Robert Merriman

There's an underlying superpower of all high impact Technical Program Managers. Do you know what it is? Getting the right people to talk about the right problem with the right context. Do this and your impact as a TPM will sky rocket.

55

10 Comments

Since recently joining the Scale AI team as SVP of Engineering, I’ve been inspired by how deeply the team understands what it takes to operationalize AI at enterprise scale. Over the past several years, Scale has helped some of the world’s largest enterprises integrate AI across their most complex workflows. Today marks the next chapter in that journey. We’re excited to open-source Agentex, the agentic infrastructure layer in Scale GenAI Platform, built to enable enterprises to manage secure and reliable enterprise AI. We believe Agentex will become the standard layer for hosting and orchestrating AI agents by enabling developers to build freely while giving enterprises the control and reliability they need for mission-critical systems. Starting today, it’s open-sourced and available to everyone. Learn more about how Agentex powers enterprise AI workflows and try it out yourself: https://lnkd.in/gP7kQsMW

473

40 Comments

The Best Leaders are Often the Ones You Don't See The most impactful work I do as a TPM is often the work no one sees. Recently, my team was collaborating with another team on a critical project. We were clearly misaligned on dependencies and timelines, and the project was at risk. I knew we needed alignment, but an official escalation felt premature. The lead engineer was tasked with sending an email to clarify things. He came to me first, and together, we drafted a clear, diplomatic message that explicitly laid out our position and asked for alignment. The email went out, and it worked. The message landed perfectly, and it opened the door for a productive dialogue. Another lead on my team then reached out to me for help in crafting their own response to the other team's reply. Together, we drafted a response that perfectly articulated our next steps. By the end of the day, both teams were aligned, and the project was back on track. My name wasn't on a single one of those emails, but my fingerprints were all over the crucial parts of the conversation. The lesson here is knowing when to lend your voice, not so you can get the credit, but so the message lands where it needs to. 👻 Be the 'Ghostwriter' of Clarity: Use your expertise to help others articulate complex ideas. A well-phrased email can be the forcing function that gets everyone on the same page. 🤝 Empower, Don't Command: True influence often comes from empowering others to act. Your job as a leader is to build a team of leaders, not a team of followers. 🎯 Prioritize Impact Over Credit: The goal is project success, not personal glory. By staying in the background, you build trust with your team and ensure the right voices are heard at the right time. That's the power of leading with empathy and a focus on the collective win. It builds trust, and it moves mountains. – 👉 Just sharing my journey and the lessons learned along the way. If you're on a similar path and this resonates, follow me, Rony Rozen, for more real-world reflections from the world of tech and leadership.

159

28 Comments

One of the best pieces of advice I ever got from a Sr. Engineer at Google about designing for failure was this: “A resilience that you don’t practice is a resiliency that you don’t actually have.” Yesterday, half of the internet was down for hours due to a single DNS issue. A single DNS issue, just a name that wouldn’t resolve, took down apps, login flows, even doorbells and games. It was not because AWS didn’t have backups or regions, or fancy diagrams. But because too many teams believe redundancy on paper means resilience in reality. What happens when the main database endpoint is gone? What if your code only knows one region, one address, one set of credentials? What if everything you rely on, feature flags, logins, and user IDs, vanishes with a single cloud hiccup? Most teams hope they’ll fail over. Most teams assume their system will just work in a disaster. But hope is not resilience. Assumptions are not resilience. Pretty diagrams are not resilience. Resilience is built when you simulate failure for real. Resilience is built when you break things in staging, on purpose, and learn what dies. Resilience is built when you practice what to do when the map you drew is wrong. If you’ve never pulled the plug on your main region, you don’t know what breaks. If you’ve never lost your trusted config store, you don’t know how lost your app really gets. If you’ve never watched your fallback routines kick in, you’re betting on hope, not engineering. This is what separates textbook architects from true engineers: – They rehearse the worst day, not just plan for it. –They discover the ugly gaps before customers do. –They don’t just design for “up”, they practice surviving “down.” There’s a chance, no matter how small, that: - Every cloud will fail. - Every network will stall. -Every dependency will have its bad day. The only question is: Will you find out how brittle your design is in a quiet test run…or when the world is watching? Practice failure. Make it boring and routine. —-  P.S. Follow me for more system design insights and check out Layrs - the Leetcode of system design: layrs.me Built to help you crack interviews with: - 60+ problems - Interactive canvas - Instant feedback - Easy-to-hard learning flow Join our Discord:https://lnkd.in/gccq4Xrx

