Peter Potapov 🧙‍♂️

252 posts

Peter Potapov 🧙‍♂️ banner
Peter Potapov 🧙‍♂️

Peter Potapov 🧙‍♂️

@peter_potapov

Head of Technology @ Kick. Ex-founder @ Telestat, ex-Head of Eng/ML @ Osome. Dophamine + AI combined

Palo Alto, California Присоединился Nisan 2022
730 Подписки334 Подписчики
Peter Potapov 🧙‍♂️
Peter Potapov 🧙‍♂️@peter_potapov·
The biggest improvement for us so far has been building a loop-based pattern at the PR level, inspired by the ralph loop: one agent implements and another reviews its work (accompanied by multiple CI/CD checks), while engineers oversee the process. There are still plenty of guardrails to dial in based on each team's setup, strengths and weaknesses, etc., but directionally that's where we're headed. Hint: you don't need a manual review for 95%+ PRs.
English
1
0
2
31
Alex Belov
Alex Belov@justalexagain·
@peter_potapov it’s impressive! could you share more about your workflow? working on my productivity too
English
1
0
0
41
Peter Potapov 🧙‍♂️
Peter Potapov 🧙‍♂️@peter_potapov·
We 2.5x'd our weekly PR volume by deeply integrating AI agents into every coding workflow: discovery, implementation, review, refactoring, etc. Same team. Same hours. Just workflows designed around AI agents as the core, not an addition. The gap between teams that get this and teams that don't is widening fast. And we're on track to double it again next month.
Peter Potapov 🧙‍♂️ tweet media
English
1
0
1
65
Peter Potapov 🧙‍♂️
Peter Potapov 🧙‍♂️@peter_potapov·
I fully agree and that's why I don't understand all this X hype around "Cursor is dead". There's still no better IDE alternative for the agentic experience. Claude Code's UI version isn't even close.
Andrej Karpathy@karpathy

Expectation: the age of the IDE is over Reality: we’re going to need a bigger IDE (imo). It just looks very different because humans now move upwards and program at a higher level - the basic unit of interest is not one file but one agent. It’s still programming.

English
2
0
2
170
Peter Potapov 🧙‍♂️
Peter Potapov 🧙‍♂️@peter_potapov·
@_chenglou Love “verifiable software” term and also having same thoughts on future software commoditization
English
0
0
0
110
Cheng Lou
Cheng Lou@_chenglou·
The cost of any verifiable software will trend toward 0
English
15
1
93
11.1K
Peter Potapov 🧙‍♂️
Peter Potapov 🧙‍♂️@peter_potapov·
The main source of bugs in most vibe-coded products, e.g. Claude for web, isn't AI code generation itself - it's probably the underlying culture of shipping without QA and the fact that reliable automated E2E testing tools still aren't there yet
English
2
0
1
94
Peter Potapov 🧙‍♂️
Peter Potapov 🧙‍♂️@peter_potapov·
Starting to think those conspiracy theories about AI companies quietly degrading models in production to cut costs might actually be true… Something is definitely off with Opus today - first time it's felt this dumb, had to switch to GPT-5.4 instead
English
1
0
1
101
Daniel Dhawan
Daniel Dhawan@daniel_dhawan·
The only thing that matters in software is the user experience. In 2011, Ryan Dahl, the creator of Node.js, wrote: I hate almost all of the software. It's unnecessary and complicated at almost every layer. At best I can congratulate someone for quickly and simply solving a problem on top of the shit that they are given. Who knew that in 2026 it would become evident to everyone
English
7
0
25
2K
Peter Potapov 🧙‍♂️ ретвитнул
Nabeel S. Qureshi
Nabeel S. Qureshi@nabeelqu·
If you'd shown Claude Code or Codex on 5.4xhigh to any reasonable person in 2020 they'd have concluded it was AGI
English
163
136
5.5K
220.3K
Peter Potapov 🧙‍♂️
Peter Potapov 🧙‍♂️@peter_potapov·
Damn, there’s so much interesting stuff happening in the world and I only have 24 hours in a day… never been more excited about the world! (2:03am and I still can’t go to sleep)
English
1
0
2
47
Peter Potapov 🧙‍♂️
Peter Potapov 🧙‍♂️@peter_potapov·
We’re already living inside the singularity and most people don’t even realize it… wild!
Andrej Karpathy@karpathy

