@patrickdevivo

software engineer in NYC. prev https://xitter.cf/t.co/Bx70AilrPF and https://xitter.cf/t.co/ZTUpgwSh6a. Now @powersetres.

 New York
Joined November 2007
  • Tweets991
  • Following1.2K
  • Followers875
65Media
on @PowersetRes we run an LLM embedding pipeline to categorize repos based on their READMEs and website content. we use embedding similarity to a category centroid to determine membership- it works, but seems like a place Jev could really shine could simplify a whole lot and probably be faster/cheaper
2
226
Jev-style realtime video question/answering with DiffusionGemma from @mmastrac running on @modal 😳
3
1
1
19
3,616
question json, editable in the web ui
1
131
1
4
234
Patrick DeVivo
@patrickdevivoAug 17
1
209
1
297
Patrick DeVivo retweeted
Tyler Bosmeny
@bosmenyAug 17
12,000 tok / sec on a $250 board is nuts Waiting for LLMs to stream text is going to seem so quaint in a few years...
Lamb Labs (YC S26)@LambLabsAug 16
12,000 tok/s on a 250$ FPGA board. It's a small model, and doesn't answer coherently to any question or prompt really. try it out on trylittlelamb.com/ but don't expect a good chat. at least its VERY fast
12
11
2
176
37,716
I remember interviewing back in 2015 thinking it would be edgy and a way to differentiate myself to answer a question: "how would you implement a photo gallery feature?" truthfully by saying I would search stackoverflow for the most suggested "photo gallery" library and just use that... didn't get the job
sathvik
@vikvang1Aug 13
“How would you design a queue system with Kafka” “Yeah so I would first start my agent and ask it how it would design a queue system with Kafka. Then it will give me a couple options, but one of them is recommended. i then click the recommended option and i should be good”
1
2
377
agents > tickets
Vorflux
@vorfluxaiAug 13
2026, the death of the backlog
6
1,193
faster tok/s will be a phase shift for AI
1
2
108
Patrick DeVivo retweeted
David K 🎹
@DavidKPianoAug 7
I can't believe I used to type this out every time: "git commit -am wip && git push" Now all I have to do is type this: "please commit and push my changes"
113
16
11
486
43,540
Patrick DeVivo retweeted
Prasanna S
@myprasannaAug 6
Going live shortly.
Prasanna S
@myprasannaAug 5
Live soon with @jake_zeller and @patrickdevivo !
3
16
3,306
I'm surprised there's not more hype for @GoogleDeepMind's Gemma 4 It absolutely flies on @cerebras and has been just as good for certain research tasks (and orders of magnitude faster and cheaper) than OpenAI models we were using
1
4
379
switched from Claude Code to Codex CLI today and am experiencing this in real time. I can’t articulate exactly why…it’s just better Less sycophantic? Feels more comprehensive? 5.5 seems to just miss fewer things
Nick Schrock
@schrocknJun 5
Underreported story in AI Eng: Just how much better it is to work with gpt 5.5 than Opus 4.8. It's smarter *and* has a decisive speed advantage. The speed difference also underrated.
1
191
Patrick DeVivo retweeted
Powerset Research
@PowersetResMay 27
The best recruiting signal is often sitting in public GitHub data. Today, we’re open sourcing the data layer we use to help founders recruit exceptional engineers. research.powerset.co/posts/r…
2
5
2
13
19,071
the job of software engineer is so clearly no longer about directly managing code, but managing the systems and rails for AI tools to ship the best possible codel, given the domain and its constraints
1
1
42
e.g. don't spend cycles on frontend code, spend time getting the agent connected to a design system, following best practices, and self-verifying its own code changes. the work is now building and operating the machine that ships and maintains code, not the code itself
1
32
Patrick DeVivo retweeted
Jake Zeller
@jake_zellerApr 29
1/ Applications are now open for Powerset Cohort 4, where top founders get their own $ 1M+ fund for angel investing. Last year, Anton Osika (Lovable) and other elite founders joined Cohort 3. Our founders have deployed $ 20M+ so far and are outperforming top VCs.
10
17
1
68
18,312
I think there are interesting use cases for sandboxes that have nothing to do with AI code execution. If you can isolate tenants and pin them to their own compute, that’s a pretty powerful primitive for multi-tenant software generally. CI/CD systems, per-tenant queries with variable compute/memory, third-party plugin systems, user-defined logic, etc. Cool to see AI infra make infra better for non-AI products too.
2
1
1
5
934
1/ If history is any guide, I think the end state of agentic coding is something far more declarative than imperative (relative to where we are today). We’ve headed far in that direction, but there’s probably more to go…
1
1
86
more replies
6/ I think there’s more to climb on the “declarative-ness” ladder though. Especially when it comes to on-going maintenance and day two development.
1
1
53
7/ Instead of a human prompting specific features and fixes to move a codebase forward, do we end up in an even more declarative/spec-driven world, where human changes are text updates to a CLAUDE.md spec and agents continuously work to reconcile to a final state?
43
Load more