Suhail Doshi

tweet shall self-destruct: https://keymake.ai/ come here and try it

hype

Perhaps the most addicting quality about AI coding is that it starts to feel like spinning plates at different speeds, and since you’re already waiting, you might as well toss one more into the air. So much so, it never seems to end until it's 2 AM.

We seem close to: - Give an agent access to a competitor app on a computer - Tell agent: Rebuild this app by using all its features - Agent tries app -> documents all flows/features/edge cases - The other agent builds all flows/features - They iterate trying/testing until done

Engineering is now this: new agent -> shift+tab -> wispr plan -> wait -> look at X -> review plan -> make adjustments -> approve/build -> verify it works -> integrate tests -> merge -> go to app

Verification: ac-a64dfe1c95e9

Become a polymath entrepreneur or suffer the consequences.

Last night I looked at my GitHub contribution graph for fun and was surprised “huh, how can I have only done 7 PRs all day?” and then I realized it was 1:30 am the next day. This is our new life isn’t it?

why did everyone have to take all the good domains?

AI agents running computers in the cloud that you can watch in real time. What a ridiculous idea!

Seat based pricing now just screams: “automate”

RT Aravind Srinivas What has Perplexity been up to last two months? We've silently been working on the next big thing: Perplexity Computer. Computer unifies every current capability of AI into a single system. Files, tools, memory, and models, orchestrated together, working for you. Original tweet: https://x.com/AravSrinivas/status/2026695864039911684

Have entered my making unhinged software era because why the hell not at this stage of takeoff.

I am in big trouble aren't I?

Make the margins next to zero for all these AI models. It was trained on humanity's data, it should be gift to ourselves. Doing so will save us from a few in control of our species. Distill at industrial scale! Distill, I say!

The average HN comment. Always at the top.A few good ones left

Ahahaha - I love it

I think a lot of doing something ambitious is figuring out how to wake up tomorrow to just keep going when today was tough or disappointing. It's a mental game since what you need to keep going is often just around the corner.

HN is insanely jealous of Peter. They are either shocked that a guy's weekend project potentially made him generational wealth or that it landed him a job they couldn't get. Of course, they have no idea. HN is a slow aging elder class of tech.

Had no idea OpenAI was doing billboards for this already.

Re

This and next year we will see some of the largest companies on earth pivot in surprising ways. Everything that seemed safe is unsafe. You either strike yourself first or you will get struck.

slowly but surely…

RT Ankur Goyal We sent this note to our customers to let them know that Braintrust has raised a new round of funding, and thank them for their support. While the money is exciting, our focus hasn't changed: we're building Braintrust to help our customers ship quality AI products. In 2026, AI is moving to production but teams have never had less conviction about what will fail next. Our customers are building AI products that serve millions and simply need to work. If Braintrust makes their lives easier and their products better, I know we are doing our job. Thank you to @ICONIQCapital for leading our Series B, and to @a16z, @GreylockVC, @basecasevc, and @eladgil for doubling down. Thank you to the Braintrust team for all the incredible work you've done over the past year. And thank you to our customers, who have made this growth possible. Original tweet: https://x.com/ankrgyl/status/2023810273598128588

Confusing even though I load unpacked the chrome extensionIf you provide your open ai api key, why won’t it use the native web search tool part of the api?

I feel like AI is using me as automation.

The funny thing is that if the product is excellent, this would’ve been too polished on day one. However if the product sucks, no amount of differentiation on branding will save it.

RT Amanda Askell WSJ did a profile of me. A lot of the response has been people trying to infer my personal political views. For what it's worth, I try to treat my personal political views as a potential source of bias and not as something it would be appropriate to try to train models to adopt. Original tweet: https://x.com/AmandaAskell/status/2022778351744581779

oh the memories of this road - iykyk

A reversion to being hardcore and ambitious again. 🎯

sigh

Been reading Garry's posts on this for a short while. They're so good if you care about San Francisco. If you care about California.

