Ish

0 to 1 Builder✨ | Prev. partner at @elixir_capital & @Woodstockfund
| Engineer @bitspilaniindia

[Ish @DecisionTree_gg]

Everyday it feels is like I’m waking up in a Sci-Fi dream

[Blue Origin @blueorigin]

Melt. Extract. Breathe. Repeat. 🧑‍🚀

From Moon dust to fresh air, our Air Pioneer technology turns lunar regolith into breathable oxygen, ready for astronauts returning to the Moon. At our Space Resources Center of Excellence in LA, we developed a reactor (left) that melts regolith simulant and passes a current through it to release oxygen and other gases. The gases flow into the purification system (right) and emerge as medical- and propellant-grade oxygen. A flight-qualified Air Pioneer at this same scale could provide the first breath of life for a sustainable Moon base. 🌕

[Ish @DecisionTree_gg]

but ma'am...an alien barfed on my science project

[Ish @DecisionTree_gg]

@andy_matuschak the noticing has made me suspicious of my own writing. now i overcompensate by leaving in typos

[Ish @DecisionTree_gg]

@CassiniRes hey Juno!

[Ish @DecisionTree_gg]

As a new coder, may I please request @github to fix it's search.
Yes, I just did find a repo on Insta that I then searched for on @github and the search sucks. I understand why it wasn't a priority so far - but I present to you my user journey.

[Ish @DecisionTree_gg]

@ozfurka Thanks! The messaging really spoke to me - see you!

[Ish @DecisionTree_gg]

what are the best co-working places in San Francisco and around the city for start-up people?

[Ish @DecisionTree_gg]

The shared channel is our internal Slack #llm-agents-raw (not GH PRs).

we post the full live trace + complete prompt history the moment the agent run finishes. the trace has every tool call, all intermediate outputs, and the final diff so we catch loops or drift right away.

GH PRs only come later for the actual code changes.

Claude Code buries the full prompt chain pretty badly. we pipe everything through our tracing wrapper (LangSmith/Langfuse style) so the raw prompts + system messages are one click away.

[Ish @DecisionTree_gg]

Actually, more than co-workings, I'm seeking good communities to hang in

@lans_sf @fdotinc @theresidency @joinodf HF0 @southpkcommons @jointhebridge have popped up so far - but i need help in curation -> airbnb location selection lol

[Ish @DecisionTree_gg]

it's truly sad that the product sucks, still - your marketing team is doing all the directionally correct things. but if dara's delphi avataar can't keep me engaged for more than 1 mins - then truly, what is the point?

the form factor of human-AI interface at scale will HAVE TO, look diff

[Ish @DecisionTree_gg]

@McNamara_Philip @AnthropicAI @OpenAI @meetgranola @googlecloud Woah...this feels - too good to be true?

[Ish @DecisionTree_gg]

too many audio prompt drops today - with @ChatGPTapp :(

[Ish @DecisionTree_gg]

HOW AM I TODAY DAYS OLD for finding @BlackFlagVC

[Catherine Yeo @catherinehyeo]

Excited to see Altara (@altaratech) on this year's Black Flag 100, a list of one hundred deep tech startups nominated by 500+ deep tech founders.

Grateful to be on a list with so many incredible companies pushing the frontier.

[Ish @DecisionTree_gg]

@Jas_Jaski @NousResearch wow, et tu @GeminiApp

[Ish @DecisionTree_gg]

@Jas_Jaski me on the same day with @NousResearch / Hermes

[Ish @DecisionTree_gg]

i love being a vibe coder. fwiw i did code in college, and i am a trained engineer. my career in VC took me away from code

but we are back, and how!

[Ish @DecisionTree_gg]

@rstagi_ +12.2pp and -85% tokens in the same run is wild. is glm-5.1 winning because the model is better at tool selection, or because BM25 + gateway hides the catalog complexity from any model?

[Ish @DecisionTree_gg]

@isaac_ts_way escalate's the easy part. the hard part is firing it at the right time — you usually only catch 'going in circles' after 2-3 wasted turns. could a separate watcher agent read the trace and call it? happy to dig in

[Ish @DecisionTree_gg]

@guilhermeotina @yoheinakajima only way out i can think of: different agent writes the test vs the implementation, different prompt. and you grade the test-writer on whether it catches stuff it didn't design for