Rethinking AI with Jev: Instant, Nearly Free, and Unlocking What's Next - MEGA.dev LIVE
Most AI in apps still lives in a chat box. You send text, wait a few seconds, then parse the text that comes back.
In this session John Lindquist walks through a different setup: Jev, a model from TypeSafe that takes messy input (speech, documents, live feeds) and returns a typed decision your code can act on, along with how confident it is.
The numbers TypeSafe publishes: responses in 70-500 milliseconds, $0.042 per million input tokens, and output that costs nothing. In John's own tests, voice to-do decisions took a median of 156 ms across 28 test phrases. A timer and a cost counter stay on screen for the whole session, so you can check each claim against the run in front of you.
What we have planned (demo walkthrough, no live coding)
Words into actions - a tool picker, a voice-driven browser and an app launcher that says no when it can't help
Messy data into structure - matching records, splitting a PDF into sections, fixing mixed units without guessing the unclear ones
Live data - earthquakes, GitHub issues, Hacker News and a Wikipedia link race where every hop is a decision and the whole race costs less than a penny
Where it breaks - the first version saved "uh" and "you know" as tasks. The fix came from asking better questions, with the same model
Jev as the front door - a ladder from plain code to a big reasoning model, where Jev's confidence decides whether a request gets answered or passed up
Start with inference, then switch to code - watch the decisions, and once a pattern settles, turn it into a plain rule
Who it's for: developers who already build with LLMs and want AI in places where a chat box and a multi-second wait don't fit.
Format: about 60 minutes of demos, a short word from Greg about MEGA.dev, then about 30 minutes of live Q&A with John.
The fine print, up front: the speed and pricing figures are TypeSafe's published numbers. TypeSafe says the pricing may be subsidized and expects it to go down. John's measurements were taken live and depend on the network (TypeSafe's service runs on the US West Coast). Being fast doesn't make an answer right, and the session shows how to handle that.
About the host
John Lindquist (egghead.io) is the Author of Week 4: Fast Takeoff in MEGA.dev, a 4-week program for developers who already build with AI agents. Program runs Oct 26 - Nov 23.