:)

vibing with models, prompt me

[Ole Lehmann @itsolelehmann]

HOW TO GET CLAUDE CODE TO "VIBEHACK" ANY SMART DEVICE IN YOUR HOUSE

(its actually much easier than i thought):

1. open Claude Code on a laptop that's on your home wifi

2. ask it to "scan my network, list every device with its make and model, and tell me all the possibilities of what you could fix, edit, or control on each one"

3. pick one and let it cook (it'll ask you if it needs you to press a button or plug something in)

for example...one guy added AirPlay to their 15-year-old TV.

another turned an old Kindle into a family dashboard with the day's schedule and incoming deliveries.

another got their washer and dryer to announce out loud when the laundry's done.

there's way more possible than you'd think, so let Claude show you what your own stuff can do.

just run it on your own network, and read the commands before you approve them.

[Linus ✦ Ekenstam @LinusEkenstam]

Want to vibe code your TV?

Here is a collection of posts with everything you need to know

Bookmark and come back to this tonight, and re-claim your TV

https://t.co/G5be4RwB89

[Mitchell @MitcheIl]

hands down one of the best articles i’ve read on how to make money with X's API:

[@levelsio @levelsio]

Here is the prompt to build your own music MP3 player and internet radio for Amazon Alexa:

[ 🎶 Jukebox for Alexa ] by @levelsio

Build me a private Alexa skill that plays my own music from a cloud bucket, plus internet radio

I want my Amazon Echo to play my own MP3s instead of being forced to use Amazon Music or Spotify. Build a private Alexa skill for this. Do as much as possible yourself: use CLIs and APIs, deploy and test it yourself, and only ask me for things only a person can do (logging in, clicking Allow, testing on the Echo).

Setup
- My music is in an S3-compatible bucket (AWS S3 or Cloudflare R2) under a music/ prefix. Each folder under it is a genre/playlist (music/Coffee Jazz/) or an artist/album (music/Artist/Album/).
- I'll create an access key for the bucket and give it to you. Keep it in a config file and never print it. The skill only needs read access; the upload tools need write access.
- My Echo and my Amazon developer account use the same Amazon login.

Build
1. An Alexa-hosted (Node.js) custom skill with invocation name "jukebox" (use two words if Amazon rejects one) and the AudioPlayer interface enabled.
2. Voice model: a PlayIntent with an AMAZON.SearchQuery slot ("play {query}", "put on {query}", "tune in to {query}", "shuffle {query}"), a ShuffleAllIntent, a WhatsPlayingIntent, a ListStationsIntent, and all the built-in AudioPlayer intents (pause, resume, next, previous, loop on/off, shuffle on/off, start over).
3. Library: list everything under music/ with ListObjectsV2 (follow the continuation token), build folders from the key prefixes, keep only .mp3/.m4a/.aac files, and cache the result for 5 minutes. Match spoken phrases to folder and file names loosely: ignore track numbers and spacing ("ama piano" = amapiano), turn number words into digits ("thirteen beaches" = 13 Beaches), and handle genre aliases (dnb, drum n bass, hiphop, lo fi).
4. Playback: give Alexa a presigned HTTPS GET URL for each file. Make it valid for at least 12 hours so long mixes and pause/resume don't expire mid-play. Keep the bucket private. Store the play queue in the AudioPlayer token (folder, shuffle seed, position) so there's no database, and enqueue the next track on PlaybackNearlyFinished. Folders shuffle and loop. A folder whose files all start with track numbers is an album and plays in order. A single song carries on with the rest of its folder.
5. Radio: a stations list of HTTPS MP3/AAC/HLS streams, with good underground electronic stations (Rinse FM, NTS, Bassdrive, Intergalactic FM, Kiosk Radio, Refuge Worldwide, and so on) plus news (BBC World Service, NPR). Test every URL with curl first and drop the plain-HTTP ones, because Alexa refuses them. Support "play rinse fm", "drum and bass radio", "techno radio" and "the news", and make next/previous switch stations. Live streams drop and Alexa treats a drop as the end, so re-enqueue the same stream with a new token on PlaybackNearlyFinished, and retry once on PlaybackFailed.
6. Log one line per request (type, intent, token, offset, error) so playback problems show up in CloudWatch.

