eric provencher (@pvncher) / X

eric provencher

16.7K posts

eric provencher

@pvncher

Codex DX @Openai | built @repoprompt | prev XR @unity

Montreal, Canada

Joined June 2013

Pinned
eric provencher
@pvncher
Jun 5
Happy to announce that @RepoPrompt Community Edition is now live on @github Repo Prompt started as a tool for copy pasting, but soon stumbled into a world of context engineering. Today, it has evolved into an extremely capable multi agent orchestration tool, by inverting
22K
eric provencher
@pvncher
Jun 3, 2025
Just use o3 for everything. It's the highest roi on your time
284K
eric provencher
@pvncher
Oct 29, 2025
This feature is probably amazing if you work at cursor with an unlimited plan
Austin Nick Piel
@austinnickpiel
Oct 29, 2025
Manager: I need you to fix this bug ASAP! Me: No problem, I've got my best guys on it. My best guys:
144K
eric provencher
@pvncher
Aug 23, 2025
Replying to @daniel_mac8
As a developer of an mcp server, it’s a few things 1. It’s got support for all the clients 2. The api is discovered at runtime, so it can change, and the client isn’t hardcoded to it 3. It has features that are underutilized like elicitation, sampling and built in prompts
63K
eric provencher
@pvncher
Feb 14, 2025
This is why I built repo prompt. You can’t just put the whole codebase into the model’s context - more than 32k tokens per prompt often leads to degraded intelligence, even with o1 pro. Being selective is super important, and it’s getting easier since I added codemaps.
155K
eric provencher
@pvncher
Jan 17, 2025
Using o1 Pro so much that I’m starting to feel like a frontier junkie. Like, I cannot use anything but the best model available. Dangerous mindset to have as I watch the frontier’s cost-per-token skyrocket.
107K
eric provencher
@pvncher
Jul 19, 2023
lol to everyone gushing over meta’s llama 2 release, I invite you to actually try it. It’s the most sensitive language model I’ve ever interacted with and if this is the future of models I don’t really want it.
282K
eric provencher
@pvncher
Jul 26, 2025
I find this really funny. Sonnet 4 is a great tool caller, but it’s been long beat by Gemini, o3 and grok 4 on actual code writing. Stop conflating tool use capability with code writing capacity.
Bindu Reddy
@bindureddy
Jul 26, 2025
Hearing rumors that GPT-5 beats Sonnet-4 on coding An extremely big deal, if this is true
206K
eric provencher
@pvncher
Feb 13, 2025
Yesterday, the codemap update for Repo Prompt went live, and it's a game changer for intelligently pulling in context detected in the code you select to be a part of your prompts. This update lays the groundwork for much more automation around prompt building - stay tuned!
00:00
131K
eric provencher
@pvncher
Oct 18, 2024
Replying to @ChristianSelig
Maybe!?
Anshel Sag
@anshelsag
Oct 18, 2024
Wireless charging absolutely has efficiency problems. But this isn't inductive charging which is actually the inferior technology. This is magnetic resonance, it has better efficiency and looser coupling. It lost to Qi (which is inductive) mostly due to cost. Magnetic resonance
182K
eric provencher
@pvncher
Jul 4, 2023
The @unity XR Interaction toolkit v2.4 is out now! The biggest change is a complete revamp of the ray interaction visual logic, it's burst optimized, rays bend to their target and are retractable! Ray jitter is also way down thanks to default stabilization! See the🧵for more
00:00
54K
eric provencher
@pvncher
Jun 10, 2025
o3 now costs the same as 4.1! Cheaper than Sonnet, with perf exceeding Opus. This is by far the best deal in the llm game by a mile! No excuse to not use it heavily in @RepoPrompt now!
Sam Altman
@sama
Jun 10, 2025
we dropped the price of o3 by 80%!! excited to see what people will do with it now. think you'll also be happy with o3-pro pricing for the performance :)
70K
eric provencher
@pvncher
Oct 13, 2025
My take with codex vs claude code - Use codex if you have a detailed plan and want to walk away for 30 min. If you try and interrupt and iterate codex, you might as well restart from scratch - Use claude code if youre not 100% sure where you're going and want to iterate
53K
eric provencher
@pvncher
Sep 29, 2025
I wonder if the folks at Anthropic doing victory laps today have tried Codex, or any GPT-5 model. Sonnet 4.5 is really fast, but it's not smarter by any stretch than even GPT-5 Med.
65K