Amjad Masad

Software isn’t merely technical work anymore. It’s creative. Introducing Replit Agent 4. The first AI built for creative collaboration between humans and agents. Design on an infinite canvas, work with your team, run parallel agents, and ship working apps, sites, slides & more.

@TheRiverCard Hmm… that shouldn’t be the case. @ReplitSupport can take a look

@amasad One can't change deployed region for deployed apps ever. Only alternative is to create new repl, migrate db etc. Early customers are likely impacted. Requesting your push for prioritisation.

@mrboolean01 @esaagar My grandfather’s house in Haifa, for one.

@esaagar It’s really expensive to market murder and theft. Might be cheaper to stop killing and stealing.

We're hiring an Enterprise PM at Replit. Everyone on this app has a take on tokenmaxxing, AI sovereignty, FDEs, and agent governance. In this role, you'll turn these important, evolving concepts into products used by some of the world's largest companies. Enterprise is our fastest-growing segment. Interested? DM me.

@praveenkoka LoRa fine tuning an 8b model is cheap

@amasad Early vibe coding, sure. Except now you're $500+ deep in monthly compute costs before you even have a working model.

Getting realtime progress updates on my model training runs. This feels like early vibe coding except it’s making personal models.

@FamilyFireGames @Replit That’s that’s awesome… thanks for sharing

With @Replit anything is possible.

Our code

Paradoxically is as frontier LLMs getting better might lead to less usage because they will enable enterprises to do their own ML research and move away from expensive models and distill known use-cases into smaller models. Programming languages might be a useful analogy. You use JavaScript and Python to move fast but for hot code paths you write C or Rust code (even assembly sometimes). Dynamic languages here is like frontier/expensive tokens. It’s the easiest to build with but once you understand the use-case and have the data you can distill that into a cheaper model. Presumably most of CPU cycles today go to C/C++ because that’s what operating systems and browsers are written in but new apps will usually start with high level languages before they start optimizing. And the analogy fits because recently we’re seeing LLMs help programmers rewrite code with lower level languages (eg Bun’s rewrite to Rust was unlocked by Fable). Same thing will happen with ML.

The mega bull case for AI infrastructure would be *if* market share shifted away from certain frontier labs with 90%+ inference margins toward cheaper models, whether open-source or closed. It would increase the ROI on AI spend for end customers by increasing intelligence per dollar, which would drive incremental token demand. Margin dollars would effectively get redistributed from the frontier labs to AI infrastructure providers. The infra winners would be those with the lowest per token cost and the winners at the model layer would be those with the highest token efficiency. There are many reasons Jensen is so focused on open source, but this is likely the most important one as I think he is probably less worried about a monopsony these days. Lower margin % at the model layer = more margin $ at the infra layer all else equal. With SpaceX and Meta being vertically integrated and possessing the #3 and #4 models respectively it is more possible than ever. Note that Grok 4.5 is ahead of Fable for some useful tasks at a much lower cost, so ranking them #3 is conservative. This is not happening yet. Cheap, mostly open source tokens are likely the majority of volume today but the majority of economic value is still accruing to the most intelligent models. Might change though. We will see.

@AIVibeCoding @Replit No been there for a while. Make sure you have “app testing on” in the chat toggle

@amasad Is computer use new for @Replit? Using Replit for Vibe Coding Summer Camp at The Bolles School (Jax, FL) and this would be a game changer!

Fun to see Replit's computer use model play against my new chess engine

@danielinnov8 @Replit super cool

@amasad This is awesome, huge chess fan here. I built this Chess Coach entirely in @Replit chess-ai.app check out some of the famous games, has puzzles and lessons.

Vibe Research Fine-tuning a Qwen-8b model to play chess on Replit. Running 3 parallel branches with different experiments and making real progress. It's amazing how far models have come in their ability to do ML (they used to be really bad at it). So now someone with good intuition to guide the process could do interesting ML work, even if they have never done it before.

@NextGenMehedi Will share tutorial soon

@amasad Can you share more ? I want to do this myself too, so, I can learn the fine tuning process and see the results. I love chess too

@RaoulDukeDegen It’s fully neural (harder) unlike alpha zero which does Monte Carlo tree search (kinda easier)

@amasad whoa this feels like alphazero but way more accessible without a big team

@afanazizoutbind Real progress initially meant generating good moves then generating coherent moves then improving centipawn reward etc etc.

@amasad A chess-playing Qwen finetune is a nice demo, but what's the actual benchmark @amasad? "Real progress" against what, a random-move bot or an engine that would flatten it in 10 moves?

@ChainZenit Still can’t beat stockfish at lowest level but that’s 1300 elo. Need to find an engine that’s 200 elo to start.

@amasad that's actually sick, how's the win rate looking?

@giacomovenier Yes will publish everything including a tutorial

@amasad Hey! Can you make the project chat public so I can learn from it?

@ThibaultJaigu Replit does all that automatically. Re eval will publish everything when am at a good point.

@amasad How do you route between the branches? Would love to see the eval setup

@SoloBoardroom @Replit That’s so cute. Glad she likes it

@amasad my 6yr old daughter & I were waiting for agent to build this & she was looking at the app & she said “mum whoever built this is a genius” I said “what @Replit?” She said “yeah it’s an amazing game” she loves the magic of watching her ideas come to life 🪄