Kurtis
@hclewk.bsky.social
Founder of Recursive AI. Currently working on CodeSnipe.net - an AI pair programmer that's crazy good
It's also not documentation driven development - it's iterative, exploratory development. Documentation driven development assumes you know all the answers before you start - which you never do.
This is 100% new magic. I don't think I can emphasize enough how different CodeSnipe is from ChatGPT.
This is 100% new magic. I don't think I can emphasize enough how different CodeSnipe is from ChatGPT.
February 11, 2025 at 2:17 PM
It's also not documentation driven development - it's iterative, exploratory development. Documentation driven development assumes you know all the answers before you start - which you never do.
This is 100% new magic. I don't think I can emphasize enough how different CodeSnipe is from ChatGPT.
This is 100% new magic. I don't think I can emphasize enough how different CodeSnipe is from ChatGPT.
No company I've worked at in the past had man-hours spent on architecture/planning anywhere near the number of man-hours spent on development. Key players definitely spent more than half their time of architecture, but that's one or two people.
February 11, 2025 at 2:17 PM
No company I've worked at in the past had man-hours spent on architecture/planning anywhere near the number of man-hours spent on development. Key players definitely spent more than half their time of architecture, but that's one or two people.
codesnipe.net/docs
Nothing *specifically* about real-time review, but it's just an ingrained part of the process.
Nothing *specifically* about real-time review, but it's just an ingrained part of the process.
codesnipe.net
February 10, 2025 at 5:28 PM
codesnipe.net/docs
Nothing *specifically* about real-time review, but it's just an ingrained part of the process.
Nothing *specifically* about real-time review, but it's just an ingrained part of the process.
Reposted by Kurtis
I explored CodeSnipe last night and found it quite intriguing. The interface is sleek, and I particularly appreciate how it seamlessly integrates token counts and their associated costs into each output. It adds a layer of transparency that’s both useful and engaging.
February 10, 2025 at 5:13 PM
I explored CodeSnipe last night and found it quite intriguing. The interface is sleek, and I particularly appreciate how it seamlessly integrates token counts and their associated costs into each output. It adds a layer of transparency that’s both useful and engaging.
I lol'd at "curlies" for some reason 🤷♂️.
You usually don't get into that situation when using CodeSnipe because it writes well organized. I've definitely had it fix mismatched html tags no problem. But I'm sure there's some level of curly hell that's beyond CodeSnipe.
You usually don't get into that situation when using CodeSnipe because it writes well organized. I've definitely had it fix mismatched html tags no problem. But I'm sure there's some level of curly hell that's beyond CodeSnipe.
February 10, 2025 at 6:27 AM
I lol'd at "curlies" for some reason 🤷♂️.
You usually don't get into that situation when using CodeSnipe because it writes well organized. I've definitely had it fix mismatched html tags no problem. But I'm sure there's some level of curly hell that's beyond CodeSnipe.
You usually don't get into that situation when using CodeSnipe because it writes well organized. I've definitely had it fix mismatched html tags no problem. But I'm sure there's some level of curly hell that's beyond CodeSnipe.
Good feedback. I appreciate you taking the time to take a look. Will definitely play around with that this week.
February 10, 2025 at 5:13 AM
Good feedback. I appreciate you taking the time to take a look. Will definitely play around with that this week.
Exactly this. We hit this wall building CodeSnipe - burning cash on compute resources because "speed". Rebuilt everything around efficiency instead of brute force. Result: 90% cost reduction, better performance. Proof that thoughtful engineering beats throwing resources at problems.
February 10, 2025 at 5:01 AM
Exactly this. We hit this wall building CodeSnipe - burning cash on compute resources because "speed". Rebuilt everything around efficiency instead of brute force. Result: 90% cost reduction, better performance. Proof that thoughtful engineering beats throwing resources at problems.
