Aditya Aggarwal

I'm here to make corporate complaints and shit post. and I'm all out of complaints right now

[Aditya Aggarwal @adi__ityadi]

Claude just casually reminding me that getting a 5090 in my 6 year old desktop doesn't mean it's "latest" now.

[Aditya Aggarwal @adi__ityadi]

While building Clara, my core objective is to build a temporal contextually aware layer. $5/M models won't be for the masses unless the context layer is efficient enough.

[Aditya Aggarwal @adi__ityadi]

3 days and 40M tokens on an entry-level model later. Harness now has the right tools and performs as well on a $0.3/M token model as it does on a $10/M model.

70 out of these 76 prompts now take as low as 1000 tokens per call. T

[Aditya Aggarwal @adi__ityadi]

Corporates are already seeing the impact of "unlimited tokens" in their bills. AI apps are soon going to be about context, memory and token efficiency, more than just trained intelligence.

[Aditya Aggarwal @adi__ityadi]

Overnight token burn just testing a token-efficient harness in Clara at https://t.co/nXbpxkkeRT only for a jackass from school to tell me "But bro, friends and family ke liye to free hoga na" seeing a $10 price tag.
Doing this with Hermes/Openclaw would have burnt over 100M

[Aditya Aggarwal @adi__ityadi]

Published a small note on the most dangerous bug in solo building: the one that looks like productivity. I dub it, The One More.bug

One more feature. One more polish pass. One more day.

At some point, “not ready yet” quietly becomes “afraid to launch.”

https://t.co/dm31bq6YO2

[Aditya Aggarwal @adi__ityadi]

@UltrahumanHQ @deeppurpled Thanks, Abhimanyu from your team reached out to me, explained the issues and offered multiple solutions. He also acknowledged the gaps in support and assured that you're already working on fixing them. Takes a while to find the right channel with you folks, but I'm glad you're on

[Aditya Aggarwal @adi__ityadi]

@UltrahumanHQ I bought the Ring Air in Nov 2024 from your official site for ~₹27k, plus UltrahumanX warranty.

I am now on my 3rd ring, and it has failed again.

At this point this is not a one-off defective unit. It is a reliability and support failure.

[Aditya Aggarwal @adi__ityadi]

@UltrahumanHQ @deeppurpled Sent you a DM with my email.
Kinda surprised the 2 caseIDs I shared earlier are not enough for you to find the issue. Does the bot just generate them in chat?

[Aditya Aggarwal @adi__ityadi]

Tried posting on reddit since the support used to be very active on r/ultrahuman. Guess what? You have to create a comment in the megathread. Which is archived since atleast march. @UltrahumanHQ
@deeppurpled congratulations, you've built a company worth following (to downfall)

[Aditya Aggarwal @adi__ityadi]

I paid for a premium wearable and extra warranty. I do not want endless firmware experiments or chatbot loops.

I want a working product or a proper resolution from a human.

Anyone considering Ultrahuman Ring Air should think twice until reliability and support improve.

[Aditya Aggarwal @adi__ityadi]

While you're here:
Checkout: https://t.co/w3m9NqkjAB

[Aditya Aggarwal @adi__ityadi]

If you aren't using @openclaw already, start now.

Free models from @OpenRouter , such as owl-alpha are great. And you have a personal assistant ready to go on any platform you like. I send voice messages on WhatsApp, and everything is ready for me when I start work.

[Aditya Aggarwal @adi__ityadi]

This is awesome for solo founders who are building by themselves and need to bounce ideas or keep logs. I rarely use my agents as self-running coding agents, but they're great for researching and scraping for marketing activities.

[Aditya Aggarwal @adi__ityadi]

Switched local agent manager's dumb subagents from claude to tencent's Hy3 preview.

This is just the automated jobs. I was spending $100/week on sonnet/haiku. Claude still does the complex stuff. Hy3 does the everyday stuff for free..

[Aditya Aggarwal @adi__ityadi]

Created a small script to use kimi and deepseek as a "Junior Assistant" to claude code. Haven't hit the Pro limit on claude since then, and evidently, kimi is much more precise in summary generation and context gathering, so the quality has gone up too.

[Aditya Aggarwal @adi__ityadi]

@sarvamAI @vivekRag I see Sarvam advertises the Text Gen models as "Free" on the pricing page, and then includes the charge on the dashboard.
Is this a marketing thing since the Model is "free" on Indus, or are your pages out of date?