Adam Bulmer 🦀

Adam Bulmer 🦀

@mintuz.com

Independent AI Product Engineer Contractor, ex-Monzo, BBC & Booking.com. Web & App Engineer. Personal favourites SwiftUI and React. 💪 workoutplan.app 🏡 belongings.app 🤓 swiftforjs.dev ⏱️ timetrials.app ☀️ morningclub.app 👶🏻 wakewise.app

1.3K followers1K following1K posts

Pinned

If I were hiring today, I’d ask how you scale yourself with agents. By 2027, defining work, delegating it and checking the results will matter more than recalling framework trivia. Should interviews test the engineer alone or the engineer with their agents?

The best mental model I have for Jev right now: I could extract attributes from an image of my face, and it would deterministically tell me how ugly I am using structured data.

You can ask it three different types of questions: Single option, scale, boolean What type of ugly am I? Butt ugly How ugly am I? 99% Am I ugly? Yes

The case for Flutter and React Native feels less obvious to me than it used to. Historically, the big win was avoiding duplicate implementation across iOS and Android. But code is becoming dramatically cheaper to write with AI.

View full thread

Liquid Glass is a good example. Apple changed the look and feel of iOS over a year ago but Flutter developers have to wait for the framework to catch up. I imagine we will be looking at similar timelines for iPhone Duo support.

As the cost of writing code approaches zero, platform fidelity becomes a much bigger differentiator. Maybe "write once, run everywhere" just isn't as valuable as it used to be.

The case for Flutter and React Native feels less obvious to me than it used to. Historically, the big win was avoiding duplicate implementation across iOS and Android. But code is becoming dramatically cheaper to write with AI.

Fable 5.1 is incredible for maintaining design systems. I've been on teams that built and deployed design systems used across many apps. That work collectively probably took thousands of developer hours. Today I had Fable build a complete Storybook and component library with zero regressions.

Some of my agent skills were over a year old, written for models I no longer use. When models change, your skills need retesting too. I tested them with Promptfoo and Warps Skill Doctor.

One of the easiest ways to spot an AI-generated video is to just mute it. Without the audio distracting you, the unnatural mouth movements, awkward posture and strange facial expressions become obvious.

I’ve been building on iOS with an App Intents-first architecture for a while. It’s not well documented and articles advocate patterns like MVVM or VIPER. With Apple’s latest Siri integrations, everything has just clicked into place. Best practices are useful, but first principles win always.

I never use plan mode with Codex. Plan mode exists in Claude because it’s over-eager and jumps straight to implementation too quickly. Codex actually takes the time to understand the codebase first and makes holistic changes based on its findings.

If you wondered… X is still full of plebs and apparently I’m a bad actor. I mean I never did drama but come on… I don’t need to be called out on it.

Claude Design is cool but GPT Image Gen 2 is even better. My workflow 1. Sign up to Mobbin 2. Use the Mobbin MCP in Codex and prompt what you’re building 3. Ask Codex to use the official GPT Image Gen skill to generate 4 UI mockups 4. Pick the best design and feed it back into Codex to build

I’ve got Codex /goal on my Raspberry Pi executed from my iPhone triaging 250 backlog items refining them to create a cohesive story on how to deliver my app from 0 to 1. It’s like having a product owner in my pocket and I’m the VP of Product 😵‍💫🤩

What matters is customer outcomes. Too much energy spent on how. Sprint Planning, Retros, Cycle Time, velocity, number of weeks per sprints etc. All bollocks. The number of hours I’ve spent in quarterly/ halve planning for the plan to be derailed 2 weeks in is staggering. It’s all busy work.

We should be minimising the time between ideation and having something in front of the customer. Planning in quarterly or halves contradicts this premise.

Let the business define the goals for the quarter “release product X”, “increase Y” and let the teams figure out how to achieve that but not 6 months in advance. Plan, ideate, implement should take no longer than 2 weeks. Allows teams to learn and pivot when appropriate.

I can’t imagine using AI to write code without TDD. Tests aren’t just safety nets anymore. They are the steering wheel on the race car.

Happy Pokemon day. 2026 Christmas Day will be my personal 30th Pokemon day. I remember receiving the game from my grandad and having no idea what it was and spending weeks trying to defeat Brock because I only had Charmander on my team 😂

The human touch will be the thing that drives downloads on the AppStore. I’ve seen a bunch of competitor apps appear recently and they all look the same. What makes your app stand out amongst the vibe coded slop will be the difference between success and failure.

I’ve been using AI for things I consider myself an expert at and using time saved to level up other skills using my brain like SEO, Graphic design and marketing.

MRR is a funny metric. I’ve been measuring my apps success in bills paid. My energy bill is now paid by my apps every month 😆 Next goal: the mortgage.

I created a new Claude Code skill this evening for iOS App Store market research. Previously this was a custom MCP server that I had to spin up locally on my machine. SKILL.md files don't just have to be coding conventions and how-tos. They can allow your agents to call third parties.

Claude Long Term Memory spec-kit, bmad, task-master, beads, beans. It doesn't matter which one you pick. They all do the same thing Just use one and get familiar with it - don't get stuck in that loop of switching tools every week because someone on Bluesky recommended it.

AI has turned my iPhone into a dev workstation. A user reported a bug for an app I’ve built. I shared a screenshot + high-level architecture via Codex on iOS. It found the issue, fixed it, and deployed - entirely from my phone. 12 months ago I was just doomscrolling and reading memes 😂

LLMs, autonomous agents, and “vibe coding” reward system thinkers. Models think locally. Slop happens when humans do too. Your value is the bigger picture and to guide them along the path.

My 2026 Indie dev goals - Reach £1000 MRR - Release a new unannounced app I’ve been building for the last month. - Start posting on TikTok behind the scenes of my indie app development process.