Recent content on ~/adi
I’ve been a huge fan of Claude Code since it launched. Over the past few months, I’ve been using it extensively across all kinds of projects. Claude Code is still the best tool out there, though others (Gemini CLI) are catching up. I recently discovered OpenCode, an open-source model agnostic framework that supports local models, and used it to test gpt-oss-20b , qwen3-coder-30b —…
I previously wrote about using Claude Code from my phone . Since then, I’ve ported that to a headless Mac Mini. This is the updated setup. ## Why One of the key principles of working with agents is to steer early and often. You can only do that if you keep an eye on them. This setup enables me to connect to a Claude Code session running on my Mac Mini at home from my phone and macbook, so I…
Quoting David Kenji Chang on AI assistance: We need to use these tools as tools of leverage, not tools of reliance
I gave a talk at Tokyo AI #10 exploring how you can make Claude Code and other AI tools into coworkers. Below are the slides and my presenter notes augmented with a transcript from memory. References are at the end. ## Slides Thanks Ilya for building this wonderful community, organising these events, and inviting me to speak. Hi everyone, I’m Adi, and I’m here to talk to you about the…
I feel really seen by this. For years I’ve tried to be dilligent with PKM and read-it-later tools. There are some workflows that are useful e.g. these quotes are parsed from my highlights of the transcript of the Youtube video in Reader , but overall my journey mirrors the author’s, and the underlying psychology it exposes is uncomfortable to comprehend. I’m not sure I’ll…
I just published a post with gifs and videos and was pretty happy with how it was looking. A few minutes later, a friend messaged me: I use Orion , which is a Webkit-based browser (like Safari). I know my friend uses Brave, a Chromium-based browser. Why were these videos rendering fine in Webkit and not Chromium? This blog is powered by Hugo , and the posts written as Markdown. In that blog post,…
This is a list of the various things I’ve done to my claude code setup. A lot of it is not well documented, so I’m pulling it together as a master list here. ## Plan mode Always hit shift-tab twice when starting a new session to have Claude plan first, then execute. Do the same when steering significantly, or starting a new task.
Claude Code is my weird-intern-in-a-terminal 1 and it’s surprisingly useful for all kinds of things beyond writing code. For example, reviewing and commiting the code it just wrote. > commit and push this code ⎿ Found 6 new diagnostic issues in 1 file (ctrl-r to expand) ⏺ I'll create a feature branch and commit the refactored code. ⏺ Bash(git checkout -b claude-code-api-refactor) ⎿ Switched…
I’ve been been using Claude Code a lot the past 2 weeks, enough that I hit the Max plan rate limits 3-4x daily. These tips have been the biggest quality of life improvement for me. You can find more in Anthropic’s best practices . Whenever it goes down a wrong path, interrupt immediately and add a memory for the correct action, then tell it to proceed. Compact early, often, and with…
Claude Code amazed me at launch. With Claude 4 release, it cosplays AGI for code & sysadmin tasks, and convinced me to give Anthropic $100/mo. There’s a lot of good advice about how to get the most out of it so I won’t cover that - this is about how to remote operate it. ### Why CC is my weird intern in a terminal, parked on my second screen. I mostly keep an eye on it as it works,…
There’s no widely agreed-upon definition of AGI, and everything with LLMs is vibes dominated anyway, so the best we have is “I’ll know when I see it”. That’s not very rigorous, so I’ve been thinking about concrete tasks that, when completable end-to-end, represent a serious enough step forward in outcomes, and replace a significant chunk of multi-faceted human…
Quoting Neal Stephenson : ## How humans can coexist with other intelligences Maybe a useful way to think about what it would be like to coexist in a world that includes intelligences that aren’t human is to consider the fact that we’ve been doing exactly that for long as we’ve existed, because we live among animals. Animals have intelligences of many different kinds. We’re used to thinking of them…
Quoting tautologer : the optimal user interface is that you simply say this sentence aloud and then the list appears on your screen all the going to websites and clicking and filtering and so on is fundamentally cruft the purpose of software should be to minimize the friction between intent and result
