Ankush K R

Chikkamagaluru boy in Bengaluru.
Software architect, AI tinkerer, business builder. Full-time multitasker.

[Ankush K R @ankushkr03]

I like the direction, but for me the interesting part is not only replacing one provider with another.

While using Hermes, I am realizing the bigger shift is workflow design:

which model should do the first pass, which model should review, what stays local, what needs approval, and how to stop agent loops from quietly burning cost.

That operating discipline matters a lot.

[Ankush K R @ankushkr03]

I relate to this more after trying to build my own small AI office.

At first I thought local AI was mainly about the model or GPU.

But the real pain shows up in the harness: model detection, parsers, tool-call formats, prompt templates, context, review loops, and switching models without breaking the workflow.

That layer matters more than I expected.

[Ankush K R @ankushkr03]

This is a useful way to look at models for agents.

My own feeling is also shifting from “which model is best?” to “which model should do this job?”

For small tool workflows, cheaper/faster models can be enough.
For architecture, review, and messy reasoning, I still feel safer with stronger models in the loop.

Routing may matter more than leaderboard thinking.

[Ankush K R @ankushkr03]

I never thought I would say this, but I miss manually writing code.

Over the last one and a half years, the way I code has slowly changed so much.

Earlier, I used to sit with one problem for hours.

Write some code.
Run it.
It fails.
Add a log.
Debug again.
Call someone and say, “Can you just look at this once?”

Then both of us sit and try 3-4 silly things until one of them works.

That feeling was something else.

Even pair programming had its own fun.

One person typing.
One person thinking loudly.
Both arguing over a small condition.
Then suddenly the bug is fixed and both are like, “Ahhh, finally.”

Now AI does a lot of the first version.

Of course it is useful.
It saves time.
It helps move faster.

But sometimes I feel like I am missing the actual joy of coding.

The small struggle.
The small wins.
The “why is this not working?” moments.
The happiness when it finally works after 20 failed attempts.

AI is amazing, no doubt.

But somewhere between prompts, reviews, and generated code…

I think I started missing the simple happiness of writing code with my own hands.

Anyone else feeling this?

[Ankush K R @ankushkr03]

The interesting part is this:

The future may not be only “more powerful AI models.”

It may also be better AI workspaces where agents know what to do, where to look, when to ask, and when humans should approve.

[Ankush K R @ankushkr03]

I used to think the main job was adding more AI agents.

Now I feel the harder part is different:

How do we organize them properly?

[Ankush K R @ankushkr03]

That is what I am slowly building with Hermes.

Not just a chatbot.

More like a small AI office:
CEO agent, manager agents, worker agents, local models, task board, memory and review loops.

[Ankush K R @ankushkr03]

Same habit after 15 years.

Very different productivity.

[Ankush K R @ankushkr03]

In college, I used to chat a lot on my mobile.

Parents and friends would ask, ‘Who are you chatting with so much?’

Most of the time, it was my girlfriend 😜

[Ankush K R @ankushkr03]

I think this is where AI agents become really interesting.

Not as another chatbot.

But as a layer between us and our work, where simple messages can trigger real actions.

[Ankush K R @ankushkr03]

I’ve started using Hermes as my personal AI workspace.

For office work, personal tasks, business ideas, content planning, and the small things I usually keep in my head.

Slowly building a system to manage work, life, and business better.

Will share the journey here.

[Ankush K R @ankushkr03]

@ShishirNagaraj Good to see this development 👍🏻

[Ankush K R @ankushkr03]

@TraderLogical Angel One Smart api i ll suggest.
You can buy otm options here.
Using it from past 1 and half years.

https://t.co/SnFLZ9gUoe

[Ankush K R @ankushkr03]

Plan to work on non co-related strategies so as to keep equity curve smooth.

Dark green: Ready to deploy
Light green: backested, but not ready for deployment yet

#algotrading #csharp #algo
#angelone #dotnet #optionbuying #optionselling #azure