this is a month late, but a month ago i shipped my first end to end app with ai🤩. i'd never set up an app, a database or sign-in before. by the end i had all of it running on @nextjs , @convex, @clerk and @vercel. as a longtime jira user, i had so much fun building a project management tool from scratch with the help of ai a month is a long time in ai, but here's what i'd still tell anyone starting today 🧵 what worked: 💬 you can ask ai to teach you, not just code for you. my most used prompts: → "show me how to do this" → "what's the best way to do this?" → "what am i missing?" 🧱 you don't need to know the tools before you start. i learned how a database works, how google sign-in works and how deploys work by building them 🤝 use more than one model. i switched between @openai's sol models and @claudeai opus 5.5 📎 make your demo fun. mine is the project management tool dunder mifflin would actually use (ofc scott's tots tuition fund is in there) what i missed: 🫠 test it as a brand-new user. i forgot to build the "create a new workspace" flow because my demo data was always there 📱 check it on a phone. i built it all on desktop and never thought about mobile 🎨 define your styling upfront. ai won't keep it consistent across pages on its own. set your colors, fonts and components early, or every page ends up looking slightly different 🧠 ai forgets between chats. i built this over a few days, and each new chat lost what i'd already decided. now i keep my requirements in a file on github, so every chat starts from the same page
Clerk on X
Praise for Clerk's agent features — agent auth, value-first signup, the agent feedback tool, .well-known support — leads the talk, with easy integration close behind. Complaints are few and concrete: duplicate sign-in emails, Claude refusing hosted sign-in, and surprise per-workspace costs.
First written 27 Oct · what changed · from 52 posts by 43 people
Tracking posts that say “@clerk”
Track this for your marketPosts per day, last 30 days
as of Nov 3, 11:01 UTCWhat changed
as of 27 Oct · checked once a monthFirst written 27 Oct. When what people say changes, it shows here.
Tone· 19 posts from 17 people· written 27 Oct
How do people feel about Clerk?
The largest theme is praise for Clerk's agent features, 5 people. Easy integration and plain, unqualified enthusiasm are level behind it at 4 people each, and 2 people come away fed up.
The agent work draws the largest group. They like that the plugin serves the /.well-known files automatically — "a nice touch," one says, the bit they usually double check by hand — and that Clerk now handles auth for letting users connect their agents without building that layer. Value-first agent signup is called perfect for agent UX, and the agent feedback tool gets the fullest accounts: when an agent fails it usually retries or quietly gives up, so letting it report the snag is "the closest thing to user interviews for agent traffic"; one builder says the best bug reports on their MCP server came from the agent itself, flagging a timestamp offset and a wrong base URL.
Easy integration is level with plain enthusiasm. Builders describe one-click tooling wiring the SDK and middleware, provisioning a Clerk app and bringing in the keys, Expo support for SDK 58, and a first-timer's first end-to-end app where Google sign-in ended up "super easy to use and set up." Others just say grab it, love it, and that it treats them well.
Two people single out the design as clean and well laid out, one praising the bento layout. The disappointment is small but blunt: a duplicate sign-in email bug, and a flagged account that locked someone out.
- Easy to integrate6
- Praise for agent features5
- Enthusiastic4
- Likes the design2
- Disappointed and fed up2
See 5 of 19 postsHide the posts
@v0 now has one-click integrations for auth, email, search, and databases. I asked it to add @clerk auth to my app. It wired up the SDK and middleware, provisioned a Clerk app, and brought in the keys. Then I published to Vercel. Prompt to live app with auth 👇🏼
The agent feedback tool is the underrated one here. When an agent fails against your API it usually just retries or quietly gives up, so you never hear why. Letting it report the snag directly is the closest thing to user interviews for agent traffic.
Value first signup later is perfect for agent UX
worth grabbing @clerk link
5 of the 19 posts behind this answer, one per person. The other 14, and every new one as it’s posted, come with tracking. Get early access
Pain points· 15 posts from 13 people· written 27 Oct
What do people complain about with Clerk?
Sign-in bugs are the biggest complaint, named by 3 people: duplicate sign-in emails, and the sign-in page keeping them in the same Gmail account instead of adding another. Coding agents refusing Clerk's hosted sign-in comes next at 2 people, level with pricing and the agent-surface complaints.
The duplicate emails are the most-reported snag: one builder says it sends duplicate emails to sign in at a bad time, and someone who has had it "reported several times" reads it as an effect double triggering the sign-in rather than a tracked bug, asking whether the built-in components or a custom flow are in use. Another says the sign-in page won't let them add a second Gmail account, logging them back into the first.
