Tom Janssens

@_baretto I'd wish I didn't have to agree

@ToJans Better learned earlier than later.

$10K MRR is not $10K MRR for life. Don't fall asleep at the wheel.

@karpathy @AllWorkNoPlay (it can only run in development mode as it needs access to it's own source code, but that's easy to fix of I'd ever need it to become a "proper" app)

@karpathy @AllWorkNoPlay I have a similar setup in all my code projects, but I added skills to the mix. And I have meta skills that maintain all the docs and skills like a gardener. When I'm done with a session, I run "improving skills", and it further optimises.

LLM Knowledge Bases Something I'm finding very useful recently: using LLMs to build personal knowledge bases for various topics of research interest. In this way, a large fraction of my recent token throughput is going less into manipulating code, and more into manipulating knowledge (stored as markdown and images). The latest LLMs are quite good at it. So: Data ingest: I index source documents (articles, papers, repos, datasets, images, etc.) into a raw/ directory, then I use an LLM to incrementally "compile" a wiki, which is just a collection of .md files in a directory structure. The wiki includes summaries of all the data in raw/, backlinks, and then it categorizes data into concepts, writes articles for them, and links them all. To convert web articles into .md files I like to use the Obsidian Web Clipper extension, and then I also use a hotkey to download all the related images to local so that my LLM can easily reference them. IDE: I use Obsidian as the IDE "frontend" where I can view the raw data, the the compiled wiki, and the derived visualizations. Important to note that the LLM writes and maintains all of the data of the wiki, I rarely touch it directly. I've played with a few Obsidian plugins to render and view data in other ways (e.g. Marp for slides). Q&A: Where things get interesting is that once your wiki is big enough (e.g. mine on some recent research is ~100 articles and ~400K words), you can ask your LLM agent all kinds of complex questions against the wiki, and it will go off, research the answers, etc. I thought I had to reach for fancy RAG, but the LLM has been pretty good about auto-maintaining index files and brief summaries of all the documents and it reads all the important related data fairly easily at this ~small scale. Output: Instead of getting answers in text/terminal, I like to have it render markdown files for me, or slide shows (Marp format), or matplotlib images, all of which I then view again in Obsidian. You can imagine many other visual output formats depending on the query. Often, I end up "filing" the outputs back into the wiki to enhance it for further queries. So my own explorations and queries always "add up" in the knowledge base. Linting: I've run some LLM "health checks" over the wiki to e.g. find inconsistent data, impute missing data (with web searchers), find interesting connections for new article candidates, etc., to incrementally clean up the wiki and enhance its overall data integrity. The LLMs are quite good at suggesting further questions to ask and look into. Extra tools: I find myself developing additional tools to process the data, e.g. I vibe coded a small and naive search engine over the wiki, which I both use directly (in a web ui), but more often I want to hand it off to an LLM via CLI as a tool for larger queries. Further explorations: As the repo grows, the natural desire is to also think about synthetic data generation + finetuning to have your LLM "know" the data in its weights instead of just context windows. TLDR: raw data from a given number of sources is collected, then compiled by an LLM into a .md wiki, then operated on by various CLIs by the LLM to do Q&A and to incrementally enhance the wiki, and all of it viewable in Obsidian. You rarely ever write or edit the wiki manually, it's the domain of the LLM. I think there is room here for an incredible new product instead of a hacky collection of scripts.

@karpathy @AllWorkNoPlay I also have a CRM skill, that can access my code, my GTM, my web content, my prospect date and platform DBs. It has a html react UI and prospectname .md etc. In-between the UI and the data files I only have prompts, plus a chat window. "Add a dashboard to the UI" was all it took

Coders in 2030 be like:

Who needs DLSS when you have @threejs ?

Three.js: ✅ 175KB gzipped ✅ Plain JavaScript ✅ Desktop, Mobile, VR, AR ✅ No shader pre-compilation ✅ Instant web deploy ✅ Wrap for Steam/App Stores ✅ Huge open-source ecosystem ✅ LLM friendly ✅ No registration ✅ $0/year per seat ✅ 100% hackable ✅ 100% yours forever

@vladikk AI can write all the code you need. But you still need to do software development.

AI can write code. Time we agree on that. But software engineering isn't about code: it's about solving problems. Some problems are trivial, but the interesting ones don't have trivial solutions. They require experimentation, risk taking, and even pivoting. That's why the AI era is the third coming of Domain-Driven Design. The ability to analyze and communicate problems effectively is a strategic superpower now more than ever.

@mrdoob @perplexdotgg @threejs Why the cornell box? How about an M&M package ;)

Testing Bounce by @perplexdotgg and @threejs SSGI

.@bcherny I'm mostly surprised by smallish prompt for big assignments. F.e. my strategic advisor skill "McKinsey-in-a-box" burns a whole context window on one question. Total word count of the skill's subfolder: close to 100K

@mrdoob Awesome! mrdoob.github.io/Starter-Kit-Ra…

🎮 mrdoob.github.io/Starter-Kit-Ra… 💻 github.com/mrdoob/Starter… Build a track and share the link! 👋

Ported Kenney's Starter Kit Racing to @threejs 🏁 Also built a basic track editor so people can create their own maps and share them with friends. x.com/KenneyNL/statu…

@threejs per week!

9M downloads/week 🚀

Very first time I realized how big this is: Spent some time in the UI chat window searching for 10 new clients... Show them in a nice list with the current status Add a button to do this on command. UX paradigms will change big time in the future!

Started cold outreach last week First & second outreach: 12.5% reply rate Third and fourth outreach: doubled down on the one proposition that worked: 33% reply rate Total sample size: 40 One prospect booked a meeting last week for this week. Closed on first call High hopes!

Nice side effect: add a dashboard as a first tab! ... Wait 3 minutes ... Dashboard UI available #winning

All driven by an ad hoc "CRM": Client research, personalization, email drafts. Only UI and data, nothing in-between but prompts

Tested 5 different angles in first 2 batches, tiny sample size but it does seem to work...

@Av1dlive @grok how likely is it the incumbents will just replace the harness or provide one of their own (anthropic etc)

.@bcherny I've created a CRM app that uses Claude code as the driver, so just a UI and data, Claude code in-between... Not only does it allow you to edit data, but also it's own prompts, UI, data format etc I'd say it's the future!