Saurav

I've been doing consultation work for years, building multiple SaaS products in parallel. At some point I realized I was spending more time on infrastructure plumbing than on the actual products. So I built Wacht.

@ayushagarwal @dodopayments @Cloudflare Love this Ayush. Genuinely curious though, got anything in the pipeline for the "no one knows my product exists" problem? Asking for a friend 🫠

we 5x'd our AI agent traffic in the last 60 days at @dodopayments. didn't rewrite a single page. all we did was give every page a markdown twin that AI agents can actually read. same URL. browsers get HTML. AI agents get structured markdown. picked at the edge. @cloudflare has a button for this. it works. but we wanted full control over what AI agents see, how pages are converted, and which crawlers get what. a one-click solution doesn't give you that. so we built dualmark. open source. framework adapters for @astrodotbuild, @nextjs, and cloudflare workers. a CLI that scores any page's AI-readability on a 0-125 scale. a CI check for github actions. 12 page-type converters. apache 2.0. zero telemetry. public spec. 30 seconds to install. give it a try - link in the first reply.

@arseniycodes @katedeyneka Hey, what's the full stack look like for Superlog? Curious how it's all holding up so far, and if you've had any other Clerk issues recently beyond the reliability one

@katedeyneka Clerk has been pretty good! they’ve been having some reliability issues recently though, we had to postpone a demo once (even openrouter was down for them one day)

my tech stack for building apps, may 2026 save this for your next build 🔖

@bentossell Plans are auditable, prompts aren't. You can't go back and see why a prompt did what it did, but you can step through a plan and figure out where it broke. I think that's what the shift is about

plans replaced prompt engineering right? we do context engineering and talk 'normally' to agents but the plan is just what 'how to prompt X model' is

@EbubeEvan The agent flow you mapped out is the shape most people land on once they start building seriously. Curious how the LangChain + Gemini combo has been holding up, especially around persistence and state across conversations. That's usually where the abstraction starts leaking

Hot take: building your own auth is a waste of time if you want to ship fast. Finally building an AI customer care agent I’ve had in mind for months AI scaffolded most of it. Starting backend-first. Stack: Next.js, TS, Node, Redis, PG, LangChain, Gemini Building in public 🚀

Every team building AI products right now is gluing together their own infrastructure. The model APIs are solved. Everything around them, auth, multi-tenancy, isolated execution, persistence, isn't. We're at the pre-Rails moment for AI infrastructure.

If a system feels complex before you've even finished engineering it, that's usually a signal the engineering went wrong somewhere. Systems that grew from smaller working ones tend to stay coherent even when they get large. The mess comes from manufacturing complexity upfront instead of letting it emerge.

Gall’s law: “a complex system that works is invariably found to have evolved from a simple system that worked.” also very relevant to agent systems. most teams are trying to jump straight to autonomous complexity before they have evals, observability, or feedback loops in place. 2026 is the year of evals

Just shipped Wacht Bench. AI coding tools are only as good as the context they have on your stack. Default behavior is generic patterns that look right and break in production. Wacht Bench is a CLI and skills pack that gives Cursor, Claude, Codex, and other agents actual knowledge of how Wacht apps should be built. Skills cover Next.js, React Router, TanStack Router, React SPA, Rust/JS backend sdks, API Gateway, webhooks, notifications, agents, testing, Multi tenancy. MCP server included. Install: npx skills add wacht-platform/bench Source: github.com/wacht-platform… AI tools should know your stack as well as your team does.

@emil406p @Taniyatweets_ Curious what tips you toward Clerk or WorkOS over Supabase auth on those projects. Building Wacht (wacht.dev) in this space and trying to map how folks actually pick auth tooling.

@Taniyatweets_ Big fan of Neon for projects where I want to use something like Clerk or WrokOS for auth, but Supabase is really good if I just want everything in one platform. Firebase storage is also my go-to for storage

as a dev , which one do you prefer ?

Public beta now. Most of the source is open at github.com/wacht-platform under an OSS license. If you've been gluing together five different vendors to ship every SaaS, this is for you. wacht.dev

What's inside: Full OAuth 2.1 provider, plugs into MCP servers SAML SSO on every plan including free Agent platform with isolated microVMs and durable task queues Prebuilt UI for every surface, sign-in to webhooks to agent conversations, ready to embed

I've been doing consultation work for years, building multiple SaaS products in parallel. At some point I realized I was spending more time on infrastructure plumbing than on the actual products. So I built Wacht.