Verification is the only bottle neck! If you can make AI verify its generated stuff on something quantifiable and clear proof - its over!
Generation systems solved!
Verification systems for everything are being builtb
Chopin + Proof-Driven Development Kit from @GitHubNext and @MSFTResearch enable teams to ship AI-generated code to production without reviewing it line by line, with stronger assurance than code review and testing alone.
Now in research preview ⬇️
Hey @bot what are the most valuable "Skills" that can be used with Claude, Codex, and any coding agent?
This is what I call "machine scale software verification/testing".
This is what a software factory needs. Deliberately invest in your "verification/testing stack" built specifically for agents.
With the amount of code being written, machine scale verification is the need of the hour!
Give your agent a way to verify its work. This can dramatically improve the quality of output.
With this process in place, Claude Code was able to write 90%+ of its code itself. When generating software at a faster rate than you can review it, the bottleneck shifts to proving that it works. @bcherny and team invested enough in verification, we've been able to move fast while maintaining stability.
Code reviews aren't going away, but you need to think of it differently. One human reading through line-by-line is no longer going to work as a gate, it needs to evolve into a multi-layered verification system, built with fast deterministic feedback loops for your agents.
This is where e2e tests, feature maps (as @poteto has been advocating), property tests and multi-agent adversarial reviews come in. These all provide opportunities to catch errors early and fix them before things get out of hand.
You don't have enough eyeballs for the amount of code being written so your focus needs to be on product intent, risk assessment and evaluating evidence rather than reviewing code. Delegate the code checking work to your automated tests, scale your approvals based on risk, and turn your test failures into living evaluations. This approach helps maintain speed without losing reliability.
Let the agents do what they do best and let you focus on what makes you irreplaceable: judgement, direction, and ownership. We tried to capture how we handle code review and verification in this writeup. Hopefully it's helpful.
Neeraj Kumar retweeted
Give your agent a way to verify its work. This can dramatically improve the quality of output.
With this process in place, Claude Code was able to write 90%+ of its code itself. When generating software at a faster rate than you can review it, the bottleneck shifts to proving that it works. @bcherny and team invested enough in verification, we've been able to move fast while maintaining stability.
Code reviews aren't going away, but you need to think of it differently. One human reading through line-by-line is no longer going to work as a gate, it needs to evolve into a multi-layered verification system, built with fast deterministic feedback loops for your agents.
This is where e2e tests, feature maps (as @poteto has been advocating), property tests and multi-agent adversarial reviews come in. These all provide opportunities to catch errors early and fix them before things get out of hand.
You don't have enough eyeballs for the amount of code being written so your focus needs to be on product intent, risk assessment and evaluating evidence rather than reviewing code. Delegate the code checking work to your automated tests, scale your approvals based on risk, and turn your test failures into living evaluations. This approach helps maintain speed without losing reliability.
Let the agents do what they do best and let you focus on what makes you irreplaceable: judgement, direction, and ownership. We tried to capture how we handle code review and verification in this writeup. Hopefully it's helpful.
Claude writes about 90% of Claude Code's code.
From June to September, merged PRs more than doubled while the share reverted fell by more than 40%.
We wrote about how we changed code review to get there, and how to adopt it:
claude.dev/blog/solving-code…
. @cursor_ai I am facing this issue : Error: Server error
The server failed 4 times. Please try again, or contact support if the issue persists.
What to do?
Is @cursor_ai having some issue? Always getting this
"Reconnecting to agentn.global.api5.cursor.sh (attempt 3, 4s)" in the cli!!
I can't remember the last time product building felt this much like play. LLMs have hit escape velocity, especially with Opus 5.5 the quality-to-cost ratio finally makes sense for high quality ambitious work.
- Offloading mundane ops work to Agents so I can focus on the actual product work
- 10X more ideas on prod in 1/10th of the time!
- With Agents, Product usage is exploding and growing at rates I have rarely seen!
- So many new things to learn!
What fun thing you are exploring with with AI right now? 👇
Neeraj Kumar retweeted
LLM makes building so much fun! Built voice-to-text device with a ESP32 microcontroller with NVIDIA's Parakeet transcription model
The setup:
- An ESP32-C6 board (PCB Cupid GLYPH) with a microphone and a battery
- Flashed firmware code onto it
- Sends the audio over Wi-Fi to my Mac
- NVIDIA's Parakeet speech model runs locally on the Mac and turns the audio into text. No data ever leaves the computer
Thanks to @GrowthX_Club and #PCBCupid for the push!
youtube.com/shorts/l9fsYQ34i…
Does @HermesWatcher has a mobile app or am i missing something?
Just downloaded desktop app and i am not sure what to do?
Great article on how they designed evals for non software field like contract review.
if you are a serious AI first company, nvest in your evals system, go in debt if you have to!
Once generation of any digital artefact is solved (which it is).... judgement & evaluation will be the only thing left!
crosby.ai/blog/how-we-do-eva…
andrewchen.substack.com/p/wi…
Love this article from @andrewchen on network effects in the AI agents era!
LLM makes building so much fun! Built voice-to-text device with a ESP32 microcontroller with NVIDIA's Parakeet transcription model
The setup:
- An ESP32-C6 board (PCB Cupid GLYPH) with a microphone and a battery
- Flashed firmware code onto it
- Sends the audio over Wi-Fi to my Mac
- NVIDIA's Parakeet speech model runs locally on the Mac and turns the audio into text. No data ever leaves the computer
Thanks to @GrowthX_Club and #PCBCupid for the push!
youtube.com/shorts/l9fsYQ34i…
Neeraj Kumar retweeted
. @claudeai Code mod is such a nice feature! Built 2 customisations (token view & loading stats as per my mood and weather)
code.claude.com/docs/en/plug…
. @claudeai Code mod is such a nice feature! Built 2 customisations (token view & loading stats as per my mood and weather)
code.claude.com/docs/en/plug…
. @claudeai Tokenweather mod doing its job
