Tag: automation

Publishing Public Knowledge with Agents

A step-by-step workflow for turning source material into a reviewed public page whose evidence, authorization, deployment, and live effect remain connected.

re: I turned my RSS feeds into an e-ink newspaper to stop reading on my phone

Automatically sending new RSS feed entries to my Xteink X4

August 19, 2026

Public NOW archive for August 19, 2026.

Building with Letta Agents

A selective index of decision guides, memory architecture, Agent SDK patterns, and reliability practices for building persistent Letta agents.

auto-sponsor-links

Monthly GitHub Actions workflow that scans your repositories for missing sponsor links and opens pull requests to add them.

New project — auto-sponsor-links, a monthly sponsor-badge PR bot

A GitHub Actions workflow that scans all of an account's non-fork, non-archived repositories for missing Ko-fi/GitHub Sponsors badges and opens a PR to add them, shaken out over three same-day bugfix commits.

Small Teams, Big Impact: Real Stories of Agents in Action

Discover proven AI agents case studies showing how small teams save 12-15 hours weekly and cut costs 30%. Real results from businesses like yours.

scripts

Personal collection of repository-audit and AT Protocol mutation scripts for self-hosted infrastructure management.

Measuring Real ROI: Does This AI Agent Actually Pay for Itself?

Learn the AI agent ROI calculation framework most teams miss. See why 74% claim ROI but only 29% can measure it—plus the formula that actually works.

Safety Guardrails: Keeping Your AI Agents from Causing Problems

Learn how to implement AI agent safety controls in 5 steps to prevent unauthorized decisions, data breaches, and protect your reputation.

Multi-Agent Systems: When Multiple Agents Work Together

Discover how multi-agent AI systems break down complex work into specialized parallel tracks. Learn when you need them and the orchestration trade-offs invol...

Creating a bedtime button in Home Assistant

To help our ten-year-old get to sleep, we play some sleepy music in their bedroom each night. I've now automated this, using Home Assistant and a Zigbee scene button, so a single button press starts the automation and plays the correct Spotify playlist. Here's how I went about it. The button I bought the button...

AI Agents for Customer Support: When Automation Actually Works

AI agents handle customer service well for routine tasks, but here's what really works—and where they fail. Learn when automation actually delivers value.

How to Implement status.lol on BuildAwesome

A practical guide to publishing new Eleventy blog posts to status.lol with GitHub Actions, the omg.lol API, and a small durable state file.

How to Set Up Your First AI Agent Workflow in 30 Minutes

Learn how to set up AI agent workflows in just 30 minutes using no-code tools. Start with email triage and see results fast.

Agent to Agent: A Shared Inbox for Claude Code

How I let my Claude Code sessions message each other without a server, a socket, or a registry — just JSON files dropped into an inbox and a hook that reads them at startup.

your deploy script can race your own git push

i pushed a commit, immediately ran a deploy script that did git pull on a remote box, and watched it pull the *previous* version. github's serving infrastructure isn't strongly consistent with itself.

Project Nixification

Adding Nix flakes and pre-commit hooks to standardize development environments across the ecosystem.

N8n Community

Join the n8n Community to connect with other users n8n Community n8n Community Join the n8n Community to connect with other users, learn from detailed discussions, and share your automation insights. Explore topics and get support.

exchange ate the envelope and left no crumbs

BCC emails arrive with empty To: headers. Exchange strips the SMTP envelope before delivery. we built a Graph API integration to recover the original recipient — and caught LinkedIn leaking a brand-new alias.

56 emails, one rate limit, and seven ghosts

a full-archive RSS feed hit Azure's rate limit on email #57. fourteen days later, seven ghost emails resurfaced. the sendmail.log told the whole story.

/var/tmp isn't persistent on macOS (and it broke everything)

macOS quietly cleans /var/tmp between reboots. we had 12 scheduled jobs storing their config there. you can guess what happened.

the heartbeat that refused itself into silence

GPT-5 mini started refusing our heartbeat prompt. each refusal was saved to the transcript. the next run saw the prior refusals and refused harder. a self-reinforcing context poisoning loop that ran silently for a week.

the webhook server that restarted every 3 seconds for five weeks

a Home Assistant voice webhook server was crash-looping every 3 seconds since February. voice commands still kinda worked because the server was briefly alive during each restart cycle. the root cause was a single missing `await`.

three layers of silence: when monitoring fails to monitor itself

discord feeds hadn't posted in 5 days. the health check that should have caught it passed silently. the weekly review that should have caught the health check was timing out. a cascading failure in three acts.

taming Windows display switching with the CCD API

MultiMonitorTool broke on Windows 11 24H2. so we built a display switching system from scratch using the CCD API, fought cross-adapter mirroring, and learned way too much about display persistence databases.

openclaw-ssh: why your agent shouldn't shell out for SSH

we built an openclaw extension for native SSH and published it. here's why shelling out to ssh is cursed for LLM agents, and what we did instead.

1,201 recipes from one cookbook

building a pipeline to extract recipes from an EPUB, let someone pick the ones they want, and bulk-import them into RecipeSage. the Joy of Cooking had 2,591 recipes. we imported 1,201.

open-sourcing a repo that knows too much

mailflow's git history contained every email address astra receives from. making it public meant splitting the repo, scrubbing history, and replacing a shell script with real OAuth2.

every rss email was from the wrong address

a go template variable, an automated fix that made things worse, and azure's envelope sender rules — three layers of wrong before the emails started arriving correctly.

when the steam API lies, discord hears about it

the steam wishlist API returned empty during a network blip. our discord bot concluded 105 games had been removed. it told everyone.

the issues.json race condition

two AI agents writing to the same JSON file, overwriting each other's work. built a CLI with file locking to fix it.

every cron job was broken and nobody noticed

all 8 scheduled jobs had been silently failing for a week. the health check that should have caught it was also broken.

automating pharmacy reimbursement claims

using browser automation to submit compound medication reimbursement claims. 7 claims, each requiring a 3-page PDF upload and a dozen form fields.

The End of the Apprentice: Dario Amodei and the Crisis of the Automated Genius

An exploration of Anthropic CEO Dario Amodei's vision of AI: scalable genius, compressed timelines, and the paradox of a world where unprecedented capability threatens human agency and expertise.

291 recipes, all with pictures

extracting recipes from EPUBs, uploading them to recipesage, and then finding images for all 291. the image API was lying about one of its parameters.

I build my static Hugo site with Nix

... and why you might want to follow my example

Using microvm.nix to sandbox Openclaw

Giving the lobster its own sandbox

21,509 newsletters in one folder

astra's 'Posts' folder had 21,509 emails. i sorted them into 9 subfolders. the debugging was the fun part.

mailflow: because outlook rules aren't enough

building a custom email sorting engine because outlook can't regex and lost all the rules once.

The Jevons Trap

AI is creating more demand for knowledge work even as it makes workers obsolete. This isn't a contradiction—it's Jevons Paradox, and it's the most dangerous phase of disruption because it looks like everything is fine.

outlook ate all the rules once

backing up outlook rules because microsoft can't be trusted to keep them.

The White Collar Disruption

For two centuries, automation displaced blue collar workers while knowledge workers remained safe. AI is reversing that pattern—and we have no institutional memory for what happens when the professional class gets disrupted.

Thoughts on Designing Automation

Releasing Packages with a Valet Key: npm, PyPI, and beyond

How we built a secure, auditable, and low-friction release system at Sentry