The agent friction is specific: Claude refused to run a normal auth flow against hosted Clerk, and another builder confirms it with a Clerk test user fixture and in an isolated Convex environment, noting Codex complies. On the agent surfaces, a connect button is cut off by the bottom of the dialog in the Claude iPhone app, and one MCP environment reports "Checks unavailable".
Pricing shows up as surprise: the Pro plan billed per workspace rather than per application, plus impersonation and a BAA that appear as add-ons after launch. Two more caution that agent auth gets hand-waved until production breaks, and that an agent configuring access also decides who gets it.
- Sign-in bugs3
- AI agents refuse Clerk sign-in3
- Pricing and add-on costs2
- Agent and MCP rough edges2
- Agent auth needs human oversight2
- Locked out of their apps2
See 5 of 15 postsHide the posts
Potential evidence of @AnthropicAI AB testing Fable 5.5: I found a hidden system prompt instruction to 'never sign into hosted services except on localhost' after Opus 5.5 refused to run a normal auth flow via @clerk
Just realized @clerk pro plan is on workspace level, not application level 🤯
We have had this reported several times but have never been able to track it to a bug. It's typically an effect double triggering the sign in. Are you using the built in components or a custom auth flow?
@clerk on the sign in page, if i have multiple gmail accounts it is not allowing me to add another even if i click on add account from the profile on top right. It keeps logging me back into same gmail account i logged in before with.
@clerk the button to connect your clerk account on Claude app (iPhone) gets cut off by the bottom of the dialog and there’s no way to get to it. :(
5 of the 15 posts behind this answer, one per person. The other 10, and every new one as it’s posted, come with tracking. Get early access
Alternatives· 3 posts· written 27 Oct
What are the alternatives to Clerk?
Only 3 people weigh Clerk against something else, and each names a different path: falling back to plain email login, writing auth yourself, or moving an app over to Clerk. The three are level.
The comparison talk is thin and split three ways, one person each. A flagged account cut one builder off from Clerk OAuth along with other third-party logins, and they ended up saying they would switch to email login. Another says Claude, left to set up auth on its own, never reaches for anything with a signup screen — it just writes its own login and moves on — which makes the case for Clerk as the thing you choose rather than what an agent picks.
On the other side, one founder announces a project going live with little visibly different, along with auth migrated to Clerk. No one weighs Clerk against another named auth provider in these posts; the choice is between Clerk, plain email login and hand-written auth.
- Plain email login1
- Rolling your own auth1
- Migrated to Clerk1
See the 3 postsHide the posts
All the issue i have faced with my account being flagged. 1. Cant log into third-party Oauth apps like @convex , @clerk , @firecrawl etc. 2. @vercel deploys stopped since it was also github Oauth. 3. Maybe even affected my Capy account will talk to the devs. 4. My Profiles, Repos, Prs, everything is now hidden from public view. Now I cant even do anything. Gonna contact all the teams & switch to email login now fuck this shit.
makes sense. when i let claude set up auth on its own it never reaches for anything with a signup screen, it just writes its own login and moves on
Updated Fitme Project, just went live! Not much visible changes, mostly internal. ah yeah, and migrated auth to @clerk
Questions· 9 posts from 8 people· written 27 Oct
What do people ask about Clerk?
The most-asked question is where an agent's autonomy stops and a human steps in, 2 people: which ops tasks run end-to-end without the dashboard, and whether an agent-created signup reaches a person. Questions about picking Clerk at all, the agent feedback tool and agent-created production setup are level behind it.
The boundary question is asked in a few forms: which ops tasks an agent can now complete end-to-end without a human opening the Clerk dashboard, whether a value-first signup leads to the human behind the agent or to an agent inbox, and whether impersonation sessions would be agent-initiated or created for a human to open. One reply in the thread settles on Clerk as a data source inside a broader workflow rather than a system an agent runs autonomously.
People also ask whether to pick Clerk at all — one founder building a SaaS puts Clerk beside Better Auth, Auth0 and Supabase Auth and asks which and why; another asks why use Clerk when building your own auth is so easy.
The agent feedback questions are about mechanics: how feedback gets back to the builder, and whether it shows which tool schemas agents keep misusing, since the hardest bugs were tools the agent called wrong without an error. On agent-created production setup, people ask whether it waits for DNS verification, and how the human review stays useful without sending someone back through the dashboard steps the MCP server replaced.
