Building useful AI products. iOS Engineer. AI-assisted by default. Previously Staff Engineer @ Bumble.
Santa Cruz de Tenerife, Spain
Joined December 2024
- Tweets2.4K
- Following1.2K
- Followers255
- Likes498
Pinned Tweet
iOS engineer running a real frontier-AI dev setup on production code — parallel agents, MCP servers, sub-agents.
I build tools out of the problems that creates — Air Wisper, SimLease — and post the unglamorous specifics: what breaks, what the models actually do, and why.
Building in public, small numbers and all.
One of the most annoying problems I face since AI coding: identifying dead git worktrees and delete them.
Usually they eat up my Mac storage and my automatic AI agent does not know if they should be deleted or not...
I will build a visual tool to identify them and delete them
I'm now thinking that this is a good candidate for a Claude mod
Working on it now...
Alexis Santos retweeted
This. People keep asking me for my setup. It's really just any harness, multiple agents concurrently, barely any skills, and using adversarial reviews. There's no magic sauce (or source!). The models are great out of the box.
Alexis Santos retweeted
Day 1/
We have optimized the default speed to be ~50% faster across GPT-6 Astra and GPT-6.1 Sol through the subscription across all our products and partners using Sign in With ChatGPT (including OpenCode, Pi, Amp, Devin, ...).
No changes needed on your end and this should be felt within the next two hours.
Life of a Mobile Engineer
- A couple years ago and before... I am waiting for the compiler to finish.
- Now: I'm waiting for my agent to finish.
Been shipping lately. Launched my first small Product Hunt campaign for Breathe Now for iOS.
Specially useful in case you use Fable 5.1 and need a break to stop and breathe before killing it for being annoying🔪🔪🔪🔪
producthunt.com/products/bre…
Breathe Now: A one-minute reset between demanding moments producthunt.com/products/bre… by @deepfirstsearch
I used OpenAI new Voice mode and defaulted to normal input again after a few tries.
The voice mode is incredible, but I love listening to music while I work... 🎶
AI can make you faster and leave your brain fried.
Before you open the next prompt, take one minute away from the screen. I built Breathe Now for that exact gap: a guided reset when you’re carrying one task into the next.
No account. No subscription.
apps.apple.com/app/apple-sto…
If you want to pick the right model in Claude Code, just don't pick Opus 5.
It is quite an annoying model who never finish a task cleanly, instead stick to Fable in medium or Opus 4.8 xhigh effort.
I'm back to using the Pareto rule for my personal projects and life in general.
Finding the top 2 or 3 things that really matter every week for me to solve allows me relief from a shit ton of small tasks that don't matter.
Try and then come to thank me
When you start a new project or a new app you must visualise your user.
Not millions but only the one. And then try to solve the problem in the best way you would imagine the solution for it.
Optimise for that first one. That's your MVP before growth. Validation of your idea
Started to understand the importance of a good AI model while working with Opus 5.
I can tell it is the most annoying model I use to date.
The only 'fix' is to add a skill to respond in plain-english rather than some AI giberish but still ... many times it just does the latter
I'm going to create some of these but will trigger them on demand rather than scheduled. I do not have unlimited tokens like @bcherny but I truly like the ideas
A weird experiment I've been trying the last few weeks is having Claude take over day-to-day maintenance of our apps. Seeing early signs of life that this might be possible.
The setup is straightforward: we have a Slack channel called proj-claude-maintains-apps. In it, Claude Tag runs a bunch of daily routines across iOS, Android, Desktop, web, CLI, and Agent SDK:
- Crash fuzzer: open the app in a simulator and tap around to find ways to crash it, then root cause and fix the crashes
- Dup unifier: scans the codebase for similar-yet-slightly-divergent abstractions, and puts up PRs to unify them
- Dead-code remover: removes statically unreachable code, and adds logging to suspected dead code to check if it's really dead and if so, remove it the next day
- Abstraction police: fixes leaky abstractions
- a bunch more..
Results have been surprisingly positive. Over the last few weeks, these routines have opened 388 PRs across our repos, 180 of which we merged after Claude Code Review + human review. We're now thinking about how to streamline this to make merging these kinds of mechanical changes easier.
Claude generally gets these PRs right on the first shot, and if it doesn't, we ask Claude to tune its routines so it's better the next day. Sometimes it takes a few days of tuning.
To try a similar workflow, ask Claude Code or Tag, or create some routines directly at claude.ai/code/routines. A few of the actual prompts I used below.
Has anyone experimented with similar workflows?
Read this today. Interesting take, but I don’t fully agree.
AI makes it incredibly easy to make stuff, sure. But taking a photo doesn’t make you a photographer. I think there’s a big difference between producing things and actually building your craft.
a16z.com/the-most-human-tech…
