Not for this project specifically, but I realize that I've seen a lot of AI agents, but I've never seen something interesting build with them. Some simple website, maybe even some very simple old games like snake or pong, but nothing better. Do I miss something ?
Show HN: Plandex – an AI coding engine for complex tasks
81–90 of 112 posts
Re: Show HN: Plandex – an AI coding engine for complex tasks
#82Not for this project specifically, but I realize that I've seen a lot of AI agents, but I've never seen something interesting build with them. Some simple website, maybe even some very simple old games like snake or pong, but nothing better. Do I miss something ?
I'd say LLM agent, or multi-agents are still in the very early research/prototype stage.
You can tell because there are papers from Microsoft on this but no product: https://www.microsoft.com/en-us/research/project/autogen/
I also wrote about the L1 to L5 of AI coding here: https://prompt.16x.engineer/blog/ai-coding-l1-l5
Re: Show HN: Plandex – an AI coding engine for complex tasks
#83Congrats on the launch. Can you please compare and contrast Plandex features with another similar solution like aider[1] which also helps solve similar problem. [1] https://github.com/paul-gauthier/aider
Thanks for mentioning aider! I haven't had a chance to look closely at plandex, but have read the author's description of differences wrt aider. I'd add a few comments: I think the plandex UX is novel and interesting. The idea of a git-like CLI with various stateful commands is a new one in this space of ai coding tools. In contrast, aider uses a chat based "pair programming" UX, where you collaborate with the AI and…
> These capabilities are all available "for free" in aider, because it is tightly integrated with git.
Sounds like the right approach to me. Some quick questions:
1. Is it easy to customize the system prompt with aider?
2. Does aider save a record of all OpenAI API calls? I’m thinking I may e.g. want to experiment with fine tuning an open source model using these one day.
3. What would you say are aider’s closest “competitors”?
Re: Show HN: Plandex – an AI coding engine for complex tasks
#84As someone who is trying to build a bootstrapped startup in spare time (read: coding while tired), this is amazing. Thank you so much for creating it.
Re: Show HN: Plandex – an AI coding engine for complex tasks
#85To support many other models you should look at ollama - it provides a REST API on your machine for local inference that works just like OpenAI
Re: Show HN: Plandex – an AI coding engine for complex tasks
#86Not for this project specifically, but I realize that I've seen a lot of AI agents, but I've never seen something interesting build with them. Some simple website, maybe even some very simple old games like snake or pong, but nothing better. Do I miss something ?
I brainstormed a text game-engine powered by an llm, but relying on a non-local llm was offputting. Local llms are getting more and more viable though. A general problem I was running into was that thinking in terms of LLM queries is a very new way of computation design and adapting takes a lot of effort. Then again, my central idea was a bit ambitious too: every game character would have a unique interpretation on what was happening.
Re: Show HN: Plandex – an AI coding engine for complex tasks
#87If this thing really worked, why wouldn't you just point it at AWS documentation and ask it to implement the exact same APIs and come up with designs for the datacenters in extreme detail? Implementing APIs is completely legal.
Re: Show HN: Plandex – an AI coding engine for complex tasks
#88I wanted to get a better idea of how it worked, so I asked Claude to write up an overview. https://gist.github.com/CGamesPlay/8c2a2882c441821e76bbe9680...
This is really cool! And quite accurate.
Re: Show HN: Plandex – an AI coding engine for complex tasks
#89Earlier quoted context omitted.
Thanks for mentioning aider! I haven't had a chance to look closely at plandex, but have read the author's description of differences wrt aider. I'd add a few comments: I think the plandex UX is novel and interesting. The idea of a git-like CLI with various stateful commands is a new one in this space of ai coding tools. In contrast, aider uses a chat based "pair programming" UX, where you collaborate with the AI and…
> These capabilities are all available "for free" in aider, because it is tightly integrated with git. Sounds like the right approach to me. Some quick questions: 1. Is it easy to customize the system prompt with aider? 2. Does aider save a record of all OpenAI API calls? I’m thinking I may e.g. want to experiment with fine tuning an open source model using these one day. 3. What would you say are aider’s closest “co…
Just to note, Plandex also has integration with git on the client-side and can automatically commit its changes (or not--you can decide when applying changes).
One of the reasons I think it's good to have the plan version-controlled separately from the repo is it avoids intermingling your changes and the model's changes in a way that's difficult to disentangle. It's also resilient to a "dirty" git state where you have a mix of staged, unstaged, and untracked changes.
One more benefit is that Plandex can be used in directories that aren't git repos, while still retaining version control for the plan itself. This can be useful for more one-off tasks where you're not working in an established project.
Re: Show HN: Plandex – an AI coding engine for complex tasks
#90Not for this project specifically, but I realize that I've seen a lot of AI agents, but I've never seen something interesting build with them. Some simple website, maybe even some very simple old games like snake or pong, but nothing better. Do I miss something ?
Try to build something interesting with Plandex! Perhaps you will be pleasantly surprised. Either way, please let me know how it goes.