@tobi Who knew early singularity could be this fun? :) I just confirmed that the improvements autoresearch found over the last 2 days of (~650) experiments on depth 12 model transfer well to depth 24 so nanochat is about to get a new leaderboard entry for “time to GPT-2” too. Works 🤷‍♂️

English
0
0
0
69
Peter Potapov 🧙‍♂️
Peter Potapov 🧙‍♂️@peter_potapov·
Like if you're also grinding in the Bay Area tonight 🌃 (10:27 PM and we’re still in the office)
English
2
0
5
103
Wes Bos
Wes Bos@wesbos·
I ran the numbers on my day yesterday: 10 hour day of coding, no meetings. It was $118 worth of tokens. In one day. I’ve heard of many devs burning through $350 worth in a day. And this is at subsidized prices
Bearly AI@bearlyai

Cursor internal analysis shows how hard Anthropic is subsidizing Claude Code. Last year, a $200 monthly subscription could use $2,000 in compute. Now, the same $200 monthly plan can consume $5,000 in compute (2.5x increase).

English
52
19
290
89K
Peter Potapov 🧙‍♂️
Peter Potapov 🧙‍♂️@peter_potapov·
To me, Cursor is like iOS and Claude Code is like Android. Claude Code is always faster to ship new agentic capabilities (especially in beta), but Cursor delivers a more polished version of the most useful agentic patterns - so the overall UX ends up being better. The gap is thin, though and, worth noting, I'm always using Cursor with Opus Fast model.
English
0
0
1
465
Jamon
Jamon@jamonholmgren·
Cursor is a better DX, if you care about code quality. Claude Code is better if you don’t.
John Loeber 🎢@johnloeber

On Cursor vs Claude Code Fascinating dynamic: engineers are cancelling their Cursor subscriptions and switching to Claude Code/Codex -- but Cursor continues to grow revenue as it's gaining totally new, first-time-AI-user customers. Is Cursor screwed in this dynamic? Does everyone end up switching to Claude Code/Codex in the long term? I used to think so, but I no longer do. Claude Code isn't winning because of Claude: the same models are available in Cursor. The most significant thing about the launch of Claude Code around this time last year was that it presented a different, slightly contrarian interaction pattern for engineers: getting back into the terminal. It turns out, this just works much better than using a VSCode-style IDE. Competition here isn't really about coding models, it's about interaction patterns! The models will continue to compete tightly and be generally available. All the differentiation is in Developer Experience + Brand. Therefore, the key question is whether new, better interaction patterns are possible. Another way to frame it: is the Claude Code-style UI the best possible UI/UX for this task? I don't think so. It's pretty good, but it's not the end of innovation. And you might say that it should be hard for a new interaction model to take market share -- but that's exactly what Claude Code did. I hear that Cursor is going all-in on building their own coding models, but I doubt that's the right move. To me, the big room is in interaction patterns: building better harnesses that provide a better developer experience. I'd expect to see some competition here over the next year or so. Perhaps we'll see another competitive leapfrog, like how Claude Code did in March 2025.

English
52
14
238
53.2K
Peter Potapov 🧙‍♂️
Peter Potapov 🧙‍♂️@peter_potapov·
@levan It locks users into personal accounts instead of converting them to enterprise, which is where the real NRR lives
English
1
0
0
30
Levan Kvirkvelia
Levan Kvirkvelia@levan·
This is not subsidization; they sell it at zero markup, cursor sell it at zero markup, but there is a nuance.
Tanay Jaipuria@tanayj

Good piece on the "war time" at Cursor. Some interesting quotes: - The company’s new mandate was labeled “P0 #1”—priority zero: “Build the best coding model.” - Cursor estimated last year that a $200-per-month Claude Code subscription could use up to $2,000 in compute, suggesting significant subsidization by Anthropic. Today, that subsidization appears to be even more aggressive, with that $200 plan able to consume about $5,000 in compute, according to a different person who has seen analyses on the company’s compute spend patterns. forbes.com/sites/annatong…

English
1
0
5
1.3K
Peter Potapov 🧙‍♂️
Peter Potapov 🧙‍♂️@peter_potapov·
We found this internally a few days ago and I still can't get over how massive this subsidy is. This is what a winner-take-all race looks like. Curious how much market share Anthropic would capture without this kind of backing. What do you think?
Bearly AI@bearlyai

Cursor internal analysis shows how hard Anthropic is subsidizing Claude Code. Last year, a $200 monthly subscription could use $2,000 in compute. Now, the same $200 monthly plan can consume $5,000 in compute (2.5x increase).

English
0
0
0
92