Deploy without the web console
- Install the ASK CLI (npm i -g ask-cli). If ask configure won't work in your shell, run ask util generate-lwa-tokens --no-browser, have me paste the one-time code, and write ~/.ask/cli_config yourself. Get the vendor ID from GET https://t.co/8eS787CaBt.
- The hosted skill's code is a git repo (find it with ask smapi get-alexa-hosted-skill-metadata). Push to master to deploy, with ask util git-credentials-helper as the git credential helper. The helper needs ask-resources.json and .ask/ask-states.json next to the repo. If pushes start failing with 403, clear the cached macOS keychain credential.
- The hosted runtime may run an old Node version, so pin an S3 SDK version that supports it, or sign URLs with a small SigV4 library.
- Upload the voice model with ask smapi set-interaction-model, check the skill manifest lists AUDIO_PLAYER, and test with ask smapi simulate-skill. The web console's simulator can't play audio, so a "problem with the requested skill's response" error there is expected.

When it works, tell me how to add Alexa app routines (When: voice "coffee jazz" → Action: Customized "ask jukebox to play coffee jazz") so I can just say "Alexa, coffee jazz". And how to play internet radio stations I like.

[J.B. @VibeMarketer_]

recommended reading this week:

@ryangrowthh - Distribution 101: building a story people want to follow, choosing the right channels, and packaging your content so it actually gets noticed

@thebeautyofsaas - Regret Minimization: three questions for decisions you've been putting off

@eyad_khrais - Grow Without Headcount: using your hiring plan to find workflows worth automating, redesigning the work, and measuring whether the same team can handle more volume

@danhockenmaier - How to tell if a startup has a moat: why fast growth alone isn't enough, and how to spot advantages that get stronger as a business grows

@a16z - This Week in Charts: Agents Denied: where AI fits into the shopping journey, why customers still want control over purchases, and how concentrated sales have become on TikTok Shop

@thisissethsblog - The race to the bottom feels inevitable: how chasing traffic can erode the trust that made a business worth following, and why holding your standards matters

@emollick - The Dot and the Swarm: an essay from earlier this month on why managing AI work may require less coordination than expected, while people still need to decide what's worth doing

essential reading: Ryan's distribution guide + Eyad on growing without headcount + BOSS on regret minimization

[kepano @kepano]

This is Pele, my new rendering library for diagrams and charts, compatible with the latest Mermaid syntax.

- 100-500× faster than mermaid.js
- 6× smaller bundle
- No dependencies
- Easy to theme
- Responsive to device width

In pursuit of making Obsidian insanely fast and cohesively designed, I always found the Mermaid library to be too big, too slow, and too hard to theme. Pele will be the new renderer in the next version of Obsidian, but you can start using it now.

Pele [ˈpɛlɛ] is named after the goddess of volcanoes and fire. It's open source under the MIT license. Try the playground!

[Dara A. @daradoescode]

adding this alone to my agents.md has 2x'd the amt I ship

[Jason ✨👾SaaStr.Ai✨ Lemkin @jasonlk]

Just another $100k deal 10K brought in today

He didn't close this one himself. The humans did.

But he found it, followed up, sent the contract, got it signed, prepared the invoice, and got the payment in today.

All by himself.

10K our AI VP Revenue. That kid's good.

[Ish Verduzco @ishverduzco]

I should have done this from the start...

But I just setup my 24/7 Community Manager bot with Notion so that everyone at the company can see the most relevant tweets, daily digests, & sentiment recaps

Easy way to scale yourself as a social/community marketer

@bot 🤝 @NotionHQ

[️ @blankspeaker]

Getting Grok Blocked by sites that dont like agents?

Here is a way around that.

[Noah Kagan @noahkagan]

Another one bites the dust in our War on Expensive Software. 😵

Zapier.

An old friend. We’ve used it for 10 years.

We paid for 123 automations. 63 hadn’t run in a month. 2 did 88% of the work.