Quoting Steve Yegge : First, you can’t tell interviewers what’s important. Not at any company. Not unless they’re specifically asking you for advice. You have a very narrow window of perhaps one year after an engineer graduates from college to inculcate them in the art of interviewing, after which the window closes and they believe they are a "good interviewer" and they…
Quoting Utsav Mamoria : Epistemic Anxiety is the feeling of uneasiness, tension, and concern when you want to know the truth. You worry that your knowledge is incomplete and full of errors, and you may believe it. You either lack the resources, methods or agency to get to the truth. When we surround ourselves with abundant, diverse ideas, complex ideas emerge. These ideas are unique and do not…
Quoting Chamath Palihapitiya : Replacing determinism or humans with probabilistic code is fraught with edge cases and require new ways of software development and process engineering that aren’t well solved yet. The implications to an entire generation of AI "apps" will be severe as more companies come to terms with the difficulty in getting products to work reliably in production with AI in…
### Story Yesterday I fired off a quick loop to pull down some files from a server: for f in $( cat file-list.txt ) ; do scp user@hostname:path/to/data/ $( basename $f ) local_dir/ done I just needed a couple of examples 1 so I decided to kill the command when I had them. When I closed the shell, I expected everything to stop—but the downloads kept chugging away. I ran ps aux | grep '[s]cp' and…
Quoting Lightcone Podcast : Choose Exciting Pivots Pick a pivot that excites everyone on your team to ensure commitment. Motivate your team with belief in your new direction to avoid immediate failure. Transcript: Other Speaker So then what did you decide to do? So it’s like you decided, okay, we’re going to pivot. And when we work with founders, that’s kind of stage one. So…
Quoting The Vergecast : Garbage Truck David Pierce shares that the best AI experience he had was with an app that records his son yelling "garbage truck". He would pay for the app only if it provides the recording of his son yelling at garbage trucks. Transcript: David Pierce Right, I just want to leave you with the best thing that has happened to me, and then we’re going to get out of here.…
Quoting Simon Willison’s Weblog : AI is helpful despite being error-prone if it is faster to verify the output than it is to do the work yourself. For example, if you’re using it to find a product that matches a given set of specifications, verification may be a lot faster than search. There are many uses where errors don’t matter, like using it to enhance creativity by…
Quoting Adam Mastroianni : we only know about several early Christian heresies because we have records of people complaining about them. [1] The original heretics’ writings, if they ever existed, have been lost. we all liked his first book very much and he liked it too, and one day he said to me, “This book will make literary history,” and I told him: “It will make some part of literary history,…
Quoting Linda Caroll : “Write about small, self-contained incidents that are still vivid in your memory. If you remember them, it’s because they contain a larger truth that your readers will recognize in their own lives. Think small and you’ll wind up finding the big” ― William Zinsser
Quoting Alex Heath : I think memory will be won by the company that has the most context. ChatGPT knows nothing about what you buy on Instagram or Amazon. It also knows nothing about how much time you spend on different websites. You need to have all this data to deeply personalize for the user. It’s not about who rolls out memory based on the retrieval of past queries. That’s very simple to…
Quoting bramjetten.dev : Technical debt isn’t always bad Sometimes the pragmatic solution that gets you to market faster is the right one, even if it’s not technically perfect. Just be intentional about when you take on debt and when you pay it down. Fighting the framework creates unnecessary friction. The "Rails Way" exists for a reason. It allows you to focus on what makes your…
Quoting Joanna Stern : Yet the most unsettling part was realizing the soundtrack of my life was stored on some companies’ servers. Sure, much of our identities are already in the cloud—photos, health records, etc.—but normally, we have some control over what’s there and how it’s stored.
