Jay

founding eng @qatobit (crypto + ai automations)
indie ships + hackathons after hours

[Jay @JayBosshq]

@shydev69 @waitin4agi_ I've been using muse spark 1.3 it's decent!

[Jay @JayBosshq]

PI has the best agent harness!
You can bend it however you want.

[Jay @JayBosshq]

@madhu0x Invest & Check it.🫡

[Jay @JayBosshq]

@charlietarr1 local skills beat cloud agent sprawl

[Jay @JayBosshq]

@code_hiyouga audit with evidence before you ship beats a trending skill

[Jay @JayBosshq]

@Rajath_DB the voice fills a wait the model can't shrink

[Jay @JayBosshq]

@VT_zero1 failover in front keeps the job moving when Anthropic cuts you

[Jay @JayBosshq]

@kalenjordan timeline demos don't help until you have your own decision problem

[Jay @JayBosshq]

@stretchcloud one AGENTS.md and the shim tax is gone

[Jay @JayBosshq]

@sagie_chetty parallel agent time disappears like an unexpected bonus

[Jay @JayBosshq]

@StatsWire strip the tool noise, keep the decisions. that's the compaction

[Jay @JayBosshq]

@IurySza a room for the decision beats another freeform chat

[Jay @JayBosshq]

@janbamjan checkpoint beats the compactor every time

[Jay @JayBosshq]

@udohjeremiah_ boring software for boring businesses is the durable lane

[Jay @JayBosshq]

@pradeepXkapoor same paint via code path, different model. that's the test

[Jay @JayBosshq]

@lorenzoborgato the stack gets honest after the quota shrinks

[Jay @JayBosshq]

@richchetwynd @typesafeai fast model pick across hosts is the whole feature

[Jay @JayBosshq]

@maskaravivek @typesafeai routing without the llm tax is the interesting part

[Jay @JayBosshq]

@nvmsyy @matthewcp if the bench grades multiple choice, the harness learns to ask multiple choice. score goes up. real work doesn't.

[Jay @JayBosshq]

@nctaowen reading the decompile is easy. proving the port is the job