@illyism@igormomentum Not mutually exclusive, and a healthy customer base should have both. Tokens are cheap so it’s just lazy not to best serve both. And a text only mcp isn’t serving anyone but developers. If you’re not mixed chat/ui as plugins you’re probably just doing the bare minimum.
Both can happen and are not mutually exclusive. When an organization has enough funding and folks working at it with existing things in place the pain has to be pretty high to change cause change is a cost and risk. They’re also more familiar with op-ex costs vs risky cap-ex to change with unpredictable long tail op-ex (cost of ownership). For them it’s a more important question during original signup then this week let’s code our own x. For smaller orgs they don’t have to deal with strict negotiation for budget allocations, yearly reviews etc. There are exceptions always but the majority is like trying to turn the titanic.
@JackEllis@davidgomes@poteto@ericzakariasson You can set the project workspace to always use a preset list with some conditions then you don’t have to worry about it (local or cloud)
@brankopetric00 doesn't matter if its ai or hand made, its off brand for him, if you have watched his other videos you might spot the difference between his story telling and this. Brand and real story telling are the difference.
@jcinjpn Jev isn’t reasoning. So no forward planning, no strategy over multiple moves, of course they’re bad. It’s zero reasoning intelligent eval. Fast and cheap as with structured output. Great worker for specific tasks. Not coding or solving the next 1000 yr math challenge
@jessethanley@Bento given data and sample size, I don't know why you haven't created a private classifier model that you update nightly at this point, single sample from batches to feed to supervision that then trains allows EVERY message to be validated.
Na I think it should just default to the fastest models, then have a yolo mode for when you don’t want to iterate. this is for local. Cloud agents should default to yolo.
This assuming faster are less reasoning and less expensive. Yolo being higher reasoning models.
I would expect the cost to be effectively the same for small to medium sized well defined features.
One you drive the other you are paying the model to drive. Many fast cheap iterations are going to be more expensive then a single high reasoning pass, but your trading time, involvement and possibly tighter fit to what you have in mind as a cost burden assuming planning is a sunk cost for any approach.
@theo Broadly speaking this is skill issue, and the amount of micro vs macro vs 10k ft understanding a dev has had is pretty directly proportional to their ability to summarize and synthesize the value vs the fluff. Inversely also why IC specializing in a space drove outsized value.