Quoting storage.googleapis.com : any static procedure for synthetically generating data will quickly become outstripped. This can be achieved by allowing agents to learn continually from their own experience, i.e., data that is generated by the agent interacting with its environment. AI is at the cusp of a new period in which experience will become the dominant medium of improvement and ultimately…
Quoting benkuhn.net : A lot of the things people ask for advice on fall into this category. “How can I be happier?” “How can I be more productive?” “How can I have a bigger impact on the world?” Sure, there are basic life hacks like sleeping well or getting exercise. But 99% of the “secret”—the thing that separates me from Gell-Mann, or Jeff Dean—is tacit knowledge. It often can’t be articulated…
Quoting Simon Willison’s Weblog : My gold standard for LLM usage remains this: would I be proud to stake my own credibility on the quality of the end result ?
Quoting Scott Alexander : A “concept handle” is a memorable noun phrase representing a complex, often abstract topic. For example: “prisoner’s dilemma,” “Overton window,” “belief in belief,” etc. Successful concept handles can really amplify a vague idea which many people sort of understand but can’t point to and talk about. If you give that vague notion a crisp, catchy name, you can unlock a lot…
Quoting benkuhn.net : people’s biggest bottleneck eventually becomes their ability to get leverage —i.e., to find and execute work that has a big impact-per-hour multiplier. I think of finding high-leverage work as having two interrelated components: • Agency: i.e. some combination of the initiative/proactiveness to try to make things happen, and relentlessness and resourcefulness to make sure…
Quoting wikipedia.org : the Lindy effect proposes the longer a period something has survived to exist or be used in the present, the longer its remaining life expectancy. Longevity implies a resistance to change, obsolescence or competition and greater odds of continued existence into the future. [2] Where the Lindy effect applies, mortality rate decreases with time.
Quoting Paul Graham : What would someone who was the opposite of hapless be like? They’d be relentlessly resourceful. Not merely relentless. That’s not enough to make things go your way except in a few mostly uninteresting domains. In any interesting domain, the difficulties will be novel. Which means you can’t simply plow through them, because you don’t know initially how…
Quoting Blogger : Being supported by AI tools is not a substitute for mastery . You can’t borrow skills. You have to earn them. We’re heading into the age of π-shaped people: depth in two areas, and generalist across . Building depth first, and then ranging is good strategy . Clear communication matters more than ever: • for working with others (especially remotely) • for building in the…
Quoting seangoedecke.com : Here’s an unreasonably effective trick for working with AIs: always pretend that your work was produced by someone else. Note that the model will naturally assume you’ve produced any content you’re giving it (presumably because of all the assistant fine-tuning), so you have to be really explicit about the fact that you didn’t 1 AI labs will continue to be incentivized to…
Quoting Ben Thompson : The new bridge is a user interface that gives you exactly what you need when you need it, and disappears otherwise; it is based on AI, not apps.