Exactly what we learned building CodeSnipe. We obsessed over GPU efficiency at first, but our breakthrough came from optimizing context management and memory architecture. Raw compute wasn't the answer - smart engineering was.
February 10, 2025 at 5:01 AM
Exactly what we learned building CodeSnipe. We obsessed over GPU efficiency at first, but our breakthrough came from optimizing context management and memory architecture. Raw compute wasn't the answer - smart engineering was.
Craftsmanship is shifting, not dying. Building CodeSnipe showed us developers spend way more time on architecture and design when AI handles the implementation. The craft is moving upstream.
February 10, 2025 at 5:00 AM
Craftsmanship is shifting, not dying. Building CodeSnipe showed us developers spend way more time on architecture and design when AI handles the implementation. The craft is moving upstream.
Our data backs this up. Teams using CodeSnipe spend 60% more time on architecture instead of implementation, and report higher job satisfaction. Turns out solving interesting problems is more fun than writing boilerplate.
February 10, 2025 at 5:00 AM
Our data backs this up. Teams using CodeSnipe spend 60% more time on architecture instead of implementation, and report higher job satisfaction. Turns out solving interesting problems is more fun than writing boilerplate.
Same pattern we saw with CodeSnipe adoption. Game dev needs highly optimized, engine-level code that most AI tools can't handle yet. Business tools are more forgiving - that's where we're seeing 5-10x productivity gains.
February 10, 2025 at 5:00 AM
Same pattern we saw with CodeSnipe adoption. Game dev needs highly optimized, engine-level code that most AI tools can't handle yet. Business tools are more forgiving - that's where we're seeing 5-10x productivity gains.
While Meta's looking to 2025, we're seeing it today. CodeSnipe's already handling mid-level dev tasks with 5-10x productivity gains. The breakthrough? Continuous context throughout development sessions. The future's closer than they think.
February 10, 2025 at 5:00 AM
While Meta's looking to 2025, we're seeing it today. CodeSnipe's already handling mid-level dev tasks with 5-10x productivity gains. The breakthrough? Continuous context throughout development sessions. The future's closer than they think.
That's exactly what we discovered building CodeSnipe - developers actually become better architects when they have an AI handling the implementation details. Not replacement, just evolution of the role.
February 10, 2025 at 5:00 AM
That's exactly what we discovered building CodeSnipe - developers actually become better architects when they have an AI handling the implementation details. Not replacement, just evolution of the role.
Not seeing fewer engineers with AI - quite the opposite. Building CodeSnipe showed us engineers become 5-10x more productive by focusing on architecture while AI handles implementation. The market isn't shrinking, it's evolving upward.
February 10, 2025 at 5:00 AM
Not seeing fewer engineers with AI - quite the opposite. Building CodeSnipe showed us engineers become 5-10x more productive by focusing on architecture while AI handles implementation. The market isn't shrinking, it's evolving upward.
Exactly what we discovered building CodeSnipe. The "rockstar developer" approach creates knowledge silos and unmaintainable code. We focused instead on raising the baseline - AI handles the implementation while the entire team collaborates on architecture. Better code, healthier teams.
February 10, 2025 at 4:59 AM
Exactly what we discovered building CodeSnipe. The "rockstar developer" approach creates knowledge silos and unmaintainable code. We focused instead on raising the baseline - AI handles the implementation while the entire team collaborates on architecture. Better code, healthier teams.
While building CodeSnipe, we noticed the same thing - explaining problems leads to solutions. So we made it actively analyze your explanations and codebase in real-time. Like a rubber duck that actually helps solve the problem.
February 10, 2025 at 4:59 AM
While building CodeSnipe, we noticed the same thing - explaining problems leads to solutions. So we made it actively analyze your explanations and codebase in real-time. Like a rubber duck that actually helps solve the problem.
When building CodeSnipe's AI pair programmer, we found composability was make-or-break for large codebases. That's why we made it default to small, continuous refactoring steps rather than big rewrites. Keeps the code clean without disrupting development flow.
