building traces.com

Joined February 2008
374 Photos and videos
Pinned Tweet
i've basically dropped issue tracking the best way to see what our team is working on is to look at the traces
18
8
284
38,431
Tarun Sachdeva retweeted
Cue the music 😁
266
9,237
50,957
1,113,083
Never, ever throw away your coding agent traces. If it wasn’t obvious before it should be now.
1
2
14
3,128
agent-written PR descriptions are terrible and they silently slow down review we just added a "ELI 15" skill to make PR descriptions easier to review
2
13
1,700
fable in ultracode mode with workflows is really, really good. definitely a “switch back to Claude” moment
1
259
Tarun Sachdeva retweeted
Replying to @tarunsachdeva
did this for our interns, traces has been amazing!
1
1
1
224
hearing this a lot more these days
3
1
31
1,672
I'll be speaking at the AFK coding workshop tomorrow talking about all the value contained in your agent traces. Come check it out! Register here 👇
The agents are all alright. The hard part of AFK coding is writing acceptance criteria sharp enough that "done" means done. Live webinar tomorrow: @anil_d93 kicks off the loop and walks away while it builds, validates, and reports back. Then @tarunsachdeva (Traces) on storing your agent traces: review a run by reading its trace, not the diff, and hand it off without re-explaining. Live demo. 9am PT / 12pm ET / 5pm BST luma.com/afk-programming
1
5
474
Tarun Sachdeva retweeted
A product I find using more and more. Tarun has also been great a helping out with any issues over text. Simple use case right now is sending a session to my co-founder and engineers where I plan out an idea, and then can take it form there and implement it. There are so many other ways to use it though!
we didn't have to make our TUI responsive and pretty, but we did anyways Traces v0.6.1 out now with session resuming, search via cli, and tons of performance improvements
2
1
6
604
we didn't have to make our TUI responsive and pretty, but we did anyways Traces v0.6.1 out now with session resuming, search via cli, and tons of performance improvements
1
2
45
3,706
cool to reflect on this original launch moment sometimes the progress has exceeded all my expectations and this is easily the most interesting products I’ve ever shaped over the next few weeks we’ll be rolling out one of the most consequential updates since launch. can’t wait
Introducing Traces A new way to share and discover traces from coding agents, and a small step to make AI more multiplayer. Here's how it works:
1
11
931
Tarun Sachdeva retweeted
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)

4
2
17
1,013
Tarun Sachdeva retweeted
announcing Falcon GX the world's most powerful brand engineering tool now in public beta
49
56
557
100,653
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)

4
2
17
1,013

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
4
308
Tarun Sachdeva retweeted
what if you could see how many people downloaded your ai prompts now available on traces.com profile pages
3
6
24
2,447
Tarun Sachdeva retweeted
nice
what if you could see how many people downloaded your ai prompts now available on traces.com profile pages
1
1
2
1,024
Tarun Sachdeva retweeted
Homecoming 🇹🇩📾
11
9
199
21,442
what if you could see how many people downloaded your ai prompts now available on traces.com profile pages
3
6
24
2,447
Tarun Sachdeva retweeted
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.
12
75
651
64,028