The Client Sent a Paragraph. Show Them an App Before the Next Call.
Give AppX the words your client used. Get a concrete app preview to discuss, test, and refine before a full build consumes the week.
Make your app idea real.
Practical guides for turning a clear brief into an app you can test and own.
How to describe your app so AI builds the right one
The gap between a great AI-built app and a mediocre one is almost never the idea — it's the describing. The four moves that collapse a vague sentence into your exact app.
What AppX actually feels like: sentence to phone in 30 seconds
A second-by-second walkthrough of using AppX — empty box, one sentence, watch it build, scan a code, and your idea is running in your hand. The experience is the product.
You have an idea and no code. The wall is gone.
For thirty years, turning an idea into software meant learning to code, hiring, or finding a technical co-founder. AI removed that wall — and the skill that replaced coding is one you probably already have.
I turn sentences into apps. Here's what it cost to get here.
The honest founder story behind AppX — the doubt, the pivots, the nights spent fixing the unglamorous two seconds at the end of a forty-step pipeline. Not the architecture. The human part.
The native-build path: getting AppX apps to TestFlight without writing native code
Expo Go is a preview surface, not a release artifact. Building the path to .ipa and .aab via EAS — what's shipped, what's still rough, and what we won't do.
What we're adding to AppX in Q3 2026
Native iOS/Android builds via EAS, custom domains, multi-user collaboration, expanded template gallery, runtime-error-into-chat, voice input. Honest gates and confidence levels — no date promises.
May 2026 at AppX: 18 wins from the persistence-rearchitecture month
The headline: we found a structural bug in agent-turn persistence (the 0ms-zombie), fixed it across five engineering waves, AND launched this engineering blog, AND shipped a dozen smaller polish items.
April-May 2026 at AppX: Forge stabilization, Edit Engine v2, and Bloom-parity chat
Two months of engineering recap. Forge containers stopped getting stuck, Edit Engine v2 made surgical edits the default (+25 pts edit success), Bloom-parity chat closed the visual gap with our closest competitor.
Why every AppX user starts with 150 credits (and what we learned from 350)
Free-trial sizing is pricing physics, not marketing. We dropped from 350 to 150. Conversion held; cost per signup dropped ~55%. The lesson on what the free tier is actually for.
What 80 user app containers on a single 24GB VPS taught us about memory packing
Naive packing = 12 containers per VPS, swap-bound. Tuned packing = ~80 containers, 80-100MB each. Four wins, one ceiling we hit, and the suspend pattern we wish we'd built earlier.
The screenshot that triggered a 5-wave rearchitecture
'Done · 0ms' looked cosmetic. Eight hours of investigation revealed it was a structural bug class. A week of work shipped a 5-wave persistence rearchitecture. The narrative of how a UI screenshot turned into the biggest backend project of the month.
Three things we built and threw away this year
The Railover orchestrator fork, the one-shot bundle server, the V1 chat intent classifier. Each was a real working system in production. Each got deleted when the shape of the problem changed underneath it.
How the AI pipeline evolved: single-shot to plan-then-build to architect+editor
Three architectures in eight months. Each pivot was triggered by failures the prior architecture couldn't fix without adding the next layer. The shape we landed on, and what we think Era 4 looks like.
Why we moved from direct provider SDKs to OpenRouter
Three SDK clients, three streaming parsers, three error shapes. We switched to OpenRouter in 48 hours. The dividends — and the honest tradeoffs we're paying for.
The four breakthroughs that turned AppX from demo to product
Smart retry on critical files, intent detection with word boundaries, planning before generation, and credit holds. None was a single algorithm — each was a structural change to the loop the LLM lives inside.
The first AppX architecture, and why we tore it down
v1 was a web-first React generator wrapped in Vite hosting. It demoed beautifully and failed in production. The honest autopsy — what stayed, what went, and why the demo and the product are different things.
Forge deep-dive: webhook-driven state, zero polling, no drift
Sequel to why-we-built-forge. Polling produces a structural drift window. HMAC-signed webhooks on every sandbox transition take state propagation from 5-10s to ~1ms. Idempotent handler, state-machine guard, replay-on-restart, and one slow reconciler kept as safety belt.
Forge deep-dive: the warm pool that gets a user from chat to live phone preview in under 2 seconds
Sequel to why-we-built-forge. Sub-2-second cold-to-QR is impossible if every request spawns a fresh container. The warm pool primitive: pre-provisioned idle sandboxes, atomic claim, load-aware sizing, ~80 containers per 24GB VPS.
