
VP of Engineering and Infrastructure at @bitrise and Dad | formerly: Facebook Infra, IBM Cloud, Ustream | Tweets represent my own thoughts
San Francisco, CA
Joined June 2012
- Tweets844
- Following475
- Followers584
- Likes704
ARTEX response: Build kill switches now, encourage your teams to use them.
We live different times. Last Saturday we woke up to the reality that coordinated AI-accelerated attacks can do incredible harm very fast.
It wasn't just Bitrise under attack. Throughout the week i spoke with countless security and threat response teams, they all confirmed seeing the attacks across the board against their systems. They are exhausted and there is no end in sight.
Every day new vulnerabilities and new exploits are coming out. The pace for defense and threat response teams went from relentless to humanly unmanageable.
My wholehearted advice: take some people from your team aside and implement kill switches to any system that contain sensitive information, so if an AI attack swarm gets in, you can pull the plug and stop before it is too late.
This is what helped us at Bitrise to contain and stop the swarm and practically keep our customer's data safe, limiting the impact to very minimal.
It became so easy to find a weak point, one mistake, one mis-configuration, one system that was not updated, or just a zero-day that was released a couple hours ago and got already weaponized.
All it takes is a single entry point. And the swarm will try until they find it.
Now is the time to build and test your response capabilities and shut down systems if a swarm gets somehow in. They will coordinate intentionally to overwhelm SOC (Security Operations Control) response and cross-company collaboration.
We have discovered this with multiple of our customers as we coordinated through our incident response and forensic investigation, the attack came at the same time, independent from each other.
One of the silver linings is how people in the industry responded, we shared all the details about the ARTEX-driven attack publicly and even our competitors came to start collaborating and sharing what they see.
We are in this together. There is something to this and i have never felt sthing like this before.
It's been a rough week. I've never learned this much this fast, never built defense systems this fast (with the help of AI). Truly we entered the time when we are fighting machines with machines, we are scaling our Defense Agent swarms (called PurrGuard and Purrifier after our cat mascott called Purrequest).
We are going to get through this, take care of yourself and stay healthy. 💜
🤖 Made with AI
4
2,260
Last Friday i made a call to shut down all Bitrise services in the middle of a security incident.
ARTEX got in, this is how it went down. Have fun scrolling
bitrise.io/artex-incident-re…
Bitrise was targeted by a sophisticatedly prepared and AI-accelerated cyber attack.
We closed the investigation by now, i am sharing the timeline and what happened, while we are compiling the incident report. Planning to release that as well.
I believe in today’s world of AI-accelerated and executed attacks, time is of the essence. We are sharing findings for security and threat-response teams so they can look for similar activity in their own environments, both retrospectively and through earlier detection or for anyone who wants learn about how such attacks are put together and executed. The patterns, how the attack was conducted, the source, username patterns, user agents, how they try to decoy. Sharing what we saw and learned.
The in addition to the review, i also prepared the package, which includes:
- A technical advisory describing the observed attack behaviors, patterns and detection ideas.
- A CSV indicator list with source IPs, account and username patterns, user agents, observed time windows, confidence levels and usage guidance.
- A STIX 2.1 JSON bundle containing selected indicators and MITRE ATT&CK technique mappings for threat-intelligence tooling.
We learned an incredible amount: how to detect suspicious activity earlier, which assumptions to challenge, and what it takes to rebuild three data centers with thousands of servers from scratch over a weekend (well this was fun exercise, full scale live DR test and we passed).
I’m proud of the team’s work. That does not lessen the disruption this caused our Customers.
And it reminded us how much there is still to learn, even about systems we build and operate every day.
The cat-and-mouse game of detection and evasion does not end. I hope sharing what we observed helps other teams recognize similar activity sooner, question their own assumptions, and strengthen their defenses.
Walkthrough: bitrise.io/blog/post/the-bit…
2
7
347
Xcode projects as readable JSON? Yes please. 🎉
Xcode 27.2 can save your project as JSON instead of project.pbxproj, and our Xcode steps on Bitrise now read and write it.
Friendlier for humans. And for AI agents.
discuss.bitrise.io/t/xcode-s…
1
443
Need your monitoring agent on every dev environment?
We've released Workspace script in Bitrise RDE: shared setup on every boot, before template hooks.
macOS and Linux, including sessions without a template.
docs.bitrise.io/en/bitrise-r…
216
We've released lifecycle hooks in Bitrise RDE.
On create: prepare the machine
On claim: run personal setup
On restore: restart services
Warm pools can do the slow setup before someone needs a machine.
docs.bitrise.io/en/bitrise-r…
1
155
M5 Pro or M6 for iOS builds?
That question kept coming up in discussions, so we put together a benchmark across M2 Pro, M4, M4 Pro, M5 Pro and M6 VMs on Bitrise.
What I was especially curious about: how do the CPU and memory differences translate into actual build times? Particularly with Bazel vs. non-Bazel builds, where the workload can look quite different.
So, beyond sysbench, we built Firefox, Signal, DuckDuckGo, Telegram (Bazel) and XcodeBenchmark across 11 VM configurations.
M6 has great single-thread performance. But I wouldn't choose a build machine from that score alone.
One example: Telegram's clean Debug build took 9m35s on M5 Pro XL and 10m16s on M6 XL. For the Release build, M6 XL was slightly ahead. Same app, different build configuration.
These are different VM shapes, including different core counts and RAM, so this is a comparison of the machines you can actually run your builds on, not an isolated CPU comparison.
