

The results surprised us: Developers thought they were 20% faster with AI tools, but they were actually 19% slower when they had access to AI than when they didn't.
July 10, 2025 at 7:47 PM
I haven't deleted the internet, but Gemini-Pro has now informed me that it is self-destructing.
June 26, 2025 at 9:18 PM
Been looking for a coding model that allows local agentic flow for some time now. Gave Mistral Small 3.2 three tools: ‘list_files’, ‘read_file’ and ‘write_file’, and it works - see for yourself in a 3-minute screencast in my blog post here: www.software7.com/blog/mistral...
#AI #Dev #Mistral
Local Agentic Flow with Mistral Small 3.2 | Software 7
www.software7.com
June 24, 2025 at 9:56 PM
The new reality of data scraping. But I wonder: it costs real money on this scale. The data doesn't change every hour.
...btw, as of a few days ago (at least temporarily), browsing PRs/code of blender requires a login. Why? Because otherwise the servers get DDoS'ed by the AI training scraping bots; looks like some of them are on the loose. Again.
June 24, 2025 at 7:20 PM
It looks like Cursor has changed its Pro plan from 500 requests to “unlimited agent requests”. That's interesting. Why?
#AI #Cursor
June 17, 2025 at 8:53 PM
New reality in software development: productivity drops during rush hours. 😅
#AI #Dev
June 17, 2025 at 3:28 PM
SwiftUI App With AI from Scratch

- Human-like communication with the LLM via screenshots
- If the LLM loses faith, reassurance helps.
- AIs use tools cleverly, even AppleScript to control Finder

More details in my full blog post: www.software7.com/blog/buildin...

#AI #SwiftUI #Development
June 15, 2025 at 8:40 PM
Reposted by Stephan Schlecht
Ichi: On-device conversational AI using MLX Swift and MLX audio
github.com/rudrankriyam...
GitHub - rudrankriyam/Ichi: Experimenting with conversational AI in iOS, macOS and visionOS apps
Experimenting with conversational AI in iOS, macOS and visionOS apps - rudrankriyam/Ichi
github.com
June 1, 2025 at 11:35 PM
Simplifying (digital) life. I have reduced the number of messengers I have an account with. The first thing I did was delete my WhatsApp account. Bonus: I don't have to read through what exactly they want to use for AI training. Like I said, simplification 😀
May 25, 2025 at 2:24 PM
Just read that Murderbot is coming to Apple TV. So, so cool. The book series (The Muderbot Diaries) by Martha Wells is definitely absolutely brilliant. I don't have time just right now, otherwise I would watch this IMMEDIATELY.
May 16, 2025 at 3:48 PM
Why I often don't polish and finish many projects and why that's perfectly fine and wasn't the goal at all.

I've learned that finishing the last 20% is often not worth the 80% extra effort.

www.software7.com/blog/struggl...
The Struggle of 80/20 in Software: Why Finishing Is Not Always the Goal | Software 7
www.software7.com
May 10, 2025 at 10:40 PM
Just published: Balancing Agentic AI with Traditional Engineering

www.software7.com/blog/balanci...

Includes a screencast with JetBrains Junie

#AgenticAI #Dev #JetBrains
Balancing Agentic AI with Traditional Engineering | Software 7
www.software7.com
May 4, 2025 at 8:50 PM
When relaunching my blog, I deleted all the old posts, but server logs showed that two particular posts were still being accessed. So I decided to republish them. Actually programmatically querying specific DNS servers on iOS (www.software7.com/blog/program...) seemed very niche, so it surprised me.
Programmatically Query Specific DNS Servers on iOS | Software 7
www.software7.com
April 24, 2025 at 11:35 AM
Git will be 20 years old on April 7 🎈. And I was able to publish my macOS Git stats tool in time for the anniversary. Yeah. www.gitchronicles.com
GitChronicles - Git Repository Analytics & Statistics Tool | Software 7 GmbHDownload_on_the_Mac_App_Store_Badge_US-UK_RGB_blk_092917
GitChronicles provides comprehensive Git repository statistics and analytics that reveal development patterns, contributor impact, and codebase evolution. 100% local processing with no cloud dependenc...
www.gitchronicles.com
April 4, 2025 at 6:53 PM