Making a lot of music lately has reminded me of the tremendous value of trial and error to discover new things. I think to the degree to which can try more, the faster we will learn. Soon, I think we will be able to turn many problems into exploration problems with compute.

It seems like the number of things people claim they are building is up but the number of great products is roughly the same. I find this particularly interesting about what it means is actually hard.

🚀

Easily some of my happiest memories from startups are from this.

RT @mashadrokova: Growing up in post-Soviet Russia, I joined the regime's youth movement at 15, believing it was the path forward. By my la…

RT Garry Tan Boil the Oceans You know the phrase: “don’t boil the ocean.” Everyone’s said it in some overly ambitious meeting. It’s good advice in normal times. It keeps teams focused. It prevents scope creep. But we are no longer in normal times, and I think it’s time to retire saying it. Artificial Superintelligence means it’s time to boil the ocean. We’ll start with a few lakes first. I was recently with a university endowment’s head of private investing who told me their engineers were terrified for their jobs after seeing what Claude Code could do. And I get it — that’s the natural first reaction. But it’s the wrong one. It’s a zero-sum reaction to a positive-sum moment. Instead of worrying about doing the same thing we’ve been doing for cheaper, why not focus on doing the thing we never even dreamed of doing? Why can’t that endowment achieve 50% net IRR instead of 10%? Why can’t a startup deliver a service that is 100x better than the incumbent? Why can’t we have fusion energy? Why can’t we talk to every single user and have a perfect understanding of every bug in our product? These aren’t rhetorical questions anymore. They’re engineering problems with paths to solutions. Here is what I think is actually going on with the fear: our fear of the future is directly proportional to how small our ambitions are. If your plan is to keep doing exactly what you’re doing, then yes, a machine that can do it faster and cheaper is terrifying. But if your plan is to do something dramatically bigger, then the machine is the best news you’ve ever gotten. If you’re a worker — someone who trades labor for a living — this is the moment to become a builder. Start a business. And if you’re already management or capital, it’s time to go 10x more hardcore on what your aspirations could be. Not eking out 5% efficiency gains. Not increasing profit margins 2% by lowering cost and firing people. Those are the old games. The new question is: what would it look like to build a prod...

The Great Compression is upon us. From an explosion of software to a re-consolidation back due to AI.

post agi type music

Touched grass tonight and god damn it felt good.

The rate of your growth in an area is directly proportional to how cringe you view your recent progress once shipped / released / posted.

We are still in the era of 4MB of memory on DOS. Amazing how a weekend of tokens, is this expensive. In time, it’ll be too cheap to meter.

The past two decades of product analytics and data science is over. A new age has begun. Just dump all your data somewhere and ask questions:

RT Mehul Announcing $60M for @maticrobots. We didn't ask "What's the most impressive robot we can demo?" We asked "What's the most useful robot we can ship? What comes after Roomba?" Customers answered with their wallets: It's Matic. Original tweet: https://x.com/mehul/status/2016936862716448873

RT Alex L Zhang We just updated the RLM paper with some new stuff. First, we just released RLM-Qwen3-8B, the first natively recursive language model (at tiny scale!). We post-trained Qwen3-8B using only ~1000 RLM trajectories from unrelated domains to our evaluation benchmarks. RLM-Qwen3-8B works well across several tasks and delivers a pretty large boost over using an RLM scaffold with the underlying Qwen3-8B model off-the-shelf, and even larger gains over directly using Qwen3-8B directly for long-context problems. Original tweet: https://x.com/a1zhang/status/2016923294461476873

Epic post on "Vitamin D & Omega-3 have a larger effect on depression than antidepressants" Curious what @bryan_johnson thinks!

Congrats! @bridgitmendler is a real one.

remember when all we wanted ai to do was finally read those crazy math symbols in papers and turn it into code? I was happy in the moment.

Much of AI is rate limited by TSMCs speed to build and manage the risk of building fabs.

All knowledge work problems become compute problems. The next few years will require an intense rewiring of what we thought mattered to be successful. As the AI systems get smarter, they can more efficiently brute force the solution.

algo is so cooked. god damn.

