DigestAvi's weekly digest, 2026
The week of Sep 26 to Oct 3.
Sep 26 to Oct 3, 2026. 24 bookmarks saved this week, sorted into 4 topics.
Agent infrastructure and the software factory
9 bookmarks
-
Factory Log
A macOS app that builds a timeline of what your coding agents did each day, per project, with all data staying local.
Why it matters. you're running multiple agents against Rails apps and need observability for the factory.
-
"Factory as code" — Ben Holmes
A Terraform-like YAML pattern defining agent factories: a foreman plus specialized agents, runners, MCP servers, webhooks.
Why it matters. it maps almost directly onto your OpenClaw software factory vision.
-
Copilot Triage
A tiny Ruby agent that triages GitHub issues, built for RubyLLM — a lightweight alternative to heavyweight agentic workflows.
Why it matters. Ruby plus agents, two of your favorite things, and a pattern worth stealing.
-
Cloudflare OS
A managed agent workspace tying into company data and systems, with new GitHub and Google Workspace integrations.
Why it matters. it's the "fully managed" end of the spectrum versus your self-hosted instincts — worth knowing the competitive shape.
-
manychat-lite
Self-hosted ManyChat-style Instagram auto-DM on Cloudflare Workers plus D1, with keyword rules and an admin dashboard.
Why it matters. social automation without the spam fits your interest in growing presence without low-quality output.
-
Depot: CI is the bottleneck
As codegen velocity rises, validation has to move into generation, not after commit — CI's purpose is trust.
Why it matters. your factory lives or dies on this.
-
Irina Nazarova: skip GitHub runners for private CI
Buildkite and Ubicloud are faster and cheaper than GitHub's per-minute runners once projects scale.
Why it matters. cost control for the factory.
-
steipete on inter-agent chatter
Agent-to-agent messages in the chat stream are getting irritating; harness visibility should collapse to a single line.
Why it matters. it's a UX lesson for your own harness.
-
Hiten Shah open-sourced 16 competitive-intel skills
Positioning, pricing, battlecards, win/loss — each a method for using AI, with a product reveal teased.
Why it matters. relevant to PostMoney intel work and your skill-building generally.
Rails and Ruby
5 bookmarks
-
Ruby AI News: "The Future of Rails"
DHH's keynote versus Dewayne VanHoozer's take, and how Rails conventions are influencing AI tooling and code generation.
Why it matters. it's your core stack's direction in the agentic era.
-
Ruby Weekly 819
Lexxy rich text editor hits 1.0, Herb becomes default HTML+ERB compilation in Rails 8.2, Matz and DHH fireside on Ruby in the AI era.
Why it matters. staying current on Rails.
-
José Valim: make Ruby 10x better
Don't just admire Rust's speed — push Ruby concurrency, type systems, embedded SQLite with replication.
Why it matters. it resonates with your Elixir/Phoenix Gateway work.
-
Evil Martians: SF Ruby conference
A pitch for the SF Ruby Conference, aimed at helping new builders choose Ruby/Rails as AI enables more app building.
Why it matters. ecosystem pulse check — possibly a trip worth taking.
-
GitHub async merge API goes GA
The only merge API supporting stacked PRs, designed for automations — submit with PUT, poll with GET.
Why it matters. your agents merge PRs; this is the API they'd use.
Self-hosted and privacy infra
3 bookmarks
-
Kaunta
Privacy-focused, self-hosted web analytics as a single Rust binary — no cookies, respects Do Not Track, with an MCP endpoint for agents.
Why it matters. analytics for your products without surveillance capitalism; the MCP endpoint is the differentiator you'd care about.
-
ytfast
A native, audio-only YouTube Music player for Omarchy, written in Rust with egui, signed in via browser cookies.
Why it matters. filed under your Omarchy cloud desktop goal.
-
Cloudflare Artifacts open beta
Every application and agent gets its own repository you can manage programmatically; integrates with Workers Previews.
Why it matters. an infra primitive for agent-built apps.
Takes and tools worth a skim
7 bookmarks
-
Karpathy: deciphering LLM outputs
We'll spend more time deciphering model outputs — use controlled language like ASD-STE100, diagrams, explainer videos.
Why it matters. practical craft from someone you trust.
-
sam lessin: no moat in AI agents
With near-zero switching costs, the only remaining moat is user setup cost — winner-take-some, not winner-take-all.
Why it matters. a strategic lens for your agent products.
-
Tavus: Griffin passes the video Turing test
48% of live interlocutors mistook Griffin for a real human — the first "Human Interaction Model," with safety reviews before release.
Why it matters. what's possible in video agents, plus the safety questions.
-
Nan Yu: "One True SaaS Layout"
A product design opinion: there is only one true SaaS layout, and it represents peak usability.
Why it matters. a quick opinionated read from a strong product mind.
-
Ivan Morgillo: Claude Code skill for beat-synced promo reels
A free MIT-licensed Claude Code skill that creates beat-synced promo reels.
Why it matters. content automation, relevant to your social presence systems.
-
Exec: AI sales training platform
AI roleplays, call scoring, and coaching for revenue teams and leadership development.
Why it matters. likely competitive intel or curiosity about AI roleplay products.
-
sam lessin on X
Couldn't be summarized — X blocked the crawl, so only the post link is available.
Why it matters. you saved two lessin posts the same day; the other was his "no moat in AI agents" thread, so this one likely continues that line of thinking.