249

9 Comments

One of the most powerful ways to grow as an engineering or product leader is to put yourself on-call for your own products. It’s not just about handling incidents—it’s about gaining a true pulse on your observability, operational posture, and customer experience. It’s also where you see the reality of AI in action: where it’s helping, where it’s hurting, and where it still has room to grow in reducing MTTR. I often remind my team: we’ve come a long way—5x better than before—but this is only 5% of what’s possible. The opportunity ahead is inspiring, and the path forward is in our hands. With my team and AI, together, there’s no challenge too tall. We can get anything done. I truly love my team. ❤️ #Leadership #EngineeringLeadership #OnCall #Observability #AI #Teamwork

221

6 Comments

Most managers think they’re giving helpful performance feedback. Most employees disagree. So how can managers deliver valuable written and verbal reviews that really help their reports...? I shared my thoughts with LeadDev! https://lnkd.in/eDyXAVBW

38

At AWS, we often say internally that our desire is for our service availability to be "indistinguishable from perfection by our customers." When we say that, we acknowledge that perfection isn't possible in the world we live in where "everything fails, all the time." But we simultaneously acknowledge that it is our job - to the best of our ability - to make the imperfect look and feel perfect. On October 20th, lots of our customers had a bad day because we failed to meet that goal. The reasons they had a bad day are now well documented publicly - the transparency with which AWS is sharing the details make me proud to be an Amazonian. The things you can see from the outside are the tip of the iceberg of the response that's happening internally - I don't know what the numbers are, but I would bet there are well over 1000 teams authoring "Correction of Error" documents that lay out everything they can possibly find that we should learn from our errors and those rascally imperfections. But that's not what I wanted to tell you about, really... I'm on one of those teams that's writing a document. I'm summarizing the impacts and learnings from a group of our customers. And I found real data supporting something that I had thought might be true before - but when these things happen so rarely, you don't get many chances to collect data to prove it. Here's what I think I learned: The strongest predictor of whether a workload was available during the service event was the frequency of failovers that workload had experienced historically. This seems obvious - of course workloads that failover often can failover reliably. If that's true, then there's a simple answer to the question, "How do I build a workload that will reliably failover when I need it to?" And that answer is, "You fail it over - in production - alot." If you're doing that, everything else will get ironed out in the process. You'll find all the canaries that don't work, automation scripts that only work on Tuesdays before 9am, weird circular dependencies. You'll find and fix (almost) all of those things. Just by doing it often. You'll find and fix the majority of the remaining issues with some fault injection testing. Now, don't hear me say that this is easy. It's not easy. You're going to find all kinds of things that have to be patched up and fixed and changed and refactored. While you get the system into the groove of regular failovers, you're going to have to constantly prioritize resiliency risk mitigations vs new feature builds - you're going to have to make hard decisions. But the data now tells us: "How do we reliably build systems that reliably failover during a disaster?" has an answer that is very simple, but not easy: "Just do it." (h/t Husin Chandra). Or, more specifically, "Just failover often." (While I am employed by AWS, on social media including LinkedIn, I speak only for myself, not my employer).

220

13 Comments

Need a low-risk way to test big changes without impacting customers? Try shadow mode. Shadow mode (or dark launching) lets you test new features in live environments without affecting end-users. This way, you test real-world performance, gather data, and identify bugs without impacting user experience. Shadow mode allows us to refine and debug behind the scenes—so when we fully launch, we’re ready. Why it’s powerful: ↳ Detects issues before full release to customers ↳ Operates in real-world settings ↳ Validates performance at scale ↳ Provides insights on scaling ↳ Validates functionality in real environments without impacting customers. Engineering and product win-win! Help others and answer in the comments: 💬 Have you used shadow mode in your projects? ---- ♻️ Repost and share these leadership tips ➕ Follow me, Ashley VanderWel, for more unfiltered leadership truths ---- New here? Welcome—I'm so glad you're here. I post daily to help leaders grow and level up their careers. Follow along, and let’s level up together.