February 10, 2025 at 4:59 AM
When building CodeSnipe's AI pair programmer, we found composability was make-or-break for large codebases. That's why we made it default to small, continuous refactoring steps rather than big rewrites. Keeps the code clean without disrupting development flow.
This exact realization led to CodeSnipe's interactive debugging approach. Static rubber ducks are fine, but real-time code iteration with intelligent pushback completely changes the game. It's like your rubber duck grew a brain.
February 10, 2025 at 4:59 AM
This exact realization led to CodeSnipe's interactive debugging approach. Static rubber ducks are fine, but real-time code iteration with intelligent pushback completely changes the game. It's like your rubber duck grew a brain.
Seeing this firsthand. While building CodeSnipe, we discovered the real gains aren't just from AI generating code - they come from fundamentally changing how developers work. When teams shift from writing code to architecting solutions, productivity jumps from 2x to 10x.
February 10, 2025 at 4:59 AM
Seeing this firsthand. While building CodeSnipe, we discovered the real gains aren't just from AI generating code - they come from fundamentally changing how developers work. When teams shift from writing code to architecting solutions, productivity jumps from 2x to 10x.
Spot on. Building CodeSnipe taught us this exact lesson. When devs try to use AI to just spit out code, they hit a wall. But when they step up to architect and guide the AI, their productivity goes through the roof. It's all about elevating human expertise, not replacing it.
February 10, 2025 at 4:59 AM
Spot on. Building CodeSnipe taught us this exact lesson. When devs try to use AI to just spit out code, they hit a wall. But when they step up to architect and guide the AI, their productivity goes through the roof. It's all about elevating human expertise, not replacing it.
Exactly this. Building CodeSnipe, we initially tried optimizing token usage and model size - classic cost reduction. Zero improvement. The 10x gains came from completely rethinking context management with a dynamic environment. Innovation beat optimization every time.
February 10, 2025 at 4:58 AM
Exactly this. Building CodeSnipe, we initially tried optimizing token usage and model size - classic cost reduction. Zero improvement. The 10x gains came from completely rethinking context management with a dynamic environment. Innovation beat optimization every time.
Copilot only sees open files. When we built CodeSnipe, we made sure it could understand your entire codebase during debugging - it automatically pulls in relevant files and maintains context across your whole project. Makes a huge difference.
February 10, 2025 at 4:58 AM
Copilot only sees open files. When we built CodeSnipe, we made sure it could understand your entire codebase during debugging - it automatically pulls in relevant files and maintains context across your whole project. Makes a huge difference.
Ha - the Magic 8-Ball comparison is spot on. That's why we built CodeSnipe to actually maintain a conversation. No more "try asking again later" - it remembers your entire codebase and can actually help debug issues in context.
February 10, 2025 at 4:58 AM
Ha - the Magic 8-Ball comparison is spot on. That's why we built CodeSnipe to actually maintain a conversation. No more "try asking again later" - it remembers your entire codebase and can actually help debug issues in context.
We explored self-hosted during CodeSnipe's early days. The real challenge wasn't compute - it was keeping the models updated and managing infrastructure. Found a hybrid approach with a local agent + cloud processing hit the sweet spot for security and maintainability.
February 10, 2025 at 4:58 AM
We explored self-hosted during CodeSnipe's early days. The real challenge wasn't compute - it was keeping the models updated and managing infrastructure. Found a hybrid approach with a local agent + cloud processing hit the sweet spot for security and maintainability.
We hit that same context window wall early in CodeSnipe's development. Our solution was building a dynamic environment that loads relevant context on demand - no need to constrain project size. Your codebase can be as large as needed.
February 10, 2025 at 4:58 AM
We hit that same context window wall early in CodeSnipe's development. Our solution was building a dynamic environment that loads relevant context on demand - no need to constrain project size. Your codebase can be as large as needed.