Abinash Meher

Crafting agents as CTO @itoflowai

[Kyle Mistele 🏴‍☠️ @0xblacklight]

if you're building an agent or a harness, you need to be an order of magnitude closer to the code than you would be normally

models are VERY bad at building harnesses & agents - not sure if it's lack of training data or smth else

but they have exceptionally bad intuitions about context management and the agent loop and caching and a lot of other components that matter a LOT

it is very high-leverage to design the code yourself - if not typing it by hand, still do the program design yourself

[Abinash Meher @AbiMeher]

@NathanFlurry I’m somewhat confused about the “flavour” of isolate here and the one on CF. https://t.co/e1OliipIeL claims it would run CPython, does this mean it can run cpython inside the isolate?

[Abinash Meher @AbiMeher]

Rancid is the perfect addition to the vocabulary to describe some of the stuff I’ve been seeing

[zeb @zebassembly]

So far I've only used Rust with Astra, but so far this code is like the sloppiest slop i've ever seen. It all works well and the abstractions are pretty damn good, but holy shit this code looks rancid.

[Abinash Meher @AbiMeher]

@threepointone The asymmetric advantage would most likely be talent and distribution

[Abinash Meher @AbiMeher]

@alexanderrX_ At the end of the day it's going to boil down to value provided per unit spent, it's much more obvious in tech right now, but probably the patterns in finance haven't been as well known yet

[Abinash Meher @AbiMeher]

Most of these AI models are trained to do great on coding tasks which are easy to test, the same cannot be said for a lot of finance related tasks, there's no feedback loop to say that the model is doing a good job.

Some folks in investing we showed it to either refused to admit it did a decent job for a draft from one shot but also realise that the wall of text probably needs someone sitting down to ensure it followed proper rigour and all numbers are cited, so they are jaded.

We are trying to improve and change this at @itoflowai

[Gabriel Chua @gabrielchua]

With GPT-6 Astra, we've released three new API updates for building agents

@OpenAIDevs

[dex @dexhorthy]

SlopCodeBench Update - Astra has entered the chat

[Abinash Meher @AbiMeher]

@dexhorthy This is my guilty pleasure, like watching those marble races

[Abinash Meher @AbiMeher]

@arcyton @thomas_ankcorn We are only thinking about "bookmarking" interesting / complex prompts thrown at us, it's still very manual

[isabel @artficialisabel]

the openai huggingface incident, from an agents pov.

(part 1)

[Abinash Meher @AbiMeher]

@ashleypeacock A brand new DO; a concrete use case, I was researching something in an agent and it brought up an alternative research path. I want to follow both of them, or fork it into a new one to hand it off to a colleague

[Abinash Meher @AbiMeher]

@DhravyaShah @_prasanna_ap Is this Linear?

[Abinash Meher @AbiMeher]

@dok2001 @Cloudflare An email inbox as a Think primitive for an agent

[Abinash Meher @AbiMeher]

@theo Used up 20% of my weekly budget in under 2 hours

[Abinash Meher @AbiMeher]

@thomas_ankcorn Not much, just this

[Matt Pocock @mattpocockuk]

AI has eaten tactical programming, so juniors need strategic experience in a low-blast-radius environment.

Here's one way to make that happen:

1. Give them responsibility over a large chunk of work
2. That work must be low stakes, but not no stakes - ideally an internal tool (AI makes these kinds of projects easy to justify)
3. Let them move fast and fail fast. Give them the same AI budgets as your senior engineers.
4. When failures happen, work with them to figure out what went wrong (learnings can also be pulled upwards into the rest of the org)

In a sentence:

Hire juniors. Give them work that matters. Watch them fail. Pick them up.

[Abinash Meher @AbiMeher]

@dok2001 @Cloudflare A full blown Data warehousing and analysis platform, like Snowflake