Apps Built With Codex That Have Revenue Evidence

ProvenStartups indexes 10 businesses whose breakdowns name OpenAI's Codex. Seven disclose a monthly figure, from $20,200/mo to $250K+/mo, and three were verified by something other than the founder's word. Ten is a thin file — we hold 49 Claude Code cases and 44 Cursor cases in the same index — so read this as ten receipts, not a trend. Five of the ten name Claude Code as well.

Contents

• ·How many apps built with Codex have revenue evidence?
• ·What do the ten Codex apps actually earn?
• ·How did the best-documented cases get their first dollars?
• ·What patterns repeat across the ten?
• ·What would we copy, and what would we skip?
• ·Frequently asked questions

How many apps built with Codex have revenue evidence?

Ten cases, graded 3 ✅ verified, 4 🗣 founder-reported, 2 📎 creator-relayed and 1 🔮 unproven. Nine of the ten sell something today. Against 49 Claude Code and 44 Cursor cases in the same 406-project index, this file is small enough that one entry moves every percentage in it.

The grades matter more than the dollar signs. ✅ means a dashboard, store payout or tracking tool was shown on camera. 🗣 means the operator said it and nobody checked. 📎 means a breakdown channel passed on a figure it never audited. 🔮 means there is no revenue at all — only a product.

The split is even: four consumer apps, four SaaS products, one platform plugin, one AI service. Six are solo operations. Three sit in Tier 1 (copy this now), five in Tier 2, two in Tier 3. More teardowns live in our built with AI hub.

What do the ten Codex apps actually earn?

The top of the file is $250K+ MRR and the bottom is zero, with six cases in a $20K–$50K/mo band that is where this cohort really lives. Read the grade column before the money column: the two largest numbers are also the two nobody outside the company confirmed.

Do not average this table. Clarvo reports an annual figure, and Super Demo's founder is a repeat operator with a team, so $250K+ MRR is no solo benchmark. Dialogue's $50K/mo is an estimate relayed from Sensor Tower, not a payout screen. Of the seven monthly figures the middle one is $30K/mo — and even that mixes Profit AI's app-only MRR with LipPal's all-sources month.

How did the best-documented cases get their first dollars?

All three ✅ cases sold into an audience that existed before the product did: a competitor's users, a live-stream chat, a consulting client list. None of them started from a blank market. That is the most repeatable thing in this file, and it has nothing to do with which model wrote the code.

How did Profit AI turn a spreadsheet into $30K MRR?

Its founder is non-technical and had maintained a profit-tracking spreadsheet for Shopify clients for years. He exported it as a CSV, handed it to an AI editor, said build it, then spent his time catching every calculation the AI got wrong. $147,000 has come through the Shopify partner dashboard since December.

The rest is less flattering: 122 app users, 38–39 paying, and 47 installs against 43 uninstalls in the window shown on camera. That leak is covered by a services line worth just under $40K MRR, and the paid plan fell from $5,000/mo to $800/mo.

How did BridgeMind reach $20,200 MRR on no ad spend?

By building the product on camera every day until the audience became the customer list. Every paying customer came from streaming. The founder's previous startup burned thousands a week on ads at $20 per keyword click; this one has spent nothing and sits at $242,964 ARR with 13,700+ Discord members.

He also turns down $10,000-a-video sponsorships to keep the streams clean — a real cost, deliberately paid, and the part most people copying "build in public" skip.

How did Prayer Lock reach $21K/mo by copying a competitor?

It did not invent demand. Prayer Lock copied Prayscreen, an existing faith-based screen-time blocker, and out-executed it: 58K downloads in six months, $21K/mo. Writing the app took three days. The work was finding one short-form video format that converted, running it to volume, and squeezing onboarding.

What patterns repeat across the ten?

Three things: tools get stacked, distribution decides the outcome, and the same niche shows up twice. Nobody here used a single AI tool, nobody won because of the tool, and two of the four consumer apps in the file are phone blockers sold to the same religious community.

That duplication is the most instructive fact on the page. Prayer Lock at $21K/mo ✅ and PrayLock at $40K/mo 📎 are near-identical — lock the phone, pray to unlock — and Prayer Lock is itself a copy of a third app. A community with its own vocabulary and channels markets for you, and it supports more than one winner.

On stacking: of these ten, five also name Claude Code, five name ChatGPT, three name Cursor, three name Gemini and one names Lovable. "Built with Codex" means Codex was in the toolchain, not alone in it. The tool-by-tool picture sits in AI coding tools by revenue; the neighbouring files are apps built with Claude Code and apps built with Cursor.

One absence needs a warning label: we file 38 documented failures across the index and none names Codex. That is not a safety record — failed projects rarely get a teardown, so any tool with ten cases looks clean.

What would we copy, and what would we skip?

We would copy the move that keeps working here: sell software to the niche you just left. We would skip the unmonetized personal app and the anonymous portfolio play — not because they are bad, but because neither produces a number you can plan against.

LipPal AI is the cleanest version: its 17-year-old founder was an Amazon KDP self-publisher who stopped competing with self-publishers and started selling to them. The software cleared $30,894.74 in April 2026, ten times what his own books ever made. Profit AI is the same trade with a spreadsheet, and Clarvo says the quiet part: everyone can build the software now, so the moat is what to build and how to price it.

The other two teach something as well. The anonymous 3D web studio turned a burner account into 5 client projects and 1 job offer — a good outcome with no revenue you can size. Thinking Space is a local-first health app its builder made for himself while ill, the only 🔮 here and never sold. Portfolio pieces, not plans.

Starting from the tool instead of the business? Read OpenAI's Codex documentation, then pick the audience before the stack. More in our built with AI hub.

Frequently asked questions

How much do apps built with Codex make?

Seven of the ten indexed cases publish a monthly figure. They run from $20,200/mo to $250K+/mo, and the middle of that set is $30K/mo. Businesses that agree to a filmed teardown are the ones with numbers worth showing, and three of the ten publish no monthly figure at all.

Were these apps built with Codex alone?

No. Of the ten, five also name Claude Code, five name ChatGPT, three name Cursor, three name Gemini and one names Lovable. Founders switch tools inside a project and rarely attribute a result to one. The index records which tools a breakdown mentions, not which tool produced which file.

What do the evidence grades mean?

✅ means a third party saw the number: a dashboard, a store payout, a tracking tool shown on camera. 🗣 means the founder stated it. 📎 means a creator relayed someone else's figure without auditing it. 🔮 means no revenue figure exists. The grade tells you how much weight a claim carries, not whether the business suits you.

Why are there no failed Codex projects on this list?

Because failures rarely get filmed. The index holds 38 documented cautionary tales and none names Codex, which reflects who agrees to be interviewed rather than how the tool performs. With ten cases, survivorship bias is strong. Treat a clean file as missing data, not as a high success rate.