For the clean-build tests, we intentionally left build caches out. No fan-out either. I wanted to understand the speed of an individual machine before adding those optimizations.
The results and methodology are here: benchmarks.bitrise.io
How leading teams scale their engineers and their agents.
RBE now available in Bitrise Build Cache: give your Bazel build a whole fleet to play with. 😎
Remote Build Execution on Bitrise runs your build and test actions in parallel across Linux and macOS workers at once. Pair it with Build Cache and build times drop from both directions.
Read how it works: 🔗 eu1.hubs.ly/H0zxxGf0
2
57
Scary fast github actions builds 👻😆 You have been warned! 🧛♂️
Which night would break your team? 🦇
Night 1: a build queued for 2h 51m.
Night 2: a runner drops at 84%. It passes on re-run #10.
Night 3: 3:14am, and the iOS build has been running for 6h 41m.
We made a short horror film out of the CI problems mobile teams tell us about.
Night five is the fix: Build Hub, one line in your GitHub Actions workflow.
Start your 14-day free trial: 🔗 eu1.hubs.ly/H0zvbHz0 🔮
1
43
Before AI-assisted dev, the @BitcoinCom team would typically have around 10 to 15 open pull requests at a time. Soon 20 became the new normal. 👀
Here's the story of how their team relies on Bitrise for handling ~60% more PRs and as a quality gate for AI-driven dev: 🔗 eu1.hubs.ly/H0zqVMg0
Shopify's agents author 61% of merged PRs for mobile apps, laptops became the bottleneck. One of the ways they scaled further is with Remote Dev Environments built on top of Bitrise RDE. Bitrise Remote Dev Environments that run the same VM/tooling stack as CI, instantly available from warmed up pools for coding agents as well as let software engineers scale beyond their laptop. Engineers work on 5-6 things in parallel while fans hum silently. Agents have the exact same dev environment as humans.
It's incredibly fulfilling to work with leaders in the industry such as @Shopify with @mustafa01ali, @colinta and @joyofmed providing the building blocks for their Software Factory.
Local machines had become a bottleneck for agentic mobile development at @Shopify, so we gave agents their own remote dev environments. They now author 61% of our merged PRs 🧵
7
207
Friends, need opinions/advice. What is a Software Factory?
@GlncsrSzabolcs and i are experimenting to show and explain what a Software Factory is in practice. With actual use cases, actual tools. Show the fundamental building blocks and operation of a Software Factory you can build today.
Feedback request please, does this resonate, how? What would you change/add/remove? Shoot! 🙇♂️
software-factory.tools.bitri…
1
94
Free trial open if you want to give it a try: bitrise.io/platform/remote-d…
1
31
"Bitrise is powering your Software Factories" --
I was catching up with old friend talking about the changes in the industry. During the chat i realized this really condenses what we do at Bitrise.
In the agentic era of software/product development you need three major components:
1.) Intelligence to build/debug code (LLM)
2.) Orchestration + Harness to bring the intelligence the right context and execute
3.) Software+Hardware infrastructure (platform) that is optimized for the Software Factory to build and test
I listened to the podcast with @GergelyOrosz and @mattpocockuk , where Matt mentioned that he does not feel his local dev environment makes sense anymore. It was such a joy to hear, like feeling understood! 🤩
At Bitrise we built #3, because we think exactly the same (our experience matches). Our cloud based on-demand macOS and Linux dev/build/test environments enable both our teams and our Customers' teams to build and ship faster.
Customers bring their intelligence/harness and we scale the rest, long running remote dev environments, from the same golden image as the short running CI/test environments. All orchestrated with choice of CLI/MCP/API, and yes we still have UI as well. 😄
Such an amazing time to be alive to experience and work in this evolution.
76
As a regular user of ForeFlight super proud of help building this product <3
Pilots and flight crews trust @JeppForeFlight to plan, optimize, and monitor 60M+ flights a year. ✈️
They first replaced self-hosted CI with Bitrise, then added Build Hub to keep GitHub Actions while builds run on our managed Macs.
Their full story: 🔗 eu1.hubs.ly/H0ywnJS0
2
54
Finally your coding agents have a home on macOS also.
On-demand, cloud-based macOS and Linux VMs for coding, testing, agents and people.
On a personal level: Super excited about this launch, all the work we put in at Bitrise has come together in a single coherent platform serving the range of indy devs to the likes of industry leading Shopify.
Helping software engineers and companies ship better products is incredibly fulfilling, having engineers as customers and build together is just next level thrill.
Most of the Customer company names we built this with are under NDA and can't name them, but oh-boy it was so much fun. One name i can mention and a massive shoutout to them is Jeppesen ForeFlight. As early adopters they helped shape the roadmap, while we were helping them build and test the software that is used in most cockpits globally.
So anyway what is this?
✅ Bitrise Remote Dev Environments gives you on-demand cloud machines that start in seconds, running the same stacks, caches and hardware as your Bitrise builds.
✅ Work in one yourself, or give every coding agent its own and let several run in parallel, each on its own branch, so they are not competing for your CPU or your iOS simulator.
Give your agents a machine of their own 🤖
🎉 Bitrise Remote Dev Environments is now generally available, and we're launching it on Product Hunt today:
producthunt.com/products/bit…
Would appreciate some love! 🤩
#BitriseRDE #AIAgents #ProductHunt
7
5
271