Much of AI is entirely limited by the calculated risk of building fabs. It’s still very early. Great analysis by Ben @ Stratechery.

hey everyone! i am working on new music and am trying an experiment to release a new song every week (~friday) this year. i call the project 'fiftytwo' for the 52 weeks in a year on some level, i know it's going to be hard because the reality is you're wondering if anyone even is listening/noticing and it requires fighting that feeling as you progress—even on the meh weeks for me, it's about putting the reps in to become a better music artist / producer. i know most of what i make might be a dud but i want to make making music less of a precious thing i hold on to tightly to and just release. for others, i hope i make song you love. if you do, please lmk - it means a lot.

1/ Here are my 5 tips on using claude code:

RT philip lewis Ye, fka Kanye West, takes out a full-page in the Wall Street Journal to apologize to the Black community, and for antisemitism: “I lost touch with reality” Original tweet: https://x.com/Phil_Lewis_/status/2015795684901810395

RT Garry Tan Pretty cool. HN understands exactly what we were going for when we redesigned http://ycombinator.com There is no place in the world quite like YC. None. Original tweet: https://x.com/garrytan/status/2014920182175826168

RT Richard Sutton Yann is right about everything (except RL). Original tweet: https://x.com/RichardSSutton/status/2014875521650475416

on my post-agi life. made this song earlier this week and just wanted to share it because i feel like it's an uplifting song for anyone going through a hard time in life 🔊 'lift me through'

RT Odyssey Introducing Odyssey-2 Pro—a frontier world model that generates long-running, interactive simulations in 720p! We're also launching the first world model API, to enable devs to build magical apps. We're now in the GPT-2 era of world models. Let the explosion of apps commence! Original tweet: https://x.com/odysseyml/status/2014780522116091927

i just want to say that @rabois is most the quotable in clips because he's just so damn based that you will always listen.

RT Sebastian Raschka If you are looking for reading material for the upcoming weekend, (the long-promised) Chapter 5 on LLM self-refinement is now finally out in the early access. Here, we continue the inference-time scaling theme, but we move beyond self-consistency and voting. More specifically, we implement a self-refinement loop, where a model iteratively critiques and improves its own answers. Along the way, the chapter implements several core pieces that will become essential in the upcoming reinforcement learning chapters like log-probability scoring (I finished the RL chapter last week, and it will hopefully be out soon, too). Like always, all of this is implemented from scratch, step by step, of course. I think that seeing it all in working code really helps with understanding how LLM reasoning methods work (versus just looking at the equations). Anyways, with Chapter 5 out, the early-access version of the book has grown quite a bit (~300 pages) and now finally wraps up inference-time reasoning before we transition into the (even more) fun part: reinforcement learning (from scratch!) in the next chapter. 🔗 Here's the link to the book's early access: https://mng.bz/dWGv Happy reading (PS: Sorry for the delay. It turns out the submitted manuscript got a bit backlogged in the publisher’s processing pipeline. The silver lining is that the reinforcement learning chapter hopefully now follows sooner! ) Original tweet: https://x.com/rasbt/status/2014341187008602162

RT Gaurav Ahuja Today we continue the 100 Year Conversation with Max Levchin (@mlevchin), co-founder of PayPal and CEO of @Affirm. A project to capture timeless wisdom from leaders who have built institutions that endure. The central question: How do you build something that lasts 100 years? Original tweet: https://x.com/GauravAhujaK/status/2014075488353001759

It was never about “coding”, it was about solving the unsolved. Understanding how things work. Making it indistinguishable from magic.

.

PSA: It seems like the published algo for X doesn't include any of the weights for the final score so there's no way to really know what actually matters. I get that they'd do this to avoid gaming or because they're constantly subtly tweaking them. Anyway, I got tricked by some people who claimed to know on X but maybe they got tricked by the hallucinating AI slop that summarized it (this happened to me until I pushed the model to prove it). If I am wrong, please let me know where the numeric weights are. I couldn't find any.

