Alex Gonch

Founder & CEO @ - AI employees that turn meetings into action 🚀. Dubai 📍 | PM geek 🤖 | Cat person 🐱

[Alex Gonch @AlexGonchX]

@adhamuxi what did it beat for you? curious if this is benchmark good or you actually gave it a messy repo

[Alex Gonch @AlexGonchX]

@DanDr1s companies won't replace people the day a model clears some benchmark. they'll spend years arguing over who owns the mistakes.

[Alex Gonch @AlexGonchX]

@ChrisGPT the first image looks incredible until you zoom in. half the text is still nonsense

[Alex Gonch @AlexGonchX]

@TTrimoreau trust. i'm asking people to connect their customs data and let us touch a filing that can cost them real money if we get it wrong. attention doesn't solve that.

[Alex Gonch @AlexGonchX]

@DanDr1s the chart doesn't say 93% of all answers hallucinate. it says 93% of non-correct responses were incorrect rather than partial or refused.

[Alex Gonch @AlexGonchX]

when you remove a condition from a rules engine, what happens to the old rules using it? keep the old behavior forever or migrate them and hope nothing silently changes?

[Alex Gonch @AlexGonchX]

@Da7_Tech what made it fall off for you? worse output or the usage limits?

[Alex Gonch @AlexGonchX]

@sandislonjsak one thing like this and now i'm checking every button before i click it

[Alex Gonch @AlexGonchX]

TechCrunch says “our team handles the rest.” then the FAQ says you pay for it and handle registration, promotion, insurance, legal, and the actual event. getting listed is useful. but you are very much handling the rest.

[Alex Gonch @AlexGonchX]

@sandislonjsak that screen glare would end my research immediately. i'm going swimming

[Alex Gonch @AlexGonchX]

@wagslane half the chat features I try are just a blank box asking me to explain the product's own workflow back to it

[Alex Gonch @AlexGonchX]

@AdolfoUsier @opencrabs what got better in practice? does it notice when it took a bad turn, or do you still have to drag it back?

[Alex Gonch @AlexGonchX]

@BrandGrowthOS nothing leaving the phone is great. i still want a record of what it saw and what it called before it took an action. debugging that after the fact sounds rough.

[Alex Gonch @AlexGonchX]

putting a human approval step into harness. sounded like two buttons. it’s now 3 states, 2 ways to answer, expiry cleanup, and a crash case that leaves the approval sitting there.

[Alex Gonch @AlexGonchX]

Dutify rule editor has 73 conditions in my test account and the UI already started pretending that's fine. it is not fine. now I'm making the ugly cases take more space.

[Alex Gonch @AlexGonchX]

@wagslane it’s always some boring line too. the agent deletes a weird guard, the tests pass, and now checkout breaks only for annual plans created before 2021

[Alex Gonch @AlexGonchX]

@DanielGlejzner that orange wall is probably a lot of agents running create-next-app because every tutorial on earth taught them the same default

[Alex Gonch @AlexGonchX]

@stanlee0nX $99 is nothing if you’re already a company. it’s a lot if you’re 16 and just trying to put some stupid weekend app on your phone

[Alex Gonch @AlexGonchX]

@IamAroke it gets expensive when the weird case disappears into the abstraction. six months later nobody knows if the rule is in code, config, a toggle, or someone's ops spreadsheet

[Alex Gonch @AlexGonchX]

@GohilHardy it’s the part where the patch looks fine and still breaks billing for one customer on an old plan. the better dev smells that before merging