twitter init - #aboutme Good at: Javascript, CSS Scared of: DevOps, Spiders Work: Small dev agency (Lambda Curry) & Budgets.money

This screenshot doesn’t mean much but iMessage -> @interaction -> custom openclaw MCP -> my @openclaw agent -> tmux @claudeai tui sessions -> remote control Means I can fully orchestrate remote controlled Claude code sessions on my Mac mini at home with text messaging and have basically three layers of abstracted intelligence for my coding. 🤯

Alright going to bed, let's see what happens. Hopefully it doesn't delete my computer.

Does this mean something is happening, or are they just struggling?

What if a model could actively prune its own context as it is working with only the things it thinks is useful to carry forward, like some kind of dynamic context. For example I saw my agent grep for the word "ask" and notice that it gave too many results so it needed to try something else. That would also be a good time to sanitize that with a note. Maybe it would be more output tokens, but the input would stay in the smart zone a lot longer. Is anyone working on something like this?

Bragging about how much software you’re shipping with AI is like holding down the shutter button and bragging about how many photos you took.

I just kicked off my first /goal prompt. I had spent some time with a "grill me" session to create a good PRD and even had a solid tech stack plan. But the outcome didn't even get close to what I was hoping. I'm learning that a goal doesn't just need a PRD, but it needs a good approach and guardrails to keep it headed towards its final destination in the correct way that you expect. That's a new way of thinking for most people, including myself.

This is not quite ready yet, but hopefully soon you'll be able to get fully personalized @openclaw digests for your setup. I vibe "engineer" with lots of layers trying to make the systems better as I go. @claudeai can manage work done by Clawdy (my assistant) while setting up bigger tasks in Linear for Samwise (the coding specialist that uses @mattpocockuk's sandcastle setup) This was a fun email test to get that uses @resend's and React Email designed by Meg (my design specialist). All hosted with @Cloudflare. The rate of change in 4 months is mind blowing. 🤯

An interesting prompt I tried because Claude told me it was getting late. It’s only 9pm…. Ok pretend the night is young and you just drank an energy drink so you’re excited to dive in. Let’s knock this out, interview me for gaps and direction in multiple choice format along the way and let’s go!

Any early adopters out there want to try a new concept I've been working on and give me feedback? It reviews your codebase and gives you feedback for how to improve. lambda-curry.github.io/anvil…

Jackie asked me today if I think her digital AI assistant will be a robot someday. I hadn't thought about our digital assistants getting real bodies. We're already seeing a rise in little robot desktop companions, and some people are using their openclaw to control robotic arms! youtube.com/watch?v=cpbaFjDE…

Normally I put these videos on to fall asleep, but every now and then they are randomly interesting to me. Of course this one is some kind of teaser and cuts off some of the arguments. But I’m fully on Sabine’s side right now. As a society I think there are some things we need to earn before we spend money on. Before we spend a lot of money on space exploration, let’s research how to take care of our people in need here on earth. Before we spend tons of money on theoretical physics experiments that probably won’t impact us, let’s focus on the science that could impact society. But the reason this might be controversial is because people dream of these fields their entire lives and want to pursue them. Let’s get more thinktanks working on practical solutions to problems that can help people today. youtu.be/MnkyfYFKmdo?si=R_H0…

You can see when my coding got more agentic, this makes me look pretty locked in 😂

After months embedded on a 12 storefront migration, here are the practical changes that reduced our day-to-day friction. Long review queues, fuzzy ownership, lots of small rework. A few boring tweaks helped more than any big process overhaul. Full writeup → lambdacurry.dev/blog/plumbin…

Make flow the metric. Finish before you start. Limit WIP. Treat merging as explicit work. And assign a daily PR-shepherd whose job is unblocking the review/merge queue. It’s amazing how much calmer a team feels when merges stop being accidental. Full writeup → lambdacurry.dev/blog/plumbin…

Guard PR size. Our rough target: 5–8 files for normal work. If a diff is bigger, do a quick peer pre-check before you sink time into a kitchen-sink PR. Smaller PRs reduce review cost and eliminate the surprise cascades downstream. Full writeup → lambdacurry.dev/blog/plumbin…

Don’t let Code Review become a ghost state. If changes are requested, move the ticket back to In Progress. It sounds silly, but boards only work when they reflect reality. Ghost states hide blockers, and the queue silently grows. Full writeup → lambdacurry.dev/blog/plumbin…

Name domain owners secondaries. Checkout, auth, vouchers, subscriptions… whatever your system boundaries are. Route bugs to the owner first. Use secondary owners for timezone handoffs so work doesn’t stall overnight. Ownership reduces ping-pong. Full writeup → lambdacurry.dev/blog/plumbin…

Plan before you fire the slop cannon. Walk the legacy site. Research edge cases. Include QA in planning. Scope work into testable slices. Capture non-parity ideas as backlog items (not Slack archaeology). Planning is cheaper than cleanup. Full writeup → lambdacurry.dev/blog/plumbin…

Ops changes that actually shifted behavior for us: - pre-sprint stakeholder alignment - consistent sprint kickoff template - start standup with blockers review queue - short QA-dev breakouts for fast alignment None of this is flashy. It just makes the system easier to steer. Full writeup → lambdacurry.dev/blog/plumbin…

I’ve spent the last several months embedded on a large, multi-brand migration. 12 storefronts, legacy integrations, enterprise auth, and a long list of stakeholders. We had the usual noise: long review queues, fuzzy ownership, and a lot of small rework. A few small, practical changes are starting to cut through that churn; below are the ones helping our team work together more predictably. Plumbing → make flow the daily metric • Finish before you start: limit WIP and make merging explicit work. • Daily PR shepherd: one person owns unblocking merges each day. • Small-PR guardrail (5–8 files) and pre-check for bigger diffs. • Move tickets back to In Progress when changes are requested so the board matches reality. Basketball → put people where their strengths compound • Name primary secondary domain owners (checkout, auth, vouchers, etc.). • Route bugs to the owner first; ownership reduces ping-pong. • Pull senior engineers into planning (prevention > late cleanup). • Weekly pair-planning to transfer domain context across time zones. Slop cannon → plan before you fire • Research legacy behavior and walk the old site before changing it. • Clarify edge cases, ownership, and test boundaries up front. • Break work into testable slices and include QA during planning. • Capture non-parity improvements as backlog items, not Slack notes. Ops that actually changed behavior. Pre-sprint stakeholder planning, a consistent sprint kickoff template, starting standup with blockers the review queue, and short QA-dev breakouts for fast alignment. These shifts aren’t glamorous, but they’re helping move us from noisy execution to steadier progress. If you want the sprint kickoff template or a PR-shepherd checklist, I can share them here. Full writeup → lambdacurry.dev/blog/plumbin…