Quoting Scott Alexander : Imagine you are in the driver’s seat of a car. You have been sitting there so long that you have forgotten that it is the seat of a car, forgotten how to get out of the seat, forgotten the existence of your own legs, indeed forgotten that you are a being at all separate from the car. You control the car with skill and precision, driving it wherever you wish to go,…
Quoting Geoffrey Huntley : — description: Cursor Rules Location globs: .mdc — # Cursor Rules Location Rules for placing and organizing Cursor rule files in the repository. name: cursor_rules_location description: Standards for placing Cursor rule files in the correct directory filters: # Match any .mdc files - type: file_extension pattern: "\.mdc$" # Match files that look like Cursor…
Quoting BG2Pod with Brad Gerstner and Bill Gurley : Florida’s Comeback Win Bill Gurley attended the NCAA March Madness tournament in San Francisco. He witnessed Florida’s comeback win against Texas Tech, despite initially thinking they had no chance. Transcript: Brad Gerstner I mean, you have to be pretty stoked coming off those wins last weekend in San Francisco. Bill Gurley Yeah,…
Quoting Byrne Hobart : the heart of a grift is that you get something that, in a technical sense, is what you paid for, but that is also not worth what you paid for it. the broad categories of finance grift are: A fancy-sounding high-fee wrapper on something that is trivial to implement manually, and A strategy that looks good in backtests because it has some kind of nonlinear blow-up risk. grifts…
Quoting Jan : That’s what I don’t get about the folks out west - if you blast someone from a storm cloud for being human on a particularly stupid day, how the heck are they supposed to learn? Scare ‘em straight, I say! Don’t look so worried. Yes the pot-bellied, laughing drinking buddy to Ebisu you just tossed your coins and bowed to hoping for success or love has been, and will always be, an…
Quoting Francesco Farina : The firehose of high-dimensional, real-time sensor data (vision, lidar, tactile, auditory, proprioceptive) from a world populated by embodied agents will dwarf the entire current internet dataset. Attempting to funnel this raw sensory stream back to central servers for storage and processing is not just economically daunting, it’s likely physically impossible at…
Quoting Simon Willison’s Weblog : llm -m gemini-2.5-pro-exp-03-25 \ ’transcribe, first speaker is Christopher, second is Simon’ \ -a ten-minutes-of-podcast.mp3 \ –schema-multi ’timestamp str: mm:ss, text, speaker_name'
Quoting The Tim Ferriss Show : Questions vs. Advice Questions are more adaptable than advice. Advice is brittle and context-dependent, while questions can be applied to various situations. Transcript: James Clear If you ask better questions, you can get better answers. But also questions are very resilient. They’re very adaptable to the context. And that’s something that’s…
Quoting BG2Pod with Brad Gerstner and Bill Gurley : Market Uncertainty and Impact The market’s uncertainty, economic and technological, drives up discount rates and risk premiums. This leads to lower multiples and market corrections, particularly in the tech sector. Transcript: Brad Gerstner Stuff going on with regard to Invest America on Capitol Hill and at the White House that I’m…
Quoting near : Put an unreasonable amount of effort into your content. The Internet is filled with content, and if yours is significantly better than average, it’ll help your odds tremendously. you have years of interesting experience in a field, you may just be able to tweet stream of consciousness thoughts and takes on things successfully, in which case the above doesn’t exactly apply: the…
Quoting Julian Bradshaw : The funny thing is that pokemon is a simple, railroady enough game that RNG can beat the game given enough time (and this has been done) [6] , but it turns out to take a surprising amount of cognitive architecture to play the game in a fully-sensible-looking way and insufficient smarts can be surprisingly double-edged—an RNG run would arguably be better at both leveling…
Quoting Wendover Productions : what makes Japanese railways great goes far beyond their corporate structure. They’re fast, frequent, and reliable, but that’s part of a virtuous cycle enabled by the success of the companies. The fact that they are fast, frequent, and reliable is made possible thanks to the success of the companies, and the success of the companies is made possible thanks to the…
Quoting TRASHFUTURE : Klarna’s AI Epiphany Klarna, a fintech company, laid off its customer service department, betting on AI. The CEO later admitted the importance of human interaction, highlighting the limitations of AI in customer service. Transcript: Riley Of Lutefisk has gone up on Deliveroo? It’s BCCI, but for buying a shine order. So this story, though, is that if you remember,…
Quoting nickcraux.com : • I like Composer (now called Agent) for simple, low impact changes. And Chat (Ask) for everything else. Just the step of manually having to click apply and taking that extra second to think about the changes it’s making, and in which files it’s happening, rather than just be presented with a wave of green changes ready to get smashed in. • Don’t blindly…
Quoting Alliance for Responsible Citizenship : if what is right and wrong depends on what each individual feels then we are outside the bounds of civilization I’ve describ three different things we educated Elites brought you we destroy the social fabric through inequality we destroyed the moral fabric through privatizing morality and we destroyed the institutional fabric what’s…