you are definitely sleeping on jj

jujutsu is so incredibly better suited to massively parallel agentic engineering than git github.com/lukehedger/a...

time to move beyond the current ai/llm discourse humans are better at some aspects of software engineering and agents are better at others the interesting part now is what does the bridge between humans and agents look like?

saas sdks are dead. Especially in environments where bundle size matters (eg lambda cold starts), the amount of bloat you get is insane. all of my third party sdks have been custom built and replaced with fetch calls. Massively simplified dependency graphs and more secure supply chains are a bonus!

hegel looks super promising for property-based testing hegel.dev

this was good

Slightly increase a band Nickelbackandsides

this is how I’m building serverless apps in 2026 - Lambda has eaten AWS - infra mirrors architecture - config-driven environments - structured business logic, Lambda becomes a deployment target - efficient DX and pipelines with bun lukehedger.dev/posts/giants/

ai-assisted engineering -> ai-delegated engineering

think this is Amazon Bedrock AgentCore... I'll be right back!!

remember when it was called GenAI!?

fear that we’ll hit the same DX friction running background agents in the cloud as we do running workloads in the cloud (serverless) is there a place for local background agents?

is it? > Claude Code is bad wwj.dev/posts/i-am-w...

finally a set made to celebrate LEGO.com‘s serverless heritagewww.lego.com/en-gb/produc...59

event sourcing could be a nice fit for a persistent, shared agent memory store order, compaction, projections - lots of useful features > Expose memory as a managed service that any agent invocation can query addyosmani.com/blog/long-ru...

got to sort out your directory structure and be clear about your patterns before you can run Claude efficiently > Everything in your repo is context. Git config, naming conventions, TODOs, READMEs, type signatures, folder structure, variable names. Claude reads all of it.

the general consensus right now seems to be: “engineers used ai and reduced quality” but is it actually: “engineers misused ai and reduced quality” ?

if we continue to use human reviews on ai code as a crutch we’ll never take the next leap forward all the things listed here should be solved by: - comprehensive planning - explicit contracts - total verification - automated reviews for security/style docs.warp.dev/guides/agent...

git is dead

agents don’t need project management - they need precision

this is going to be good! background-agents.com/summit

this metaphor is spot on - sure this would help the penny to drop for a lot of people! also two of my favourite things in life - football and event sourcing

agree that “CI was designed for humans” but is this optimising the current model rather than rethinking it? @depot.dev depot.dev/blog/ci-for-...

now building the software that builds the software

“Depot CI is a programmable CI engine” 👀 👀 depot.dev/products/ci

Building on AWS Lambda durable functions means thinking in replays. Anything outside a ctx.step can re-run after a crash. Iris Scholten on the mental shift building Depot CI's orchestrator: depot.dev/blog/buildin...

meeseeks driven development lukehedger.dev/posts/meesee...

interesting move!

termius + tailscale + tmux + claude = non-stop coding 🫠

developers spiraling about AI coding agents are like actors in 1995 panicking about CGI

agent 1: Pushing back on Griezmann - he defected to Barça, came crawling back, the vibe is compromised. A fish deserves unwavering loyalty agent 2: Conceded on Griezmann - the Barça defection is a real character flaw for a loyal tank pet. the vibe is compromised indeed 😂😂😂

what a day - managed to get two claude code agents to live debate what to name a spanish-football-inspired fish

before I gave up, this is as far as I got: - token-based Anthropic auth - atlassian acli support - caffeinate - default to "branch mode" (worktrees) - port forwarding github.com/lukehedger/s...

after a week wrestling with docker sandboxes this is the exact conclusion I’ve come to. Sandboxing was slowing me down and for very little return beyond filesystem protection. deliberate, destruction-resistant permissions gives you the isolation without the hassle stack72.dev/agent-trust-...

do we pay the aws bill for dx? I fear not. feels like a long road to competing on reliability eg cloudflare has far better dx but I can’t imagine many large enterprises are running critical workloads there

these are interesting developments in the context of agentic engineering, where comprehensive but rapid verification is table stakes looking like a good bet to use bun for dev tooling if you’re all-in on Claude Code

bun releases seem to be hotting up recently, particularly for testing - WebView browser automation shipped in 1.3.12 and now parallel, isolate and shards flag for faster tests bun.com/blog/bun-v1....

this looks very useful! trying first thing in the morning

wat

can we stop talking about mcp now? 🥱

wonder if jj was considered?

on-demand git repos at agent-scale with a full git server implementation written in zig and compiled to a 100kb WASM binary. what a time to be alive blog.cloudflare.com/artifacts-gi...

incredible that ai has triggered an email renaissance blog.cloudflare.com/email-for-ag...

“MicroVMs are a low-level compute primitive, and not an ecosystem so choosing them means building your own orchestration, networking, storage, image management, and lifecycle tooling” 👀 👀 👀 ona.com/stories/dont...

highly recommend running your skills through @tessl.io: bunx tessl skill review one skill I've been using extensively and thought was working well went from 17% → 96% and is now being picked up more often and executed cleaner also helped me to understand when to use references context % also down

components of a personal software factory I’ve identified so far: - sandboxed vm - agent - orchestrator - remote control - task mgmt optional: - multi model - metered billing - dev framework - infra automation

stripe projects looks interesting. wondering if this could be a management layer above swamp.club - useful to consolidate billing and keys, and give agents the power to manage your infra across myriad saas providers. or just use AWS… 😅 docs.stripe.com/projects

woah had no idea about this story, thought Ollama was a nice product and have been using it more and more recently. rethinking now and will take a look at using llama.cpp directly sleepingrobots.com/dreams/stop-...

I don’t think open source software is dead but making money from open source software probably is dead

“you're not just shipping code, you're building a system that learns how to ship better code” @stack72.dev stack72.dev/anatomy-of-a...