iiinova — where your digital self keeps evolving

If agents make writing code cheap, the expensive parts are proving it works and maintaining the codebase(s). One codebase for every platform instead of one per platform continues to make sense. That's why my bet is Flutter and Compose Multiplatform.

000

Visual fidelity to the design spec: Flutter came in first, Compose Multiplatform was a close second, React Native was third, and both native builds last. The two native apps failed the same way — off-design buttons and tabs, an extra border on every card. On one run.

101

I measured the apps, not just the dev. One run each, emulator and simulator. All six: 244–421 ms cold start, zero crashes in 15k random taps. Size is the exception — iOS: 0.3 MB native vs 6.6 Flutter, 10.5 CMP. Android flips: Flutter 16.8 MB smallest; RN's 45 MB largest.

100

"Native is cheaper now" is the wrong takeaway. What speeds an agent up is checking its own work in milliseconds instead of booting a simulator for every change. That's an architecture choice, not a language choice — and you can make it in Flutter or CMP too.

100

In my tests, the native builds exposed eight real bugs each — about 4 per platform — compared to an average of 1 for React Native and 2 for Flutter and Compose Multiplatform.

100

Which makes me think "who writes the code" was never the interesting variable. Multiple codebases equal multiple chances to be wrong. That equals higher maintenance costs.

100

Ryan Edge @chimon.bsky.social · 23/09/2026

I built the same four-screen app six ways — native, React Native, Flutter, Compose Multiplatform — with agents doing the work. Some teams dismiss native's cost too quickly. In my tests, it was the slowest by a wide margin — up to 3.5× the agent minutes.

100

If app stores die, it’ll be because they’re the wrong primitive—but I don’t think a messaging-based app store is the right solution.

000

Ryan Edge @chimon.bsky.social · 23/09/2026

"Working with an AI agent sometimes feels like working with the fastest, dumbest coworker you’ve ever had."

000

Ryan Edge @chimon.bsky.social · 17/09/2026

If your UI only has a dark mode theme, it's probably not accessible for a large percentage of your user base.

001

Don't use Shopify migrating away from React Native as an opportunity to dunk on RN because it's not your tool of choice. You're just stoking the fire.

000

Native is now the future of mobile at Shopify (2026) - Shopify

The lesson isn't Swift and Kotlin. It's agent-addressable architecture: headless logic, CLI-driven state, checkpointed review. Swap the target for anything, same result — the tell that the target didn't move it. (Disclosure: from a Flutter Engineer) shopify.engineering/back-to-native

shopify.engineering

Coding agents changed what it costs to build mobile apps twice. Here’s why Shopify is moving from React Native back to Swift and Kotlin.

010

Every Helix checkpoint needs a human nod + two adversarial reviews. Agents scale; review attention doesn't, so doubling codebases doubles review load. By how much per feature? Three numbers: headless vs simulator cycle time, review-minutes/feature, first-pass rate by language.

100

Agents erase RN's human-time taxes: debugging, bridging, upgrades. That cuts both ways: if two codebases are cheap, bridging is cheap too. Caveat: models write better Swift/Kotlin than bridge-glue. Native keeps: platform proximity, first-party tooling — thinner than billed.

100

The 12-week Shop rebuild has the same problem. Helix reads the RN codebase to plan checkpoints. The RN app is the spec. So the number measures how cheap rewrites got, not how cheap building two apps from ambiguous requirements got.

100

Read their fast-feedback section. The unlock is decoupling business logic from UI and exposing it headlessly via CLI, so agents iterate in milliseconds instead of babysitting simulators. That's agent-addressable architecture, and it was fully available in React Native.

101

Shopify is going back to native from React Native. But they changed two things at once: the language, and a new agent-addressable architecture. Two-variable experiment, one-variable conclusion — the headline will read "native won," and the evidence doesn't isolate that.

100

creolestudiosnewsletter.substack.com

Jun 28 — Spec Coding Is Runway creolestudiosnewsletter.substack.com/p/spec-codi... A short-written spec — what to build and what counts as done — saves startups money before AI starts building, this argues. Without one, AI makes confident wrong guesses you pay to fix twice.

Spec Coding Is Runway

Rework quietly eats startup runway. Spec-drive development helps founders, developers, and AI tools build from the same definition of done.

010

The Cost YAGNI Was Never About

Jun 25 — The Cost YAGNI Was Never About newsletter.kentbeck.com/p/the-cost-... "You aren't gonna need it" was never about saving typing effort, even though free AI-generated code makes that reading tempting. It's about two costs: locking into a guessed design early, paying before you must.

newsletter.kentbeck.com

If you think YAGNI is about saving effort, cheap generation should retire it. It doesn't. Here's why.

100

How to Write an Effective Software Design Document

Jun 24 — How to Write an Effective Software Design Document refactoringenglish.com/excerpts/wr... A walkthrough of what belongs in a design doc: goals, non-goals, diagrams, security, open questions unresolved. It's a test for what to include: how costly is it if the decision turns out wrong?

refactoringenglish.com

Best practices for writing a design doc based on my experience working as a developer at Google and Microsoft.

100

The New Software Lifecycle | AddyOsmani.com

Jun 16 — The New Software Lifecycle addyosmani.com/blog/new-sd... A Google white paper argues most AI coding agent failures trace to their setup, not the model underneath. Writing code is basically solved, it says; the bottleneck shifted to clear specs and verifying what got built.

addyosmani.com

I co-wrote a Google whitepaper about how AI is changing the software lifecycle. I'm not going to summarize the whole thing. Instead, here are the handful of ...

100

AI demands more engineering discipline. Not less

Jun 15 — AI Demands More Engineering Discipline. Not Less charitydotwtf.substack.com/p/ai-demand... As AI makes code cheap to regenerate, teams need more discipline, not less, this argues. Fast feedback loops and treating code as disposable, not sacred, are what let teams benefit.

charity.wtf

If you lived through the shift from handcrafted server pets to immutable infrastructure, you should sense something oddly familiar about what's happening now.

100

Why AI hasn’t replaced software engineers, and won’t

Jun 11 — Why AI Hasn't Replaced Software Engineers and Won't www.normaltech.ai/p/why-ai-ha... Digging into major layoffs blamed on AI, this finds cost-cutting behind most, not AI. Its argument: writing code was never the bottleneck — deciding what to build and owning what ships are.

normaltech.ai

Coding agents as normal technology

100

Loop Engineering | AddyOsmani.com

Jun 7 — Loop Engineering addyosmani.com/blog/loop-e... Instead of typing prompts one at a time, people are building loops: systems that write their own prompts, run on a schedule, and use a second AI to check the first's work. Staying engaged still matters, or results get worse.

addyosmani.com

You don't really need to be good at prompting anymore. The thing to get good at is the loop that does the prompting for you. It's five building blocks plus s...

100

Jun 7 — A New Era for Software Testing antirez.com/news/168 AI-written code now beats typical hand-written code, even if it still lags the best. The bigger idea: give an AI agent a checklist and let it act as your QA tester, running checks you'd otherwise skip.

100

“We’re accumulating code faster than we are accumulating trust.” Sometimes a phrase just hits.

Jun 2 — Trust Factory newsletter.kentbeck.com/p/trust-fac... Beck argues writing code fast doesn't matter if trust doesn't grow at the same pace. Old team habits like testing, pairing, and frequent small releases build trust as a side effect; coding alone with AI strips most of that out.

newsletter.kentbeck.com

Trust Factory

100