We are now officially entering the slow tech news time of year, when most engineering orgs stop shipping flashy new features and instead put a few cycles into bug fixes and tech debt, before they inevitably get distracted by the holidays and/or discussions of what cool new features they’ll be shipping the following year. What’s more fun: discovering and fixing bugs or eating mashed potatoes? Based on how many bugs are out there, I think we know the answer.

One place that will always prioritize mobile engineering work over mashed potatoes is the Flight Deck. In this edition, we look at why you should stop vibe coding your unit tests, the hidden costs of mobile releases, how to make your Android notifications 100% reliable, and why Apple is paying Google a billion dollars a year to rebuild Siri with Gemini. Not one bit of potato-related news to be seen.

Posts we liked

Meetings and interruptions are still the biggest obstacles for developers, even with AI

Which year will we see lifelike enough AI avatars that they can attend all our Zoom meetings to nod occasionally and say, “nothing from my side” when prompted? 2029? 2027? Until we reach that point, all the time we save via AI-assisted coding (an average of 3.5 hours a week, according to Justin Reock in this post) seems likely to be sunk into meetings and other interruptions that no computer has the charm and people skills to handle.

How to make Android notifications 100% reliable

Does your app send notifications to users, at least occasionally? Probably. Are you confident that those notifications always fire at the right time and work reliably 100% of the time? Maybe, but I can’t say for sure since I’m not you. If you have noticed unreliability or are concerned that some devices may be delivering notifications inconsistently, Nikita Vaizin has advice that has ensured his notifications are always delivered with 100% success.

Stop vibe coding your unit tests

There is a growing sentiment that LLMs are good for unit tests, but Andrew Gallaher argues that this view is not just wrong; it is the complete opposite of reality. Either because developers are often bad at writing unit tests and LLMs learned from their bad unit tests and made them worse, or because of some unknown, ineffable quality (since no one not being paid a billion dollars by OpenAI really knows how LLMs work) LLMs write terrible, brittle tests that will make your colleagues mad at you.

The great software quality collapse: how we normalized catastrophe

The Apple calculator leaked 32GB of RAM. Android 15 launched with 15 critical bugs. Crowdstrike almost ended the world (and also stranded several Runway employees in Portland after a conference). Denis Stetskov looks at how “ship broken, fix later (maybe, if we feel like it)” has become the hallmark of software development over the past several years. This isn’t shipping fast and breaking things; it’s mostly shipping sloppily and breaking things, which isn’t quite as inspiring.

Apple nears $1 billion a year deal to use Google AI for Siri

In a move that may or may not increase your confidence in the future of AI (Apple Intelligence), Apple will soon be using Gemini’s 1.2 trillion parameter model as part of a larger overhaul to make Siri as useful as the other rapidly evolving voice agents that have mostly passed it by. Apple considered using ChatGPT and Claude, but chose Gemini, which is a capper on the positive changes that Gemini's reputation has seen over the course of 2025.

Go figure

On average, mobile engineers spend 5 hours per release on non-productive or low-value tasks — manual steps, coordination issues, approval bottlenecks, and other busywork that doesn't move the product forward.

Assuming a biweekly cadence, five wasted hours every two weeks translates to 130 wasted engineering hours annually. That's more than three full engineering weeks lost to process inefficiencies, and more than three full weeks you could have spent building features that your users care about. And that’s only if you’re wasting five hours on this stuff (37% of teams say they’re spending up to 10). Imagine if you release weekly!

There's a massive opportunity for efficiency gains hiding in plain sight. Read our State of Mobile Release Management to learn which other hidden productivity drains have impacted release management the most in 2025.

Live events we hosted and talks we gave

[Recorded event] Mobile DevEx: the hidden cost of releases

Miss last month’s live discussion where experts from Runway, Sentry, and leading mobile teams unpacked why automation hasn’t solved your release pain, how constant firefighting affects both morale and velocity (see the great software quality collapse article above), and the steps you can take to improve DevEx around releases? Good news! You can watch this excellent conversation any time at the link above.

[Recorded talk] Future-proofing your Android release process

Runway’s very own Keegan Rush gave a talk at droidcon Berlin, connecting the dots between modern Android development and release excellence, showing how forward-thinking teams like Skyscanner have transformed their processes to ship twice as fast while reducing engineering overhead. He discussed concrete infrastructure improvements, automation strategies, and cultural shifts that can position your team for the next generation of Android development.

Runway featured feature

Did you know that Runway streamlines the hotfix process by automating branch creation, versioning, and conflict detection so that you… don’t have to do any of that? Initiate a hotfix release, and Runway will handle the rest, ensuring your team can resolve issues swiftly without losing precious minutes to process-related busywork.

And recently, we added even more info to the rollout notifications that Runway sends to help you determine when you might need to get a hotfix out, including specific event-level details on crashes and exceptions, complementing the many aggregate health metrics we already surface and helping your team triage emerging issues more quickly.

When you enable the corresponding option in the rollout notification settings (enabled by default), each notification will include a bulleted list of new and trending stability issues identified via your stability monitoring integration (Sentry, Crashlytics, etc.), complete with descriptions and occurrence counts.

Events

Hoping to see Runway in person? For the moment, you missed your chance. We’ve reached the end of conference season, just in time for the US government to shut down all air travel and make it impossible for many of us to go anywhere anyway. (Update: It appears a deal was made to reopen air travel at the exact moment I was writing this paragraph.)  

We’ll be back out there next year with both LEGO sets and knowledge of mobile release management to share. Until then, we’ll see you digitally in Runway, on Zoom calls, and in this newsletter.

And so, this 27th edition of the Flight Deck comes to a close. Should you find yourself with free time during Thanksgiving in a couple of weeks (or find yourself with free time on a random Thursday because all your US colleagues are off and drowning themselves in gravy), kick back with any of our previous 26 editions.