would be so cool to have a native/managed sprites.dev -like experience on Lambda. Barebones firecracker VM environment, persistent mutable filesystem, shell access, per-use billing, service exposure via open port(s), etc. Sandboxes for all the team and their agents!
Let's pretend we discovered a way to make flying 10x faster. As a result, everyone starts taking way more flights, because now they can!
The likelihood of a crash would be the same, but there's 10x more flights, which means 10x more crashes.
That's what's happening with AI and software outages.
so far it has built:
- agentic payments platform
- managed event sourcing platform
hasn’t cost me a penny!
prototyped my own agentic software factory today:
- opencode running on persistent microVMs from @fly.io (sprites.dev)
- open-weight models via openrouter
- work described via github issues
- every issue triggers new opencode session with prompt
- agent controlled remotely on mobile
complexity which existed to manage human limitations is being automated away
what could an intent-addressed interface look like?
Claude Capybara
Just seen the proposal for branching… 👀
github.com/anthropics/c...
Soul of a New Machine is by far the best software engineering book I’ve ever read! Such a fascinating story. So long Tracy 👋
CIMD is cool
> Client ID Metadata Documents (CIMD) let OAuth clients identify themselves using a URL. No preregistration necessary
client.dev
4 figures!
metrix - claude code usage dashboard built with @ratatui.rs
- token usage per day
- activity heatmap
- cost estimates
- file ops stats
- tool call types
- longest session
- most files touched in single session
- conversation turn counts
- most edited file
github.com/lukehedger/m...
opencode webui running on iOS via tailscale
got remote ai coding sessions working with opencode webui and Tailscale this weekend!
It's inherently social/creative but we must include a non-sentient tool that is a poor simulacrum in the process
specs and docs are useful but they are not the software. Nor are they the definitions of the software - that’s in our collective heads
I do get the discourse around “write it down” but still feel software engineering is an inherently social/creative process and we need to find ways to make Claude a natural part of that
often find myself reviewing code during an implementation session and wanting to ask several questions all at once about different aspects whilst not blocking/confusing the build. takes a lot to keep track of claudes responses doing this in a single chat thread
is there anything like threads for Claude Code? Not separate sessions
/subagents/teams - the same session with the ability to branch off at certain points to discuss something deeper whilst keeping the main thread going. Am I missing something and this is possible!?
definitely ship code quicker than before but the biggest difference these days is QUALITY. I’m writing typescript with levels of type safety and error handling that I didn’t even think was possible. Expertise that would have taken me years to obtain is now available to me as I iterate towards goal
yes, Coach
That means you can load an entire codebase, thousands of pages of contracts, or the full trace of a long-running agent - tool calls, observations, intermediate reasoning - and use it directly. The engineering work, lossy summarization, and context clearing that long-context work previously required are no longer needed. The full conversation stays intact.
is that the end of context engineering?
claude.com/blog/1m-cont...
You work for Lego? Drop the manufacturing tolerances on your bricks by 5% and my kids stop buying.
Actually, that might be good for me.
Ha! I thought we were talking about software 😂
Bricks are another matter…
do one way doors still exist? 🤔
> The question isn't whether you can afford to do it. It's whether you can afford to wait.
🎯
How We Pulled It Off In Two Weeks Instead of writing a migration proposal and spending weeks on planning, Markus built 60% of the replacement and showed it to the team. This is a key lesson in the age of coding agents: If you're trying to convince a team to do a major rewrite or feature implementation, a working demo will do more than any document. A year ago this would be terrible advice. You'd risk weeks on something that gets scrapped. But now that building a working proof of concept with agents costs less than writing the proposal, the calculus has changed.
have found this to be true. intuition tells you a migration is necessary, now forging ahead allows you to dis/prove assumptions asap
strawberrybrowser.com/blog/react-t...
