_pdp_

20 ms·

421.
by _pdp_ 9mo ago
To be honest this project is solving the easy part, i.e. the compute where to run the agent. The part where the agent does something useful is like 99% of the value that unsolved for most organisations.
422.
by _pdp_ 9mo ago
It occurs to me that there is an extraordinary amount of BS coming from all the places these days and I wonder if this comes from people with actual real experience or just some hypothetical, high-level thinking game. I mean, we use coding
423.
by _pdp_ 9mo ago
We also use CLI tools for agents because it is obvious that they are good at it. Our approach is slightly more complex because we need to supported fully authenticated sessions for end users and for shared credentials. It works by creating
424.
by _pdp_ 9mo ago
We realised early in 2025 that SQL is just better for 99% of the data retrieval tasks. AS a result of that, we wrote dedicated SQL translation layers for common APIs. The agent writes SQL which gets converted to API calls. There are some ba
425.
by _pdp_ 9mo ago
Bogus in what way?
426.
by _pdp_ 9mo ago
If you are interested here is a list of actual agentic patterns - https://go.cbk.ai/patterns
427.
by _pdp_ 9mo ago
I fully agree with the comment but CLI tools don't do auth and session management in a consistent way. Besides, imagine if now every vendor out there need to distributed custom binaries for their services. Some of them do. Many don&#x2
428.
by _pdp_ 9mo ago
Because we are not hyping to lure investors to give us hundreds of millions dollars. We took the more honest route and work with actual customers. If we are to accept hundreds of millions at some point perhaps we are going to reach hundreds
429.
by _pdp_ 9mo ago
I was skeptical until 3-4 months ago, but my recent experience has been entirely different. For context: we're the creators of ChatBotKit and have been deploying AI agents since the early days (about 2 years ago). These days, there&#x2
430.
by _pdp_ 9mo ago
I mean this is what? feeding a prompt to claude. It could be any other file. llms.txt makes sense as a standard but this is unnecessary.
431.
by _pdp_ 9mo ago
Yah I wouldn't. In my opinion, these things are better run the cloud to ensure you have a properly sandboxed, recoverable environment. At this point, I am convinced that almost anyone heavily relaying on desktop chat application has fa
432.
by _pdp_ 9mo ago
We have also attempted to implement exactly this but it turned out to be really bad architecture. The file system as an abstraction is actually not that good at all beyond the basic use-cases. Imagine you need to find an email. If you grep
433.
by _pdp_ 9mo ago
The upside is huge. The only downside that I can see is that the tools become less hackable. You cannot simply patch ts or eslint via pnpm if you want to. Now you need to build your own version. So yah, no free lunch.
434.
by _pdp_ 9mo ago
MCP is also a pretty good way to circumvent normal API security and many companies bought into it - and all of that just to hop on the AI hype train. You can do your own research.
435.
by _pdp_ 10mo ago
AI agent development looks deceptively simple at first glance. The following guide contains a high-level overview of everything we’ve learned the hard way. If I had known what it truly takes to build an agentic AI platform, I would never ha
436.
by _pdp_ 10mo ago
YEP Things are changing. Now everyone can build bespoke apps. Are these apps pushing the limits of technology? No! But they work for the very narrow and specific domain they where designed. And yes they do not scale and have as much bugs as
437.
by _pdp_ 10mo ago
Yaml works really well with LLMs (not to generate but to consume). So yes, we use it all the time in our service.
438.
by _pdp_ 10mo ago
We have a lot of grunt work scheduled overnight like finding bugs, creating tests where we don’t have good coverage or where we can improve, integrations, documentation work, etc. Not everything gets accepted. There is a lot of work that is
439.
by _pdp_ 10mo ago
With everything that we have done so far (our company) I believe by end of 2026 our software will be self improving all the time. And no it is not AI slop and we don't vibe code. There are a lot of practical aspects of running software
440.
by _pdp_ 10mo ago
Over 20 years professional experience here. LLM tools feel great. A single person can now accomplish what used to require many teams.
441.
by _pdp_ 10mo ago
You are absolutely right.
442.
by _pdp_ 10mo ago
To be fair, AI cannot write the code! It can write some types of code. It is fascinating that it can bootstrap moderately complex projects form a single shot. It does a better job at writing unit test (not perfect) then the fellow human pro
443.
by _pdp_ 10mo ago
@Spotify what are my top songs I don’t have the ability to pull your personal top songs directly from Spotify because that requires accessing your authenticated listening data. You can view them in Spotify by going to “Your Library” → “Made
444.
by _pdp_ 10mo ago
ok... how do you measure the performance of a coding assistant? Counting the lines of code written, bugs closed, PRs reviewed, some fuzzy measurement of quality or something else?
445.
by _pdp_ 10mo ago
You can apply the same philosophy to employees and if you dare to do so you will quickly find out that it does not work. When a measure becomes a target, it ceases to be a good measure - Goodhart's law. I cannot see why AI agents shoul
446.
by _pdp_ 10mo ago
interesting but ... why not debug the actual code that is invoking the API.. like break point at the right place, edit state, step over, resume... it seems that the toolchain is a lot more mature and it will fit right into the specific prog
447.
by _pdp_ 10mo ago
I wanted to share an example architecture which shows how an MCP server can be composed of additional AI agents providing. This is a decent way to hide implementation complexities and to enable more advance use-cases.
448.
by _pdp_ 10mo ago
I am fan of using markdown to describe the UI. It is simple, effective and feels more native to me than some rigid data structure designed for very specific use-cases that may not fit well into your own problem. Honestly, we should think of
449.
by _pdp_ 10mo ago
I wonder if this can be used to contaminate OpenAI search indexes?
450.
by _pdp_ 10mo ago
Yes probably. Though you know how these AI systems are - minor adjustments lead to different outcomes.

More ›