Signal House

You can describe an entire app to Claude and ship it in 4 hours. Then you add SMS and suddenly it's 2015 again. CPaaS wasn't built for how we build now. We fixed that.

@theo respectfully waiting to see which one of the 322 ships the feature 'send the user a text when the deploy fails'. that's the one where every founder learns about a2p the hard way.

322 apps already deployed on my cloud 👀

step 47 omitted from the deck: your otp text never delivered, the carrier returned 200, and the afternoon is now a crash course in delivery receipt webhooks. x.com/levelsio/statu…

@elusivedev08 @TechCrushHQ huge for maternal health. one thing that bites later: a2p approval timing. if a reminder text is queued for the day the campaign clears review, the reminder hits next week. plan for that window.

Reliability is everything in maternal health. Today’s MamaCheck update: successfully transitioned to Twilio for SMS/OTPs and refined our codebase structure. Building a better experience for mothers, one commit at a time. 🩺#techcrushaluminibuildathon @TechCrushHQ#MamaCheck

@rubenboonz this is the right comparison. twilio in 2010 worked because the rails were ready, not because the api was clever. voice ai is one cycle behind on its rails. whoever wins is whoever makes the boring infra pieces invisible first.

vapi raised at $500m this month to build horizontal voice ai infra. amazon ring shipped ai-answered doorbell calls the same week. voice ai right now looks the way telephony looked in 2010 when twilio launched. one horizontal api layer locks up the infrastructure money. then a different layer of companies sits on top, one per vertical, building the actual workflow. amazon does doorbells. someone does hotels. someone does shopify phone support. someone does dentists. the horizontal layer becomes plumbing and the vertical workflow becomes the product. founders trying to be horizontal voice infra in 2026 are showing up to a fight that already had $500m raised against them last week. founders who pick one vertical and ship the workflow inside it are early.

@shoplokuli service booking rails is a real gap. fwiw the rails for messaging part is still kinda broken too. the boring sms layer is where booking confirmations live and die.

Stripe built rails for payments. Twilio built rails for messaging. We're building rails for service booking. lokuli.com/mcp — the infrastructure layer for AI-automated booking.

@Rajath_DB @gaganghotra_ phone numbers over browser tabs is the right take. sub 800ms ttfb is table stakes. real boss fight is the user texting the number back and your bot having a working sms layer too.

@gaganghotra_ yeah it's wild for browsing but desktop only is the catch. the real unlock for voice agents is phone numbers, not browser tabs. been building voxa around that gap, langgraph + twilio with sub 800ms ttfb. different use case but same underlying tech tbh

Have u tried Gemini Live in Chrome (desktop)?

@frank__rosh @launch_llama 4 lines vs 50 is the right vibe. add a note in the readme about what happens when the number gets flagged for spam mid call and the agent quietly stops working. that one bites at scale.

@launch_llama Building Patter — open-source SDK that gives any AI agent a real phone number. Handles Twilio carrier auth, STT, TTS, barge-in, AMD. About 4 lines of Python replaces ~50 of glue code. github.com/PatterAI/Patter

𝗦𝗵𝗼𝘄 𝘂𝘀 𝘄𝗵𝗮𝘁 𝘆𝗼𝘂'𝗿𝗲 𝗯𝘂𝗶𝗹𝗱𝗶𝗻𝗴.
 Drop your product in the comments.
 If we love it, 𝗪𝗲'𝗹𝗹 𝗹𝗶𝘀𝘁 𝗶𝘁 𝗼𝗻 𝗟𝗮𝘂𝗻𝗰𝗵 𝗟𝗹𝗮𝗺𝗮 𝗗𝗶𝗿𝗲𝗰𝘁𝗼𝗿𝗶𝗲𝘀. 🦙✨

@mmabouelnasr @W33Z_global the shared otp inbox problem is so real. half the staging accounts at any startup are tied to one phone that left two quarters ago. neat fix.

@W33Z_global I got tired of chasing OTP codes over Slack. Vendor accounts tied to one person. They leave. Access gone. Built Inflovy — team-owned inboxes for shared accounts, staging flows, and verification emails. One dashboard, whole team has access. Just shipped → inflovy.com

Happy mid week Founders What are you building? Let’s send traffic!

@TommyBez85 love auth.md. now do messaging.md. agent hits a send endpoint, gets a 451, reads messaging.md, learns it needs an a2p campaign, opt in proof, sender id. half the agent ecosystem stalls on this one.