- Where does the agent stop and a human step in?3
- Which auth provider, and why Clerk?2
- How does the agent feedback tool work?2
- How does agent-created production setup work?2
See 5 of 9 postsHide the posts
which ops tasks can an agent now complete end-to-end without a human opening the Clerk dashboard? that boundary is the useful detail behind “run businesses.”
Do you find the ask leads to the human behind the agent signing up or an agent inbox like the grok inbox being used?
helpful distinction. so the near-term value is Clerk as a data source inside a broader workflow, not a system an agent runs autonomously. would impersonation sessions be agent-initiated, or created for a human to open?
I’m currently building my SaaS 🚀 Confused about which auth provider to choose 🤔 1. @clerk 2. @better_auth 3. @auth0 4. @supabase Auth Which one would you recommend and WHY? Also, what other auth solutions am I missing? 👀 Would love to hear your experiences!
@Hiteshdotcom sir clerk kyu use kiya jaata hai aaj ke date pe bhi? if we can so easily create our own auth system, then why link is powered by @clerk ?
5 of the 9 posts behind this answer, one per person. The other 4, and every new one as it’s posted, come with tracking. Get early access
Wishes· 9 posts from 6 people· written 27 Oct
What do people want from Clerk?
The biggest wish is a ready-made tenant isolation test, 2 people: proof that signing in doesn't mean a second account can see the first one's records. Every other wish here is held by one person.
The isolation check is framed as a gap in what people test: sign in with a second account in a private window and make sure it can't see the first one's workspace, because sign-in working doesn't prove each user only sees their own data; the suggested starter is two organizations where a user from one cannot read the other's records even by changing an ID in a request.
The rest are one-person asks, and most concern agents. One wants tool-level permissions tested, since getting user auth right doesn't mean the tool boundaries are right. Another wants the agent's setup kept intact when a human claims the project so nobody onboards twice, and also asks for stable organization IDs and an explicit "more results" indicator so agents don't act on the wrong tenant or mistake the first page for the full list. A read-only organization access report — members, roles, pending invitations, recent changes with timestamps — would let an agent answer who has access without permission to change memberships. Smaller asks: markdown 404s that keep their status and point to nearby valid routes, and metered billing, called a strange omission when everyone is shopping their own agent products. One person holds out for free-tier passkeys before making them their only auth.
- Test tenant isolation2
- Free passkeys1
- Test agent tool permissions1
- Metered billing1
- Keep agent setup when a human takes over1
- Stable org IDs and clear pagination1
See 5 of 9 postsHide the posts
nice build. i'd add one to what i missed. sign up with a second account in a private window and make sure it can't see the first one's workspace. sign-in working doesn't prove each user only sees their own data
A useful starter test would create two organizations and verify that a user from one cannot read the other’s records, even by changing an ID in a request. Getting sign-in working is the first step; tenant isolation needs its own check.
the "can or cannot be accessed by agents" boundary is the important bit here. i'd test a valid user credential against a tool the agent isn't supposed to invoke — getting user auth right doesn't automatically mean the tool-level permissions are right.
The handoff is the part I'd obsess over... keep the agent's setup intact when the human claims the project. Nobody wants to do onboarding twice 😅
Returning stable organization IDs alongside display names should help agents avoid acting on a similarly named tenant. Clear pagination and an explicit “more results” indicator matter too, so the first page isn’t mistaken for the full organization list.
5 of the 9 posts behind this answer, one per person. The other 4, and every new one as it’s posted, come with tracking. Get early access
In their words
How do people talk about Clerk?
The phrases people use most, counted once per person.
- 1“mcp server”4
- 2“agent feedback”4
- 3“feedback tool”4
- 4“agent feedback tool”3
- 5“auth flow”2
- 6“sign-in working”2
Posts are shown the way X shows them. One of them yours? Here’s why it’s shown, and how to take it off.
Also in Developer authentication
- WorkOS+252%The talk about WorkOS is warmer than it is detailed: people praise the team, ship on its auth, and recommend it by name, while the sharpest criticism is reserved for what it charges per SSO connection — and Clerk is the comparison it keeps getting.203 posts / 7d
- Auth0+7%Praise outweighs complaint: 9 people credit Auth0's setup speed and agent-identity direction, and 7 name it first when asked which auth tool to use.111 posts / 7d
Get every post about Clerk, filtered and labelled
This page is the summary. Tracking gives you the stream behind it for any brand, keyword or @mention: each new pain point, request and question as it’s posted, with the spam and promotions already taken out.