Natansh Verma (@natansh) header
Natansh Verma (@natansh) profile picture

I make things. Sometimes that requires writing code.

San Francisco, CA
Joined May 2008
Photo by Natansh Verma (@natansh)Photo by Natansh Verma (@natansh)Photo by Natansh Verma (@natansh)Photo by Natansh Verma (@natansh)Photo by Natansh Verma (@natansh)Photo by Natansh Verma (@natansh)Photo by Natansh Verma (@natansh)Photo by Natansh Verma (@natansh)Photo by Natansh Verma (@natansh)
Was able to tell an agent how to use SwiftUI correctly. I guess I am still useful some days.
29
Powerful idea! Been asking AI to prove bugs by writing failing Unit/E2E tests myself.
🧠 Most AI bug-catching tools review your code the way a human reviewer would, only faster. What if the AI wrote a test that proved the bug instead? That’s Catching Just-in-Time Tests (JiTTests) — a novel approach where LLMs generate executable tests designed to fail, surfacing behavioral changes in a code change before it lands. After a year of research, it’s now part of Meta’s infrastructure. ✅ 589 engineers have engaged with it, reporting a 60% true-positive rate 📈 In the first half of 2026, it helped catch 74 privacy SEVs at the diff stage, lowering overall privacy SEV occurrence by 20% 🔗 Read how it works: engineering.fb.com/2026/10/2… 📄 Read our FSE 2026 paper, “Just-in-Time Catching Test Generation at Meta”: arxiv.org/pdf/2601.22832 #AI #SoftwareEngineering #SoftwareTesting #DeveloperTools #Engineering
1
47
The best way I have found to detect an architecture going sideways for AI is when a conceptually simple idea results in a complex implementation.
23
I had to get an agent to keep on aggressively deleting old build artifacts while working with Rust, sometimes as frequently as 20 minutes.
You know that your agents are programming Rust when your 2TB drive runs out free space.
23
One of the reasons I think Critical Thinking is slipping is because agents still are very slow, latency wise, so a full send helps get more throughput. The dream is to be able to have fast, totally in the flow multi-modal conversations with agents as a thought partner, really dig into the problem space, and then commit to a task being completed.
1
27
Happy “rewind all the wall clocks manually” day.
41
Many people mention core functionality missing, but to me, DeviceHub simply feels unwanted, in an agentic era. I want to be able to run many simulators visible simultaneously in a grid, efficiently, with easy understanding of what agents are doing - and those feel like problems the community is hacking and solving, rather than Apple.
After a few months of giving it a chance, Device Hub really feels like a step backward
44
As of October end Claude Opus 5.5 > GPT 6.1 Sol Codex Desktop > Claude Desktop
2
67
I'm literally hitting system limits while pushing Claude, enough that sometimes running Claude itself causes my performance measurements for another app to not be scientific anymore.
18
Natansh Verma retweeted
one day you wrote console.log("here") for the last time and didn't even know it
213
1,724
269
17,857
610,930
Very excited to see progress on visual thinking models!
Spent years watching models ace a single image and completely whiff on anything that moves or changes angle. Today we’re fixing that. Try Glimpse now, our first visual thinking model with competitive frontier performance compared to Astra on key public visual reasoning benchmarks.
1
101
“Do work that you’ll be proud of, I trust you.” is my new favorite way of working with agents. I’ve noticed agents check and critique their work a bit more when I say that.
27
Fills me with joy to see this collaboration!
Today we’re releasing Muse Spark Open, the most capable US open-weight model to date. Built for coding and agentic work, it excels at long-horizon tasks, complex software engineering, and multi-step reasoning. Key Model Specs: → 816B total / 42B active parameters → 1M token context window → Natively multimodal across text, image, video, audio
1
75
Almost every part of the software stack is ripe for reconsideration. We have to update our priors.
New on our Engineering blog: Creating in-house source control. In May, we merged around 2,500 code changes per day. In September, we merged more than 30,000 per day. As a result, we became bottlenecked by developer infrastructure built for human (rather than agent) scale. Here, we share how we built a new version control system—and fully migrated our monorepo from git—in five months: anthropic.com/engineering/re…
25
Finally!
We are making Markdown files work natively across Google Drive and Docs. You can now preview .md files in Drive, open them in Docs, edit them, comment on them, and collaborate without converting the file into a Google Doc!
25
ASD-STE 100 works well, but there are more ways to control - 1. ELI5 - Explain Like I’m 5 - Works a treat if you want to clearly understand the outcome. 2. Use principles of “On Writing Well”, automatically primes AI to write far better prose. Mix and match.
We'll be spending a lot more time trying to understand the outputs of language models. A few thoughts, tips & tricks: Writing. Something I've had success with: Ask your LLM to explain something in ASD-STE100, it's a controlled language specification originally developed for aerospace maintenance documentation. LLMs well-versed in this language and it comes with heavy constraints on clean writing style that I often find a lot more readable. Sometimes I've tried to soften it a bit e.g. ask for "80% of the way to ASD-STE100" because the spec is quite stringent. But even better: Diagrams / images. Instead of writing, ask your LLM to create a diagram. These can be a lot easier to process, parse, and understand. But even better: Web pages. Ask for output "in HTML" to get a beautiful, interactive webpage. LLMs are getting really good at frontend and can create beautiful experiences, animations, etc. But even better: Explainer videos. The output format I am most bullish on is fully custom / bespoke explainer videos generated on any arbitrary topic. Experiment with things like "Create a 3b1b style video explainer on X. Use my ElevenLabs API key for audio narration". (you'd need an API key for the latter or you can ask your LLM to find you decent free alternatives that use your local compute). This is actually starting to work! In summary: - As LLMs get better, they will do more and more of the legwork autonomously, and a lot more of our work will rise up the abstractions into oversight and understanding. - Luckily, LLMs can help here too because as intelligence and code are increasingly abundant, you can ask for large, custom, discardable software artifacts (e.g. web apps, video explainers) that would have never made sense to create before. Push the boundaries here and you'll be surprised.
53
This is amazing!
Google: “[Swift] has matured into a viable systems and cloud language, pairing Rust-like data-race safety with predictable, reference-counted performance.” Announcing their Swift-based cloud SDK for backend services: cloud.google.com/blog/topics…
38
Natansh Verma retweeted
Started doing this. Very useful for when you're dictating out a bunch of stuff and you can only half-remember what you've said. I prefer a 5-word version: Restate my intent before continuing
this has become one of my most used prompts recently: > restate in your own words what you think my goals are and what the problem i'm trying to solve is
63
123
17
2,756
219,992
The only thing to use comments for are details or context that cannot be obviously derived from code.
People talk about the diminishing returns of writing unit tests in the agentic era, but what’s even less useful? Comments. As a rule, drop all comments. If something is important enough to need a comment, make an exception for it and audit the comment so it actually makes sense.
52
Not gonna miss this one! It’s kind of a relic at this point.
we’re thinking of killing plan mode and using the shift+tab hotkey to adjust effort levels I don’t think the models need plan mode anymore, but if you’re a plan mode diehard would love to get your feedback on why
1
1
53
Should you plan? Yes. Do you need plan mode baked into Claude? No.
1
24