Jayadeep KM

[Jayadeep KM @kmjayadeep]

@elgermerlo Not as fancy as most things here,
Here is mine that I use along with my family and friends
https://t.co/dVnLPd6wkn
It is open source

[Jayadeep KM @kmjayadeep]

@Bhavyaztwt codex is the best choice for 20$ plan at the moment. GPT 5.5 is top notch for most tasks, 5.3-codex is cheap, almost like unlimited, and can be used for most lightweight tasks.
Deepseek v4 pro is on par with 5.3-codex, but nowhere close to GPT 5.5.

[Jayadeep KM @kmjayadeep]

Tried deepseek v4 pro yesterday. I'm impressed by the quality/price ratio.
Definitely nowhere close to frontier models as most people hype on x.
I would still use GPT 5.5 for majority of the tasks, and delegate some sonnet level tasks to deepseek v4 pro.

[Jayadeep KM @kmjayadeep]

@builtwithjon @Teknium @aronprins @dotta @adam_x_mentis Nothing fancy, just instruct the CEO to keep improving the agents when they makes mistakes.
Something like this
https://t.co/LTO5WYxZEB

[Jayadeep KM @kmjayadeep]

(4/4)

I’m open sourcing it here:

https://t.co/6EWHDRukTR

Built from real usage, not just demos.

I will write a blog post with more details soon

[Jayadeep KM @kmjayadeep]

(1/4)

A few weeks ago, I shared my honest thoughts after running Paperclip AI for 2 weeks.

Then I ran it for 2 more weeks on a real project: k8s-ai-sre

https://t.co/9SZ8kqRBTD

[Jayadeep KM @kmjayadeep]

(3/4)

Based on what worked, I created an agents template that can be reused with any project.

It captures the setup and patterns that were useful in practice, so you do not have to start from scratch each time.

[Jayadeep KM @kmjayadeep]

@builtwithjon @Teknium @aronprins @dotta @adam_x_mentis Love the idea! This is the same direction i was planning, you are way ahead of me :D
What worked for me is to use a frontier model for CEO, and let it do staff management + project management.

important: It needs a self-feedback loop to keep improving the agent instructions

[Jayadeep KM @kmjayadeep]

@builtwithjon workflow and config management is a pain point with papeclip. the native way of using AGENT_HOME is too unreliable.
I ended up managing the agent configs myself, and using paperclip purely as orchestration layer. Here is my reference structure
https://t.co/6EWHDRukTR

[Jayadeep KM @kmjayadeep]

I ran Paperclip AI for 2 weeks to manage open-source projects.

It’s not an “autonomous AI company,” but it is an interesting control plane for multiple agents.

My honest thoughts:
https://t.co/3zkSnwlA1s

#paperclipai #homelab

[Jayadeep KM @kmjayadeep]

Have been playing with #paperclip from @dotta .
Really love the concept. It definitely took some time to get it to a productive state (ngl, there are some bugs at the moment). But going well so far.
Detailed post coming soon. Stay tuned :D

[Jayadeep KM @kmjayadeep]

AI doesn’t reduce platform demand. It amplifies it.
Read the full article in my blog:
https://t.co/hmXiUSAlgx

[Jayadeep KM @kmjayadeep]

AI is not replacing DevOps engineers. It’s making us more valuable.

After 10+ years in Platform/DevOps, here’s what I’m seeing:

[Jayadeep KM @kmjayadeep]

4/ AI is increasing output across teams.
More code. Faster iteration. More experimentation.
That creates pressure:
• Stronger pipelines
• More stable infra
• Better security
• Scalable platforms
Faster dev cycles increase the need for mature DevOps.

[Hunter Kallay @lazukars]

R.I.P few-shot prompting.

Meta AI researchers discovered a technique that makes LLMs 94% more accurate without any examples.

It's called "Chain-of-Verification" (CoVe) and it completely destroys everything we thought we knew about prompting.

Here's the breakthrough (and why this changes everything): 👇

[Jayadeep KM @kmjayadeep]

Is there anyone who likes #opencode more than #ClaudeCode ? I really want to use it more often. But somehow claude code seems far ahead (except the stupid scrolling bug). Am I missing something?