…and create an environment where people are learning from each other
Okay yes this is why we built traces. And you don’t have to use it if you don’t want, but please don’t throw away your session data!
Traces are not digital exhaust and shouldn’t be treated as such
Judging developers based on tokens is like judging a contractor based on watts consumed by their power tools
There will be loose correlation to outcomes but really it’s useful only to isolate LLM skeptics
The better way is always to look at their new work product (sessions)…
Introducing a minimal training harness built on prime-rl and verifiers, so you can now train your own RLMs without sandboxes! All available in the `training/` folder in the RLM GitHub repo!
We train RLM-Qwen3-30B-A3B-v0.1, using RL on a separate split of environments (OOLONG-Spam, BC+ split) to greatly improve performance across the board on long-context tasks evaluated in the original RLM paper.
We trained for a day on an 8xA100 using prime-rl; code and model are open-source and available on GitHub / Huggingface.
I walked into new stadium today to demo traces, honestly didn't realize this was being recorded on video lol
anyways turned out to be a crazy fun session, we cracked a lot of jokes, there were a few bugs and i gave out too much alpha but its all good
stay tuned for more like this
A behind-the-scenes working session with Tarun Sachdeva (@tarunsachdeva) for an upcoming film collaboration.
Tarun is the founder of Traces (@tracesdotcom), a platform for sharing the work you do with coding agents.
In this conversation with @internetvin, he walks through what a trace actually is, opening with a single session and pulling on it until the whole product unfolds: profiles, teams, the CLI, integrations, and the discover feed.
Timestamps
00:00:00 What is a Trace?
00:06:55 Categorization, Status & How the Summary is Built
00:08:30 The Trace as Atomic Unit
00:10:30 What Happens Between Linear and a Code Base
00:12:55 The Middle Made Visible
00:14:00 Andrew Nesbitt & Everyone is a Beginner
00:18:00 Project Baton: The CLI
00:22:40 Publish, Resume & Git for Agents
00:25:00 Single-Player vs Multiplayer
00:30:00 Integrations: GitHub, Hugging Face & Linear
00:33:50 Why Tarun Stopped Using an Issue Tracker
00:35:00 The Discover Tab
00:37:00 JC, Jake & Reading Other People's Traces
Amazing demo of @tracesdotcom by @tarunsachdeva. Instantly one of those ideas that you know is going to be big, sharing agent sessions across teams across harnesses, make your agent sessions multiplayer. Massive unlock for teams. #TorontoTechWeek@devtoolsTO
in traces v0.6.0 you can
- search your local sessions across agents
- launch any of the sessions you see directly in the agent cli (currently mac only)
- view sessions from other people in the orgs you're in, directly from the terminal
I will be demo'ing this and a few other things tonight at @devtoolsTO
Hung out with @sojoodi and a bunch of developers at the 8090 office yesterday for a Builder Hour workshop (code and traces were shared)
First time we used Traces in a community-meetup setting, was surprisingly effective. Definitely doing more of this
Yesterday's Builder Hour session recap:
- built a UW Coop Eval form collaboration tool in "1 hour" (it took a couple of hours after the session to complete the project, but still within a day) - github link below
- traces published - link below
- deployment today (won't share the link ;) ) but screenshots below
- total tokens used, 1.8MM mostly GPT-5.5/high
- most importantly, met 15+ cracked AI builders some of whom built with me live during the session! -- THIS IS WHY I HOST BUILDER HOUR! 🙏