Story: A long time ago I met @FrakThePerson in this app called Clubhouse during COVID. People were doing freestyle rapping over the Internet which was crazy. So, I decided to start making beats for people in the room, play it on my phone, people would rap on it. That's where Frak and I became friends. Later on, he took a beat from that moment, and put it on his mixtape years later (coming out soon!). Then I heard a bunch of songs on the mixtape and loved this one with @G_Eazy (who I've been a fan of for a while) and ended up doing this remix which came together really fast. It's easy with these two guys rapping to feel inspired. Enjoy!

RT Isaiah Taylor - making nuclear reactors We have over 30 jobs listed on http://ValarAtomics.com/jobs. Come join the hardest and most important mission in America. Original tweet: https://x.com/isaiah_p_taylor/status/2013462356370587842

Yes

RT Ryan Dahl This has been said a thousand times before, but allow me to add my own voice: the era of humans writing code is over. Disturbing for those of us who identify as SWEs, but no less true. That's not to say SWEs don't have work to do, but writing syntax directly is not it. Original tweet: https://x.com/rough__sea/status/2013280952370573666

RT Daniel Lurie 丹尼爾·羅偉 This week, I announced our plan to combine the Planning Department, the Department of Building Inspection, and the Permit Center into one entity. This merger will mean better coordination across departments, saving small businesses and homeowners time and money. These three departments all handle permitting, and bringing them together into a single entity will create a more predictable process—making it easier to build housing and expand businesses in our city. Original tweet: https://x.com/DanielLurie/status/2013039591645741538

RT Ehsan Akhgari From a private account: “I just saw one of my friends. He said his cousin's husband was on his way to the supermarket when a car pulled up next to him, the window rolled down, and someone said "God bless your effort." By the time he went to respond, he saw the guy pulling out a gun. He raised his hand in front of his face and got shot in the hand. In other words, these criminals even fired at people who hadn't gone to the protests” This essentially means: target selection, scene control, deliberate decision This behavior has a name: Terror Policing. Its goal is neither dispersing the crowd nor containing the gathering But rather, in reality, the destruction of the sense of "safety in everyday life" In security literature, it's called: Collective Punishment through Randomization In fact, the randomness of the shootings isn't random at all. Because: there is no pattern, no rules exist, no safe path remains. The final message is that don't protest, even ordinary life is no longer guaranteed safe, and the ultimate result is the psychological paralysis of society. Original tweet: https://x.com/ehsanakhgari/status/2012719347223990774

RT Max Levchin This spells it out plainly. Taxing unrealized gains/“paper wealth” will decimate California’s future tax base: early employees and founders of tech startups. Why would you start a company in CA if the wealth tax man can dictate when you must sell it?! You wouldn’t. Original tweet: https://x.com/mlevchin/status/2012334979963928738

RT Naval If you are the founder of a highly valued illiquid startup, set some time aside and read this entire article. The act is designed to bankrupt you and to personally punish any appraiser or accountant who disagrees. Original tweet: https://x.com/naval/status/2012046918495715335

I am often surprised by how greedy the agents are to cornering a local maxima solution. That'll put it on the top quartile but almost never in the top 5%. It takes a monumental amount of nudging, enforcement, tweaking to make it do novel things that stay on trajectory.

do you use claude in an IDE or terminal?

It is a testament to how complex mass mfg frontier chips (TSMC) is that multiple $4T companies are competing for prioritization and don't vertically integrate that part of the process. Worthy of study to a great degree I think.

The best thing that ever happened to Anthropic was not getting to ChatGPT first. It forced total unrelenting focus on Enterprise.

are you an npc?

RT Frak Going bar for bar with @G_Eazy Heat Check Original tweet: https://x.com/FrakThePerson/status/2010793368264327567

🔊 new song: round n round ♥️

The next frontier is breaking cynicism of this belief: "many people are interpreting this as evidence that LLMs are now creative, that they can genuinely produce something new without human feedback."

Not liking what you made is not failure. It’s the cost of doing reps at your current level.