The rest was form-to-Slack and HubSpot-to-Sheets plumbing.

Claude rebuilt 1 of them. Everything else moved into tools we already pay for.

Zapier: $4,788 a year → $0.

Steal the audit. Export your Zapier task history and give Claude or ChatGPT this:

“List every automation, its owner, and runs in the last 30 days. Flag anything that hasn’t run. Show which few do most of the work. For each one, name a tool we already pay for that could replace it. Give me a cancel plan with dates.”

Bye bye, old friend. 👋

[Matt Pocock @mattpocockuk]

I think I'm addicted to getting agents to build state machines

Every time I see an agent has built a state machine (or even just a reducer) I can feel a huge bump in the robustness

And I know that future updates will be trivial and testable

Would recommend for frontend

[Sarvagya Kulshreshtha @sarvagya_kul]

same. works all the time

[Jason ✨👾SaaStr.Ai✨ Lemkin @jasonlk]

More than half our invoices were paid late, and we didn’t know it.

At month-end close, 10K, our AI VP of Revenue, reported the real number: 56% of SaaStr invoices were paid late while humans ran collections.

Part-time contractors without Salesforce access billed deals by hand a day or two a week, and follow-up was inconsistent.

After about a quarter of 10K running quote-to-cash:

-> Invoices paid late: 56% to 8%
-> Average days overdue: 17 to 6
-> Longest overdue invoice: 51 days to 12

What 10K does after every signature:

#1. Reads the full 20-page contract, because split payments and odd terms are scattered through it

#2. Moves the deal to Closed Won in Salesforce and adds missing contacts

#3. Creates the invoice in bill dot com off the actual contract terms

#4. Sends it within minutes to the signer, our deal contact, the business owner, and a finance contact it finds through enrichment if the contract doesn’t list one

#5. Checks every open invoice daily and follows up weekly

#6. Pulls a human in at 7 days past due

Along the way it found Bill reminders we’d paid for and never turned on, plus 2 customers still being billed for a product we stopped supporting 6 years ago. Then it took over commissions in about 30 minutes.

It ran on 3 supervised deals before going autonomous on the 4th, with one wrong invoice since.

Collections works well with agents because every input already sits in a system of record: the contract, Bill, QuickBooks. Very little judgment required.

At our Dreamforce booth, more people asked about this than anything else. Far fewer companies run agentic collections than run AI SDRs, and almost every B2B company has the same problem we did.

[Micky @Rasmic]

I don’t get why guys like this get so pissed lol

> I wanted to test out REA repo
> I did
> it worked, app works pretty good
> i don’t care to make it look pretty
> i provide the repo
> mad there isn’t a download button

God bless you sir… go get some air

[Vincent van der Meulen @vinvan]

how to use @bot for engineering and get a ton done:

1. create a (cursor) project for every big workstream. i have ten, such as a “mcp project” and a “bugs project”

2. make sure every project spins up cloud agents with /poteto-mode

3. in grok bot, make an eng bot whose sole job is routing requests to the right project.

4. now voice call (!) your primary bot and braindump all your tasks.

your tasks will automatically end up at the correct projects, and result in sweet prs :)

you get all the benefits of projects + pstack, but don’t have to think about anything!

[Lenny Rachitsky @lennysan]

Here it is: Benchmarks to measure a given company's moat

Image 1: KPIs to look at, by potential moat.
Image 2: What "good" and "great" looks like.

Note, these are rough and meant to be interpreted directionally, because there is so much variance by industry and business model. For a deeper exploration, see links in the PDF here: https://t.co/kbxEnO5fzU

And for a lot more context and advice, don't miss @danhockenmaier's full post: https://t.co/jMdyoPpeF6

[lauren @poteto]

perhaps one of the most durable parts of pstack are its engineering principles: very concise skills (~10-20 loc) that distill down good practices into its core.

as models get smarter, i expect most of pstack skills to compress into short principles or workflows (a generic list of steps) and that i will be able to delete a few of them.

how it works is that /poteto-mode contains a short index of available principles and tells agents "In your reply, name each principle that shaped a decision and the specific choice it changed." this is why you'll often see a Principles Named or Used section at the end of the agent's response

many of these principles were extracted from my sessions correcting agents since the start of this year. some are inspired by problems i see people talk about, like Attack the Premise, inspired by @VictorTaelin's post https://t.co/qTGVfoKSyD

you can also use the principles as skills if you want greater control, by invoking /principle-<name>

https://t.co/wuTWsflVN3

[WebDevCody @webdevcody]

@Rasmic you should test out REA on TablePlus, then archive the repo since using a GUI to connect to your db is pretty much pointless when we have claude.

