Finally. It's time. A decade after starting my first company (wow time flies), I'm working on something new! The new company is Datanaut, and while it's too early to share specifics, I'm excited to work with @scottfaust43 and our small team to re-imagine how we as humans interface with AI, each other, and the treasure trove of data that is fragmented across the apps we use every day. I'm not much of a sharer, but this go-around I'm committed to building more in public. Expect to see regular posts from me on things like ui / ux design, agentic patterns, running a lean company, and more in the weeds technical stuff - basically a log of the real-world challenges that we're faced with every day as we build Datanaut. I can't wait to share more. Most importantly, I can't wait to start shipping again 🚀.

no matter what I do in claude file, skills, output style, system prompt.. the sheer number of comments that opus/fable adds is getting out of control. it's impossible

We switched from @bunjavascript to @nodejs to take advantage of Sentry profiling (bun not supported yet) and were quite surprised by the result.. not sure if a bun + @EffectTS_ thing or what root cause was yet, but safe to say we'll be sticking with node for now.

Wishlist item for @github - would be great to have a 1 click "re-write / consolidate" extended description option when merging PRs.

🔥 #AGI has arrived

Every time i restart my computer and BetterDisplay w this option isn't running, I wonder how folks can live without it. Should be a default option in OSX!

StreamDB - a reactive database in a Durable Stream. Designed for AI apps and agentic session state. You give it a Standard Schema. You get type-safe, multiplexed data sync into @tan_stack DB. Links in the 🧵

the more claude code types for me the worse i get at typing myself

Messing around with formal Job abstraction w @EffectTS_ and @Cloudflare. Define a job like you define state - `Job.Tag`, implement a handler, then schedule from any actor. Works in memory (for tests) or on CF. Builds on x.lingyaoai.com/marbemac/status/202922… -

love that @EffectTS_ ships the ts source with the npm distribution - more libraries should do this. really helps llm's out.. don't need to clone the repo and point them at it if they can just look in node_modules

Been working on an @EffectTS_ lib to smooth over the experience working with actor patterns, with a specific focus on making it trivial to swap out backing implementation between @Cloudflare durable objects and node/bun/memory. Lot more to it than this, but it's coming along nicely!

Very excited about the @EffectTS_ v4 beta! Haven't seen much mentioned about this and curious if there's any guidance on what to expect wrt tsc type check perf, and stack traces / debuggability dx. Similar characteristics between v3 and v4, improvements, not sure yet?

Let's goo - the parallel routes proposal for @tan_stack router looks so promising github.com/TanStack/router/p…

favorite alias: `git gone` - it removes local branches whose upstream has been deleted add it: `git config --global alias.gone "! git fetch -p && git for-each-ref --format '%(refname:short) %(upstream:track)' | awk '$2 == \"[gone]\" {print $1}' | xargs -r git branch -D"`

For Christmas I'd like an e2e architecture that: 1. uses @EffectTS_ as much as possible AND... 2. with actor based backend that makes it easy to swap concrete impl from CF to node, with incredible DX ala @rivet_dev AND.. 3. with a thin RPC layer for logic that is not owned by actors, and an incredible client<->server DX ala orpc AND.. 4. an easy way to expose some of those RPC procedures over a public api ala ORPC openapi integration Is that too much to ask? 🙏😅

forget creating a mock backend - really enjoying pattern of: 1. start with local only @tan_stack DB collections 2. hook frontend up to those collections 3. once all feels good, define pg schemas, migrations, backend handlers, etc, and swap the collections to electric/query

🔥 huge

sheesh, the amount of stuff you can do with the actor model, assuming each actor is easy to create on demand / lazily, and has access to it’s own storage, is nuts. very bullish on the actor model in general

Really cool tool to explore diffs of claude code prompt/tool changes over it's version history cchistory.mariozechner.at/

Haven't had enough time to get a gauge on output quality, but hot damn cursor's new "composer" llm model is VERY FAST. like orders of magnitude faster than sonnet