Persistent Metro: 1,750x faster than spawning expo export:embed per request
One-shot bundlers pay cold-start every invocation. Persistent Metro dev server inside each preview container takes file writes to 4ms and warm bundles to 100ms. The pattern generalizes — any incremental build tool with a watch mode wants to be a daemon, not a script.
SIGKILL is the boss fight: design your sweeper as your worst-case process-death handler
SIGTERM gives you cleanup time. SIGKILL doesn't. Deploys, OOM, and crashes all skip your graceful-shutdown handler. Your sweeper IS the SIGKILL handler — and its query shape should test for invariant violations, not row age.
Dice-coefficient fuzzy match for failed SEARCH/REPLACE: turn a useless error into autocorrect
Agentic edit loops bleed on whitespace drift and one-character mismatches in SEARCH/REPLACE blocks. Replace 'SEARCH text not found' with the top-3 similar candidates scored by Dice's coefficient — the LLM autocorrects on the next iteration at the cost of one O(n*m) scan.
Run Kahn's topological sort on every AI-generated plan
AI architects emit multi-file plans that compile, run, and crash because the dependency graph has a cycle nobody looked at. Validate the plan with Kahn's algorithm before you spend tokens generating. Twenty lines of TypeScript catches a class of runtime bugs at near-zero compute cost.
Why we picked Expo Go for AI-generated apps (and rejected native builds)
Native builds = 60s+ per iteration kills the AI feedback loop. Web preview defeats the mobile-app pitch. Expo Go + Metro gives sub-second hot-reload on a real phone — and that latency budget cascaded into every architectural decision downstream.
From sentence to working mobile app: building AppX in 90 days
Origin story — what AppX's first prototype was, the three architectures we tore down to get here, and the lesson we kept relearning: the AI doesn't need more intelligence, it needs more structure around its outputs.
Stop fixing bugs. Start fixing classes.
Every non-trivial bug is one instance of a class. Patching the instance leaves the class alive — and it will mint the next bug. The lens, the patch+class split, the economics, and the trap of fixing classes that don't matter.
Wave-shaped refactors: how to modernize a hot persistence path without freezing development
Big-bang PRs get skimmed. Multi-week branches drift. Neither works for a hot persistence path. The shape that does: 5 independently-shippable waves, shadow-mirror writers, and a kill-switch wave that turns modernization into actual debt reduction.
Inventory before refactor: a case study where 3 investigators rewrote our plan in 30 minutes
Sibling to investigators-before-builders. Here's the concrete case: three parallel read-only investigators caught that only 8 of 32 fields warranted promotion — and turned a new-table refactor into an extend-table refactor, saving two weeks.
Ban raw column references in your sql templates — one rule that kills a schema-drift bug class
Typed query builders catch schema drift everywhere except inside raw sql templates. The string column reference is the hole. Ban it. What's left is enough for every legitimate raw-SQL case — and the bug class becomes structurally impossible.
Why we built Forge: how we run per-user app sandboxes at AppX
E2B, Modal, Daytona, Replit, Bloom — we evaluated them all. None fit the shape of an AI app builder that needs sub-2s phone preview, long-lived Expo dev servers, and 80 containers per VPS. So we built Forge.
Defense-in-depth for stateful AI-agent loops: 4 layers, each catching what the others miss
A frontend clamp papers over a bad column. A backend sweeper catches symptoms not causes. A DB constraint blocks bad writes but doesn't repair existing rows. You need all four — and the order to retrofit them when starting from a broken system.
Stop jamming AI-agent state into a metadata JSON column
Every soft field in a JSONB blob looks like flexibility. By month six it's a defensive ?? walk on the frontend and a JSON_EXTRACT join the planner can't index. Three forces pull fields out of JSON — and the migration shape that needs no freeze.
The 0ms-zombie: a TOCTOU class of bug in AI-agent turn persistence
Two-phase write + process death = a row that lies about its own state. We call it the 0ms-zombie. Defines the bug class formally and ranks four mitigations by structural depth, from frontend clamp to DB CHECK constraint.
Step 0: delete dead code before you refactor (or the LLM will get lost)
Every byte you load into an LLM agent's context is a byte it has to keep track of. Dead code looks like signal — and agents preserve, extend, or mimic it. Our rule: a pure-deletion commit before any structural refactor of a file over 300 LOC.
Investigators before builders: a 2-phase LLM workflow that rewrites 30% of our plans
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.
The file-disjoint rule: how to dispatch 10 LLM coding agents in parallel without merge pain
Parallel coding agents don't bottleneck on compute — they bottleneck on the merge. Two agents touching the same file cost more than one agent doing both edits. The fix is a partitioning rule we wish we'd adopted earlier.
The 5-backtick fence: a mathematically unbypassable defense against 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.