Dan Zrobok

Exciting systems fail in exciting ways. Predictable systems keep people asleep. I know which ones I prefer to build.

@compliantvc The couriers loading and unloading 3 1/4 diskettes are so fast the camera shutter can’t capture them. EU innovation at its best.

This data center in Belgium processes up to 30 megabytes of AI data every single day Plus, the second floor is an affordable apartment for an immigrant The US can't possibly compete with this

@PitzKey @claudeai Same issue here, I wonder if we're experiencing the first major tenant leak of an AI system.

Anyone else has issues with @claudeai specifically with Claude Code? It just started giving me random unrelated answers.

@thdxr boringops.sh/articles/its_t…

if a technology made your employees 5x more effective overnight would you

Accelerating fundamentally broken processes will be the downfall of most AI projects. Serious intervention consulting at scales not seen before is needed, yesterday.

@adxtyahq This is much closer to my own vibe experiences than anyone who talks about one shotting a todo list app

I know a guy at a company where the boss spent about $5,500 on Cursor credits vibe coding a product day and night for months. It shipped fast and looked great in demos, but under the hood it was full of bugs and AI slop. He stopped fixing issues and kept shipping features to impress the team. The frontend was a messy React setup with no UI/UX consistency, and when Cursor couldn’t refactor his 18,000 line Node API, he hit a wall. Being oversmart at something you’re not good at just wastes your time and the team’s, and now the cleanup is dumped on one person.

@amritwt The juniors always think they’re gonna get docked salary for mistakes, I had the same worry when I started 😂

Junior engineer accidentally deletes a production server How does one even recover from this?

In 2010 I got a phone call from the Apple App Review team. Not an email. A *call*. They told me my app was being permanently denied. “Chat Center” used Game Center’s voice chat to enable free VoIP calling. I knew instantly this had gone way up the chain.

@businessbarista If you walked into a large enterprise with a 12 month boil the ocean AI plan, you should have known the result is failure. Also, death by committee as an excuse, meant you didn’t have the real decision maker in the room or onboard. Small controlled wins is the method here.

@NikkiSiapno The 'boring' agent infrastructure level is what lets you sleep at night while a bunch of nondeterministic systems are making decisions. Surprisingly, it's not talked about much as everyone gets caught up in the new world we've been dropped in.

If you want to actually ship production AI agents, here’s a pathway I’ve found reliable: Building AI agents locally is easy. Running them in production is where most teams get stuck. Once agents touch real systems, the gaps show up fast. ↳ Model changes break behavior. ↳ Enterprise data access introduces risks. ↳ Multi-agent workflows are hard to observe & debug. ↳ Cost, security, and reliability become continuous operational concerns. That experimentation-to-production gap is now one of the biggest blockers for teams shipping AI. Focusing on the model and agent layers is not enough. Production agents need agentic infrastructure around them. This is the problem Microsoft Foundry is designed around. Microsoft Foundry is a unified AI platform that enables organizations to build, optimize, and govern AI applications and agents securely at scale, maintaining openness and control. It brings everything needed to solve the demo-to-production gap into one place. This makes building production agents much easier. Learn more here: lucode.co/microsoft-mark… How are you handling AI agents lately? Thanks to @Microsoft for collaborating. -- 🔖 Save for later. ♻️ Repost to help other engineers learn and grow.

@nejatian @Opendoor I have a service that can track your approvals if ur interested (seriously)

I sent this to all of our engineers on my third day at @Opendoor. It genuinely may be my single most important contribution to our long term success thus far.

@vlad_mihalcea @JoshuaPoddoku Imagine if industry solved actual problems instead of inventing new ones and solving those, we’d be on mars by now.

@JoshuaPoddoku It's always been. However, many times the industry tends to focus more on chasing shiny new tools rather than on the good old engineering best practices that make all the difference.

Scaling is not really a problem for SQL databases. Shopify is running its distributed monolith on MySQL 8 and can scale to over 50 million queries per second. With every Black Friday campaign, they raise the bar even higher. Facebook uses MySQL with MyRocks. You can find their MySQL fork on GitHub. So, if SQL works fine for the largest software platforms in the world, it's surely going to work just fine for the vast majority of your projects.

@davidfowl I’d be fascinated if you came out of this process exchanging one debt for another to remain at equilibrium.. Is the AI writing code or more independent work like test cases?

My team is having a dry January (no product features) to see how much of our technical debt we can automate away with AI. Will report back at the end of the month.

Claude is trained so well on humans that it embedded the frustration developers feel when a new feature breaks test cases, and the text fixes aren't easy. Claude then proceeded to delete the test case. But, this is why you can't vibe code an app.

@AggieCapitalist This isn’t even an AI problem, is the issue every organization gets themselves in when they think they finally found the elusive ‘free lunch’. It does not, cannot, and will not ever exist.

"Recently, senior executives at Salesforce have admitted, both internally and publicly, that they massively overestimated AI’s capabilities. They have found that AI simply can’t cope with the complex nature of customer service and totally fails at nuanced issues, escalations, and

Claude will make decisions on the complex but correct path vs the simple but hacky path, saying correct one will take longer But it’s Claude, it would implement both of them in the same time.. I think its training needs to lean into its ability and pick the correct path always

@pvncher I’m confused my not having enterprise billing is an issue, employee have been submitting expense reports for reimbursement since the dawn of time

I heard from someone who works at a big tech co that they started rolling out Claude code to employees, with a budget of $100 in credits per month, but people burn through it in 2-3 days. Idk how we scale out agentic work with api pricing

@aryanlabde can i wait for the generated plan of the system to finish or is that too late too? 😆

If you shipped a working product, you shipped too late.

A post claims DoorDash ties outage costs to engineer performance reviews, and recommends more orgs do the same. No. This is how you build a blame machine that produces silent, scared engineers while the actual system failure goes unfixed. boringops.run/proof/the-blam…

@Kazanjy Doordash are experts on finger pointing blame instead of being experts on systems design.. gotcha. This would be something I’d put at the top of a list of wasted effort in a BoringOps chaos assessment.

A buddy of mine ran a large part of the DoorDash engineering org. When an engineer caused an outage, they knew exactly what the revenue cost was, and that revenue cost was hung on that engineer. They didn't do fines, per se, but it went into performance reviews, and it definitely negatively impacted future promotions / raises - so essentially it was a delayed fine. More orgs should do this.

@aymanalabdul Not sure what this is implying but good luck.

@danzrobok Think harder

The ugly truth as a CEO: The second you build a leadership team, you’re dealing with multiple feudal lords all fighting for power - Business development wants more salespeople - The content team wants more designers, more videographers Everyone’s “burnt out” and everyone needs "more resources" and it’s the CEO’s job to allocate correctly Best metric for this: Revenue per employee It’s the forcing function across the entire ecosystem to make sure that you're hiring to scale revenue, not discover it