[ILIAS ISM @illyism]

bro listed 13 SaaS bills at the bottom like you need Vercel + Supabase + Inngest + Clerk + S3 + PostHog to send an email loop

1 Hetzner (€3.79/mo) + bun:sqlite + SES + 6 MCP tools

0 admin UI, 0 React dashboards, 0 SaaS / SQL wrappers

Claude is the UI, here's the entire prompt 👇

Build a single-file headless newsletter engine (`server.ts`) running on Bun + `bun:sqlite` (WAL mode) + AWS SES on a €4/mo Hetzner VPS. 0 external SaaS dependencies, 0 Admin UI, the entire control plane is a remote MCP server.

1. SQLite Schema (`db.sqlite`, WAL + busy_timeout):
- subscribers(id, email UNIQUE, status['pending'|'active'|'bounced'|'complained'|'unsubscribed'], tier['free'|'paid'], source, attrs JSON, token UNIQUE, created_at)
- posts(id, slug UNIQUE, title, mdx_free, mdx_paid, sent_at)
- send_queue(id, post_id, subscriber_id, subject, status['queued'|'sent'|'failed'], attempts, next_retry_at, UNIQUE(post_id, subscriber_id))
- events(id, subscriber_id, post_id, type['delivered'|'opened'|'clicked'|'bounced'|'complained'], meta JSON, ts)

2. Public HTTP Routes (`Bun.serve`):
- GET / & GET /p/:slug -> server-rendered HTML web archive + RSS feed at /rss.xml (unlock `mdx_paid` if signed cookie is paid)
- GET /media/:file -> serve optimized images directly from local `/var/data/media/`
- POST /subscribe -> rate-limit by IP in memory, reject disposable domains, upsert subscriber, send magic-link opt-in via SES
- GET /magic/:token -> activate subscriber + set HMAC signed session cookie, redirect to /
- POST /u/:token & GET /u/:token -> RFC 8058 1-click instant unsubscribe
- POST /webhooks/ses -> verify AWS SNS signature, process Bounce (Hard -> 'bounced', Soft -> ignore), Complaint (-> 'complained'), Delivery/Open/Click -> insert into `events`
- POST /webhooks/stripe -> verify signature, flip `tier = 'paid' | 'free'`

3. Background Worker + Email Compiler (in-process loop, 0 Redis/Inngest):
- Compile MDX to Outlook-safe table HTML via `react-email/render` and verify `Buffer.byteLength(html) < 102000` (Gmail 102KB clip limit).
- Every 1s, pull up to 50 due rows from `send_queue`, send concurrently via SES `SendEmail` with `List-Unsubscribe` + `List-Unsubscribe-Post: List-Unsubscribe=One-Click` headers, and exponential backoff on 429/5xx.

4. Remote MCP Server (`POST /mcp` with Bearer token):
- `sql_query(sql)`: read-only SQL over SQLite (replaces all analytics dashboards, cohort charts, and segment builders)
- `save_post(slug, title, mdx_free, mdx_paid)`: upsert draft, return Gmail byte-size check + web preview link
- `upload_media(filename, base64)`: resize/convert to WebP via `sharp`, save to disk, return URL
- `send_test(slug, to_email)`: send rendered preview email
- `send_post(slug, where_sql, ab_subjects?)`: populate `send_queue` for active subscribers matching `where_sql` (e.g. "tier = 'paid' AND json_extract(attrs, '$.topic') = 'ai'")
- `mutate_sql(sql, params)`: tag, import CSV, or update subscribers/automations directly