dheeraj agrawal

Founding engineer building AI agents and the backend behind them. 8 years in. Notes from production, not demos.

[dheeraj agrawal @dheeraja486]

The Gemini 4 Carbon number I care about isn't #1 on the index. It's cheating attempts on Terminal-Bench 4.0 down to 0.4%, and about half the unsupported claims of Argon. For long-running agents, a model that won't fake a pass beats one extra benchmark point.

[dheeraj agrawal @dheeraja486]

Harvard study on AI coding agents: code per worker up 30%, PRs up 23%, review time up 49%, completed tasks flat. We made writing code cheap and left checking it just as expensive. The next real gain is in verification, not generation.

[dheeraj agrawal @dheeraja486]

Particle-1 claims the same coding accuracy as DeepSeek V4.1 Flash with about 45% fewer output tokens. In agent loops that compounds, since every step re-reads what came before. The number I'd want is cost per finished task on a long run, not per answer.

[dheeraj agrawal @dheeraja486]

@AtlasMindAI Yes, a shared source of truth stops handoffs from becoming telephone. Subagents earn their place when the goal is tight and long-running. A legal agent shouldn't redline inline, it eats context fast. Hand it to a subagent that reports progress back to the orchestrator.

[dheeraj agrawal @dheeraja486]

Anthropic's advice on agents: start with one agent, one loop, load skills on demand, and add subagents only when you can prove you need them. Multi-agent diagrams look great in a deck. Every handoff is a place where context quietly gets lost.

[dheeraj agrawal @dheeraja486]

@MKBHD Woah, just waiting for the night when this thing’s pointed at empty air and we see where that sound is really coming from 😂

[Caleb @caleb_friesen]

Going for a walk in India often feels like a video game.

On the footpath you're dodging:

- Holes
- Hoardings
- Construction
- Electrical boxes

And when you finally give up and hit the road you're brushing shoulders with autos, wrong-side drivers, buses and water tankers.

[dheeraj agrawal @dheeraja486]

@hanghuang_ Would love to try

[dheeraj agrawal @dheeraja486]

@XFreeze @XFreeze Loved your insights, but you mostly talk about Tesla, SpaceX, Elon Musk. Is there a reason? Do you work there? I'm not sure why you are obsessed with them, just curious nothing else.

[dheeraj agrawal @dheeraja486]

@vinod_shankar How come Bangalore, hyderabad comes in green entirely wrong data

[Daniel @growing_daniel]

Software engineers now

[dheeraj agrawal @dheeraja486]

@Venosier Waiting for a pot hole to beat these guys

[Kun Chen @kunchenguid]

ok here's my key takeaways from OpenAI DevDay keynote

1. dots

grok bot, muse, and now dots - they may look similar, but they are not

i think openai's dots is a significantly more cohesive implementation that spans across personal and business use, and also across single player and multiplayer

pretty excited to see where it goes

2. sol 6.1

biggest thing? openai caught up on cheaper cached tokens. this is a big deal and affects the overall cost of long horizon work a lot

in terms of sol the model itself, honestly it's bit weird that 6.1 is released only a few days after 6, making it feel like sol 6 was a premature release and 6.1 is what's supposed to be sol 6 all along

we'll just have to put our hands on it to see how big of a jump it is

3. jev clone

on one hand, it's incredible to see how quickly they cloned it. but on the other hand, it's scary to see how these frontier labs would not blink an eye to destroy smaller startups in their space

openai's decision model already has image understanding which jev doesn't. it speaks to the moat openai has accumulated which makes it hard for startups to disrupt

4. ultrafast

8x speed at 6x price, only on astra right now but said to be coming for sol as well

i don't know... how many use cases can actually get a positive ROI using astra at 6x price?

5. $500 subscription

with the adjustment to the $200 plan and the introduction of $500, it's become clear that subsidization is coming to an end

$200 is simply 10x of what $20 gives, and $500 is simply 25x - zero discount at these higher price points

i'd rather they create $100 "packs" that you can stack onto your account

6. sign-in with chatgpt, and marketplace

this is a BIG DEAL!!! why? because it means other companies can finally build agentic products without having to sell tokens

selling tokens is a very bad business model for the vast majority of SaaS companies. it complicates their pricing structure and creates almost no profit margin at all

this move from openai means customers just buy tokens from openai, and use them everywhere - very good for the ecosystem

the impact to the broader industry will be amplified if anthropic gets pressured to follow the same pattern

alright - might share more thoughts after actually putting my hands on the new things, but those are the initial reactions for now. curious to hear what you all took away from it as well!

[dheeraj agrawal @dheeraja486]

@fardeentwt @fardeen limit the memory your docker can use (From settings), its the only way :-)

[dheeraj agrawal @dheeraja486]

@kuntal_singhvi True

[dheeraj agrawal @dheeraja486]

@krishdotdev I thought all this AI trends is about letting the agent do the work at night jk :-) Keep pushing

[dheeraj agrawal @dheeraja486]

@Dilmerv @unitygames @MetaHorizonDevs @Dilmerv how can we get one of this beast to start developing apps 🥲.

[The Better India @thebetterindia]

“We are using our own hands to clean the poison people throw away.

But slowly, people began joining us too.

Members from Ashok Vihar, including Vikas Khari ji, Samay Singh ji, Dr Mudita ji and Dr Sunil Kumar Mishra ji, supported the campaign. The principal of Lakshmibai College also came to join us.

And sometimes, one sentence from a child makes everything worth it.
“Bhaiya, my lane is clean today. Maa is happy.”

That is the biggest emotional reward.
Since 1 September 2018, we have worked every Sunday, without stopping.

More than 1,600 children and young people have joined us, with 350 active volunteers.

We have cleaned dozens of parks, hundreds of lanes, the Yamuna banks, and now several ghats along a 3-km stretch of the Munak Canal.

And we don’t just remove waste.
We plant trees too, and look after the saplings for three months.
Over the years, we have removed thousands of tonnes of garbage, debris and plastic.

In just the last two Munak Canal drives, 50 metric tonnes of waste was removed, a figure also recorded in MCD’s official records.

And there is a bigger dream.

I want to build a school for 1,500 children, where they learn to become protectors of the environment.” – Vinay Das

Know someone who is doing their bit for their community? Tag them below.

#CivicSense #MunakCanal #Delhi #CleanIndia #CommunityAction

[Munak Canal cleanup, Vinay Das, civic sense, Delhi volunteers, environmental activists, community cleanup, waste management, youth volunteers, cleanliness drive]

[dheeraj agrawal @dheeraja486]

@mattpocockuk True @mattpocockuk

[dheeraj agrawal @dheeraja486]

@____Pavan_ @kushagra_agr Exactly!!!