You can build your app with a coding agent. Here is why most founders should not.
Coding agents in the terminal are the best thing that happened to developers this decade. For a founder with no developer, they are a maze with no map. What the first week looks like with an agent, what it looks like with AppX, and when you should still pick the agent.
Ilyos Olimov · · 5 min read
Every week someone writes to me a version of the same message: "I tried to build my app with Claude Code / Codex for two weeks. I have 300 files and nothing that runs on my phone. Can AppX help?"
I want to be fair to those tools, because I use them. For a developer, a coding agent in the terminal is a superpower. It reads your code, it writes the boring parts, it fixes what the compiler complains about. If you can already build a mobile app, an agent makes you three times faster and you do not need AppX.
This post is for the other person. The founder with an idea, a laptop, and no developer.
Week one with a coding agent, as it actually goes
You open the terminal. The agent asks you to pick a stack. You say "mobile app". It suggests React Native with Expo, which is right. Then the questions start, and every one of them has an answer you do not have.
Which Node version. Whether to use the managed or bare workflow. Expo Router or React Navigation. Do you want TypeScript. What bundle identifier. Which simulator. Should it set up EAS. Do you have an Apple developer account. What is your Android package name.
You answer "whatever you think" and the agent thinks, and now you own a project configured by someone else's defaults, and you cannot tell which parts matter.
Then you want to see it on your phone. This is where most people stop. Metro bundler, a QR code, a tunnel that does not connect on your network, an iOS build that needs Xcode, which needs 40 gigabytes, which needs an update. The agent can describe every step. It cannot do them for you, because they happen on your machine, in your accounts, with your passwords.
Then the context runs out. You come back the next day, open a new session, and the agent has forgotten what it decided yesterday. You paste the summary. It re-reads 300 files. You pay for that in tokens, every morning.
By day ten you have a repository that a developer could finish in a week. You are not a developer. That is the whole point.
What is actually hard here
None of it is the AI writing code. The code is the easy part now. The hard parts are:
- Knowing what is what. Metro, EAS, Expo Go, TestFlight, bundle IDs, provisioning profiles, the three different places a name lives. A developer has a map of this in their head. An agent will explain each term perfectly and still leave you without the map.
- The environment. Versions, simulators, certificates, accounts. The agent cannot own these. You can, but you did not sign up to.
- Keeping the thread. A product is hundreds of small decisions that have to agree with each other. A fresh session every day forgets them.
- Seeing the result. The whole reason you wanted an app is to hold it in your hand. With an agent, that moment is the last step and the most fragile one.
The same week with AppX
You describe the app in a paragraph. The builder asks one or two questions if something is unclear, like which part to build first. It shows you a plan: the screens, the tabs, roughly what the build will cost in credits. You approve.
A few minutes later there is a real React Native app. Not a mock. You open it in the browser preview or on your phone from a link. You tap through it. Then you change things by chat: "make the primary colour green", "add a settings screen", "rename it". Each edit takes under a minute and costs a credit or three.
The environment is ours. The versions, the sandbox, the build, the preview. You never see Metro. The decisions are kept with the project, so Tuesday's edit knows what Monday decided. And the code is yours: download it on any plan, hand it to a developer whenever you want one.
Median first build in our measurement across twelve apps: 6.3 minutes. A five-screen app costs around 30 credits, about a dollar and a half. New accounts get 100 credits free.
When you should still pick the agent
I would be lying if I said never.
- You are a developer, or you have one. The agent plus your skills beats any builder.
- Your app is mostly backend: payments, real-time, heavy integrations. AppX gives you an honest front end and a place to plug a backend in; it will not write your bank integration.
- You want full control of every dependency from day one.
For everyone else: the agent gives you the pieces of an app and a map you cannot read. AppX gives you the app, and the map is our problem.
A fair way to decide
Spend one evening with each. With the agent, measure the time until the app runs on your phone. With AppX, measure the same thing. Then ask which one you would want to be doing again next week, when you need to change something.
That second question is the real one. Building an app once is a weekend. Changing it is the rest of your life as a founder.
Keep going
More practical ways to move your app forward.
- How to describe your app to AI: a prompt template
Write a clear AI app-builder prompt with a copyable template, a practical example, acceptance checks, and focused follow-up edits.
- How to turn a client brief into a mobile app prototype
Turn a client brief into a mobile app prototype with a copyable AppX prompt, a practical test checklist, and a focused client-feedback workflow.
- Three emails a solo founder sends, and what came back
AppX now sends three short emails from me: when your first app is ready, when you built one and did not come back, and when you signed up and never started. 56 sent so far. Here is what they say and the mistake that means I cannot tell what came back.
- The 5-backtick fence: stopping prompt injection via Markdown
We ship user input into an LLM architect prompt every turn. A reviewer caught that a 4-backtick string from the user could close our 3-backtick fence and run arbitrary instructions on the model. The fix is one weird Markdown trick.
- Investigators before builders: a 2-phase LLM workflow
Most teams dispatch coding agents straight to the build. We run a read-only investigator phase first — and the findings have rewritten our plan in 30-40% of cases. Here's the shape, the brief format, and what to do when findings contradict the plan.
Ready to try your idea?
Start with a clear description. AppX will help you turn it into an app you can test.