WorkOS just shipped auth.md — an open protocol that tells agents how to register for a service. Agent hits your API → 401 → reads auth.md → registers itself. No sign-up forms, no human in the loop. Two flows: • Agent-verified (automated, IdP vouches) • User-claimed (OTP, human confirms) Cloudflare + Firecrawl are first partners. The missing piece for "let my agent handle that" on any web service. workos.com/auth-md

@fin_ai the unglamorous half of ai support transform: the channel the answer goes back on. half the time the customer wants a text, not a chat reload. the channel is the experience.

Most customer service teams are experimenting with AI. Very few are transforming because of it. The gap between those two things is larger than it looks; and right now, it's where the most consequential decisions in support are being made. Moving from a successful pilot to full production is where most teams struggle, so on May 27 we're hosting a session for the people doing exactly this work. Join Declan Ivory, Forrester guest speaker Max Ball, and AppFolio's Guneet Singh, XMP for a webinar that will cover: ➤ Why AI agents are changing the economics of support delivery, not just the speed ➤ What's actually shifted in AI reliability, scale, and quality over the last 18 months ➤ Why the jump from pilot to production is harder than most teams expect ➤ How leading teams sequence for ROI without trading away customer experience ➤ What moving from experiments to production at scale actually looks like We'll also walk through how AppFolio took their AI resolution rate from 17% to 73%, while holding a CX score of 93%. That kind of result doesn't happen by accident. Register at the link in the replies.

@claudeai an opt in flow for an sms list that doesnt feel like the 2009 web. mostly making sure no checkbox is pre checked, and the keyword is not the word stop.

What are you making with Claude Design?

Tinkering, prototyping, and seeing what happens with Claude Design:

@cursor_ai composer in the sdk is fun. give the agent an sms send tool and you immediately learn that the carrier doesnt care your prompt was elegant. send tools are the boring half of agent infra.

With the Cursor SDK, you can build your own agents with Composer 2.5. It's now available in Python and TypeScript. This long weekend, Composer usage is 90% off in the SDK. We're excited to see what you build!

@twilio this is sick. flag for the prod team: ConversationRelay is the easy part. getting the post call sms confirmation through 10dlc on time is where the demo turns into a ticket.

Here's how to build a real-time AI voice assistant with Twilio ConversationRelay + Google Gemini Flash 👇

ok now do single prompt for the user side. 'register my sms campaign so my agent can text otp and confirmations.' that's the harder one. machine payments shipped this year. machine messaging still pending review. x.com/stripe/status/…

you forgot the line item where each one also has to text the user. twelve agents, twelve a2p campaigns, twelve approvals still pending. the boring sms layer is where the workflow actually breaks. x.com/HubSpot/status…

@ev_strike opt-in by default for notifications is the boring right answer. the only thing better than 'opt in to discord notifications' is letting users pick the channel. text the urgent ones, push the routine ones, email the receipts. respect the inbox, win the user.

We just shipped a hot fix for discord notifications. Now you have to opt in to notifications you want.

one prompt to your agent for machine payments. now do the harder one: one prompt to your agent for sms otp on the user side of the same transaction. cpaas approval queues are why the agent economy still needs humans to babysit signups. x.com/stripe/status/…

every agent ships with its own login, its own billing, its own legal review. somehow none of them ship with a working notification channel. the receipt to the human is still a text. boring layer, biggest gap. x.com/HubSpot/status…

@SahilExec @CaptainInsightX queues per channel + sms-specific retry policy is the move. sms costs money per attempt so dumb infinite retry is a billing event waiting to happen. cap retries by destination, route around the noisy carrier, then alert ops on dlr failure clusters.

Exactly the right instinct Queues decouple the trigger from the delivery. Each channel - push, email, SMS - fails and retries independently. One slow SMS provider should never hold push notifications hostage. That's the core idea behind every serious notification system at scale.

A startup builds their notification system like this: One function handles everything: 1. Query user preferences from DB 2. Format the message 3. Send push notification 4. Send email 5. Send SMS 6. Save to notifications table 7. Update unread count Works perfectly at 1,000 users. At 500,000 users one slow SMS provider makes every notification take 45 seconds. Users getting push notifications for events that happened an hour ago. The entire system is a single point of failure. What are the architectural mistakes here and how would you redesign this to handle 10 million users?

@cursor_ai 3.5 looking great. one feature we'd love to see in 3.6: cursor agent gets an sms when a long task finishes so we stop alt-tabbing every 90 seconds. the inbox to humans is still a number.

Download Cursor 3.5 to try these new features: cursor.com/download

You can now create and manage automations in the same workspace as your agents. Automations are now available in the Agents Window. For the next 7 days, all agent runs for newly created automations are 50% off.