19

10 Comments

Classic SRE is the only thing keeping Generative AI from being a dangerous toy. Companies are currently flooded with "AI Engineers" who can build cool prototypes but have no idea how to run them at scale, secure them, or control their costs. They are building "probabilistic software" (AI) using "deterministic ops" (traditional DevOps), and it's failing in production. Anyone agree with Gemini's take? #sre #devops #gemini

63

12 Comments

There's been some talk around storage unification and zero-copy lakehouse integrations recently, and I wanted to better define these terms as well as look at ways we should evaluate different design decisions in this space. So I’ve published a new post: A Conceptual Model for Storage Unification Storage unification isn't about “one storage format to rule them all,” but virtualizing them into a unified logical model. In this post, I lay out a conceptual framework for thinking about storage unification (aka data virtualization). I explore: 🔹 Internal vs. shared tiering (and why shared tiering is so tricky). 🔹 When materialization makes more sense. 🔹 Client vs. server-side stitching. 🔹 Lifecycle and schema management across evolving formats. 🔹 Seven key considerations for evaluating different approaches. This is the groundwork for a set of posts where I’ll dig into how some real-world systems deliver storage unification, primarily about unifying real-time and historical data. https://lnkd.in/dBiS2kxJ

45

Recently I went to a local mixer here in Sonoma County and did something radical: I left the AI buzzwords at home and just asked people, “How is technology actually feeling in your business right now?” I heard it all: “Excited, but overwhelmed.” “Curious, but I don’t trust the hype.” “I’d use AI tomorrow… if someone I trust set it up.” Those conversations did more for my roadmap than any online trend report. Face to face, with eye contact, people tell you what they’re really worried about: staff burnout, sloppy data, missed calls, chaos in the back office. That’s where AI actually belongs... quietly fixing the unglamorous problems. If you’re building in AI, go talk to the people on your street, not just your Twitter feed. The answers are usually sitting at the bar table across from you.

32

23 Comments

What does it take to build a successful engineering organization? 🚀 It’s not just strategy decks or scaling headcount. It’s the messy, behind-the-scenes work—building clarity when there’s chaos, balancing autonomy with alignment, and constantly evolving your architecture and culture as you grow. I’ve spent the last few years learning (and re-learning) this. So, I wrote a 3-part blog series to share some nuggets. If you're leading (or hoping to lead) at scale, I hope there's something useful in here for you. Would love your thoughts, and even more—your own lessons 🙂 #engineering #leadership #lessonslearned

29

1 Comment

Have you been a victim of workslop 🪣? A recent HBR report on bad AI content at work really hit home for me. I've seen it everywhere and I know you have too. Workslop is that super polished-sounding content that looks impressive but says so little and lacks specifics in your work context that it offers no real value and even wastes time (HBR says AI slop is wasting $186 per employee per month). LLMs are incredibly impressive and are starting to make a dent on actual work (see OpenAI's latest research). But if companies rush to deploy any AI solution and don't think about training, metrics, or grounding it in their own work context, it's going to lead to wasted time and workslop. Why the rush? I've heard it firsthand in customer meetings: "We need AI or competitors will eat us alive." One report found 77% of CEOs feel competitive pressure to adopt it. So here are some my tips to get more out of AI deployments and reduce that workslop. 1. 📊 Your work context matters! Don't rely on a vanilla LLM or generic "enterprise chatbot."  It must be tuned to your work content, metrics, and data. Many chatbots only connect superficially to 3rd party content, or pull in personal files but not shared ones. So they aren't very effective. (Check out Dropbox Dash, we do this really, really well). 2. 🎯 Define your success criteria Know what you're solving. If you want fewer errors or faster delivery, measure that. Don’t just declare success because "AI is deployed." For software development, many companies see success when all devs are using an AI coding copilot. Or "generating code with AI". But for what purpose? Faster deployments, fewer bugs, improved feature velocity? That's what you need to measure. 3. 🛠️ Training and culture That blinking chatbot cursor is intimidating. In one survey, 49% of workers felt pressure to adopt AI, but 55% didn’t know how. The best deployments invest in training, shared playbooks, demos, and mentors. And don’t just do it once and think you are done. Ongoing measurement, monitoring, and training is needed. Also, look for AI tools that aren't just a chatbot. Products that blend seamlessly in the background. Not everything needs to be solved with a language model. 4. 🚀 High performers are going to highly perform and vice versa Ignore AI for a second. Top performers always seem to "figure it out" in the workplace. They learn the new tool and incorporate it into their workday. And so AI deployments are just new tools. It’s critical to watch and observe their output and use them as mentors (above). But others may simply take AI output and present it as their own. Workslop! And unless you guard against that (via reviews, performance management, feedback, and coaching), you're going to get more workslop. And that annoys and slows down everyone else. AI is nothing new here. Don't assume AI is some magic bullet and everyone's performance will rise. Don't just deploy it and forget it!

40

1 Comment

LLMs are powerful, but they are stateless machines. They don’t carry memory or intent, they work with the context you provide at that moment. The same model can feel brilliant in one scenario and completely off in another. The difference is not the model, but the context and use case we design around it. The technology is already here. The real opportunity lies in how we shape it to solve real problems.

103

11 Comments