Very impressive @jgebbia

RT Adam D'Angelo i can't believe we used to write all this code by hand Original tweet: https://x.com/adamdangelo/status/2008961539836186773

RT Paul Buchheit Delete the process Raise a Series-A on your terms The deadline to apply is January 7th, 2026 at 9p.m. PST. Standard will get back to on-time applicants by January 16th, 2026 @Standard_Cap Original tweet: https://x.com/paultoo/status/2008668330446700784

RT Blake Byers In 2025 we invented and our first anti-aging medicine -- makes your liver a decade or two younger (best guess today). We'll spend another couple years optimizing and then launch human clinical studies. In parallel, we are working on anti-aging medicines for the immune and vascular system. In a few years, we will we working on many more tissue types. This is all happening much faster than we anticipated when we started @newlimit. Original tweet: https://x.com/byersblake/status/2008657713178509701

Programming was never about learning how to write if-then statements. It was about building cool things and understanding how they work. Neither of which modern LLMs take away from you. Even if they build everything, you will still need to understand to direct them.

RT Apoorva Mehta if you have started a company in the last 12 months and it's not top 1%, now may be a good time to reconsider everything. claude code + opus 4.5 has changed what's possible and it'd suck to get the timing wrong by just a few months Original tweet: https://x.com/apoorva_mehta/status/2008192076341727501

If you want your kids to have great smiles in pictures, don’t take a picture and ask them to smile. Do something that would make them have to smile, YOU smile, and have someone take a picture. Like giving them a horsey ride to bed.

Crazy how good the loop of: let’s build X, write plan/.md, review plan/.md, iterate + give feedback, execute, put up PR, merge…think of next thing to build is now. When one person can execute on the whole vision of a product, you get really special products.

RT Paul Graham Whoah, today's Times crossword puzzle. We are as legit as Michael Caine and the yapok. Original tweet: https://x.com/paulg/status/2007155333245088156

Re I pin it to the top to avoid doomscrolling in For You:

Happy new year. Now let's fucking get it!

The sooner you act like the person or surround yourself with people in #1, the sooner you’ll be able to sense where the puck is going. Put another way, if you were 13 years old today, what would you be tinkering with?

Douglas Adam’s on technology and your age. I think it’s a good moment to reflect on in the age of AI.

RT Lulu Cheng Meservey Responding to negative feedback: are you a CodeRabbit or a Claude? Getting dragged in public sucks, and it happens to every founder at some point. The natural instinct is to bristle and push back, but that only makes it worse (Obvious exceptions: if the critic is a dogmatic hater or irrelevant to the business, if you're wendys or ryanair and being mean is your brand, etc. But in this case we're talking about relevant feedback from actual customers) Instead of a kneejerk defensive reaction, try a more strategic and disciplined route: 1) Don't jump to debate the facts. All feedback contains two different things: the substance, and the customer’s desire to have their frustration heard and acknowledged. Those are discrete issues, and you need to address both. 2) Start by aligning on principles. Even if you disagree on the merits of the feedback, you surely agree that quality is important, feedback is valuable, and you're the person responsible. There’s no hope of aligning on facts if you can’t first align on principles. 3) Overindex on accountability. Take more responsibility than what seems necessary. Take so much ownership that it surprises people. This obviates their need to hector you over it and removes a lot of surface area for attack, creating space for a calmer exchange. More importantly, if you’re the founder, the reality is that every detail of your product does fall on you. 4) If you need to clarify facts, make it an explanation instead of a defense. You can state the exact same information in a way that’s either defensive and bitter, or earnest and transparent. The only difference is tone. Approach with a stance of openness and transparency. 5) If self-critique or apology is warranted, keep it straightforward. (And if it's not warranted, skip it! People can tell when it's disingenuous or forced.) No need to grovel or self flagellate. Recap the problem plainly, explain the fix, say what you’re doing to prevent it in the future, ...

Interestingly, the invention of Claude code itself has and will create many orders of magnitude more side projects than before.

Write down what you predict for 2026. Keep it private if you want or share it w some friends. It’s a great way to humble or impress yourself as the year goes on. Besides, writing it down will make you think harder about it.

❤️