@elliscs.bsky.social
How I Stop LLMs Drifting In Production Codebases: scottspence.com/posts/how-i-... heavy Svelte focus
How I Stop LLMs Drifting In Production Codebases - Scott Spence
The problem I care about with LLM coding is drift: one plausible shortcut gets copied, then future sessions start treating it as how the app...
scottspence.com
Do you like this feed? If you do repost this and share it with your friends! bsky.app/profile/scot...
I’m building my-pi.dev It’s the Pi harness but with Svelte extras (amongst other things) essentially $effect use is blocked 😅 Coding standards that are configurable, I like snek_case so that’s enforced at the harness level etc Latest on there is the observeability extension
My-Pi
A ready-to-run Pi CLI distribution with scoped MCP tools, project skills, context reduction, recall, LSP tools, redaction, telemetry, and team mode.
my-pi.dev
If you're exploring other harnesses and are considering Pi I've made a load of extensions for it - reduce context bloat - MCP management - LSP - Configure skills loaded per project/org You can also try my-pi which has a load of other stuff on there! github.com/spences10/my...
GitHub - spences10/my-pi: Composable pi coding agent with MCP, LSP, agent chains, prompt presets, and local eval telemetry
Composable pi coding agent with MCP, LSP, agent chains, prompt presets, and local eval telemetry - spences10/my-pi
github.com
The durable output of engineering is shifting from code to the harness around it: documentation, skills, intent. Externally, SaaS companies will be evaluated on agent harness quality, not dashboard UX. open.substack.com/pub/engineer...
What Outlives the Code
The monomyth of software engineering.
open.substack.com
Our friends at @ainativedev.io just dropped the first wave of speakers for their London event 👀 Speakers from Tessl, Netlify, Thoughtworks, Docker, Snyk, GitHub Security Lab and more. 📍 June 1–2, London 🎟 30% off with code SVELTE30 💻 Free virtual tickets available 👉 ainativedevcon.io
Devcon Spring 2026
The best place to master agentic coding, helping developers and organisations build with agents
ainativedevcon.io
Traditional Cost of Goods Sold tracking: Redshift, Airflow DAGs, weeks in the BI worklog. Yesterday I built the same thing with Claude Code skills in a day. Bronze → Silver → Gold, but the refining is agentic. open.substack.com/pub/engineer...
New post: Zero Trust Launch Day We couldn't predict what would break. So we built the instrumentation to search production instead. First client team fully live, sub-one-hour resolution on blocks. open.substack.com/pub/engineer...
Running 3-5 Claude Code instances in parallel. They're all productive. They're all stuck waiting on me. The speed isn't the problem. Surviving it is. New post on multi-agent orchestration, what stage 6 actually feels like. open.substack.com/pub/engineer...
Orchestrate to Survive the Speed
The factory floor is uncomfortable
open.substack.com
3 books in to Iain Banks Culture series. Its amazing how prescient he was
Build vs buy math everyone gets wrong: You're not comparing SaaS cost to engineer salary. You're comparing it to what that engineer would've shipped instead.
Build vs Buy
The spreadsheet comparison misses the real cost
open.substack.com
"AI will commoditize" crowd in shambles. OpenRouter data shows demand is price-inelastic. Users pay 40x more for Claude over commodity models. Happily. Quality > cost because what I'm comparing against is my own lost time. openrouter.ai/state-of-ai
State of AI | OpenRouter
An empirical study analyzing over 100 trillion tokens of real-world LLM interactions across tasks, geographies, and time.
openrouter.ai
Notre Dame took 180 years to build. It burned in hours. Restoration took 5 years and billions. Last week, a scoping bug wrote bad data for hours. Fix took minutes. Cleanup took days. The asymmetry between construction and destruction governed everything. Still?
The Asymmetry Between Construction and Destruction
What's worth protecting when rebuilding is cheap
open.substack.com
Gemini Flash beating Pro could just be a sign of overfitting on RL, right?
the fact that codex now supports skills is telling this type of procedural knowledge is durable and useful
Working on a new version of Svelte Form Builder What’s new: • Generates Svelte Remote Function code • New layout with fields, editor, preview • Works with Valibot, Zod, and ArkType Preview: svelte-form-builder.vercel.app/v2 Still in alpha - early stage, feedback welcome!
Your evals > public evals. $ is the only benchmark that matters.
The Only Eval That Matters
GPT Image 1.5 won every benchmark. Users still prefer Nano Banana Pro.
open.substack.com
