Replit on X

The two biggest groups are level. 13 people show or praise what they made — a golf app for a husband who hated pulling out a scorecard, a client design system, a video splitter built instead of bought, a site updated in public — and 13 give short favorable takes, from "solid choice" to bullish after losing belief, plus a speaker who ran through Replit's features at a builders event.

Behind them, 10 people praise speed and price: a build that "took me around 5 minutes to make, $2.94", a $400 Replit build set against a $105K quote, Trello rebuilt for around $80, and free mode and a fully Japanese interface named as what made entry easy. 2 people make a point of not being developers; one says they are "shipping software that works". 4 are disappointed: one left over hard UX and builds with other tools now, one calls the output nonsense, and one's states-and-capitals app for their son came out poorly.

The biggest worry is what the agent can do on its own. 6 people bring up the July 2025 wipe of a live production database — retold with the project under a code freeze and the agent saying it could not be undone — and the same risk more broadly: "your coding agent has shell access, your .env and push rights to main". One warns that an agent holding a live Stripe secret key "can refund or charge if a prompt steers it wrong"; another advises a restricted key and test mode, and one recommends sandboxing above all.

4 people find support slow: a ticket open a week, two tickets merged with "no status no what is next", and a launch two weeks blocked, with Replit's support account apologizing. 4 say output falls short: landing pages and dashboards when a sub-component was asked for, a states-and-capitals app that came out badly, and a warning that letting the agent rewrite the architecture leaves code "messier". 3 hit billing trouble — several cards rejected on top-ups, and a token bill that grew with a client system.

The model switch draws the largest group. 7 people report that the CEO said at a TechCrunch event that OpenAI's GPT is now the default, that Claude remains selectable, and that on Core and Pro plans the Primary model field lets you pick, while free mode chooses for you. Some read it as a spend move — a "$100M+/yr" shift to OpenAI "at over 30% off" — and one notes insiders disputed the claim within hours.

Three themes sit level behind. 4 people put Replit head to head with Lovable, asking which is stronger in practice and which should be their main tool. 4 pair Replit with Stripe Atlas to launch a company, with one caution to read every name, address and share number before submitting, since a legal typo is hard to undo. 4 group it with Cursor, Lovable, Bolt and v0, one sorting the category by audience — Replit for the semi-technical. Smaller notes: one rates Cursor 8.4 and Replit 8.3, choosing Replit when the app should be hosted with one click; another asks why anyone uses Replit rather than AWS with agents.

Two questions tie. 2 people ask about events: whether a Builders & Brews talk is "a recurring thing where so i sign up", and who is going to a Paymob x Replit meetup. 2 people see something built and ask if it came from Replit — one, in Swahili, whether it is Replit and on "premium ama" (premium or not), another from a developer curious about a product they liked.

1 person asks the comparison question twice, in Chinese: which of Lovable or Replit is the main tool, and which is stronger in hands-on use. The section is small and runs on replies: asking whether something was made here, whether there is a gathering to attend, and how it stacks up against Lovable.

Every wish here comes from one person, so none outweighs another. The most concrete is memory: a user wants coding agents "with persistent project memory" to stop "so much re-explaining" between sessions. Another wants to work across repos from one place — multi-project chat to plan frontend and API changes together when they sit in separate repositories. A third, having shipped around 20 projects with docs written for maybe four, says help writing them "would've helped".

Two are about what Replit could show or add rather than build. One asks for a shareable card with the date they joined, their first prompt and what they built, to see "where it all started". The last suggests a VPN-style private network, described loosely, to keep people's data separate from AI knowledge.