Activity
10K followers
-
Rob Zuber shared thisWho's AWS in 2026? OpenAI and Anthropic. Look at where they sit in the market and the comparison makes sense. This Wardley table is a helpful way to think about it. Platforms create space for other companies to innovate on top of them. As those products prove there's a market, the platform can move up the stack and absorb what was once a differentiated product. AWS has played this game for years. Foundation model companies can play it at an entirely different speed. If you're building on top of OpenAI, Anthropic, or another foundation model, you have to assume they're watching the same market you are. Something differentiated today could become a platform capability surprisingly quickly. So something we're asking ourselves as CircleCI is what space are we operating in, and how do we create enough room to build something durable there? The ground is moving too quickly to answer that question once a year. You have to keep asking where you're operating and whether there's still enough room there to build something durable.
-
Rob Zuber shared thisThe longer you've been building software, the more there is to unlearn right now. I was chatting with Michael Webster, who started professionally in 2012, and Hanabel Mengistu, who graduated last year. Between us we cover a lot of ground, but we're all still figuring out what the job is becoming at the same time. Webster said it better than I could: what happens when the stuff you like doing, the stuff you've spent years getting good at, becomes the bottleneck? Hanabel is coming at the same shift from almost the opposite direction. She doesn't have decades of habits to unlearn. And because none of us has the answers, she's been able to contribute earlier than she expected. In some ways, it's a more level playing field. That doesn't mean experience doesn't matter. But Hanabel said something I haven't stopped thinking about: some junior engineers feel like they aren't learning enough because "just ask an LLM" has quietly started replacing the mentorship they used to rely on. An LLM can answer the question. A good mentor helps you understand what you didn't know to ask. So maybe the question isn't whether we still need junior engineers, or how much code they need to write themselves. It's what we're actually growing them into. My bet is that curiosity matters more in that job than it did before. The people who are willing to keep learning, question how things have always been done, and follow the thread when they encounter something they don’t understand are going to do well.
-
Rob Zuber reposted thisRob Zuber reposted thisWe’ve been doing a lot of testing to understand what actually makes a product work better for coding agents, and some of the results have surprised us: 🧹 Removing ~99% of a build log cut wrong-approach fixes from 48% to 18%. 🛠️ Giving agents more tools didn’t necessarily make them better. In one test, an agent had 138 MCP tools available and consistently relied on 10. ✅ Even when ~90% of fixes made CI pass, only ~77% met our bar for something we’d actually want to merge. Read on in the latest Confident Commit ⬇️What we've learned (so far) making CircleCI work for agentsWhat we've learned (so far) making CircleCI work for agents
-
Rob Zuber shared thisI've been going to more in-person events lately. I'll spend a week convinced I've got a pretty good mental model for where software development is headed, then have one conversation that forces me to rethink it. Living in the Bay Area means there are plenty of opportunities for that, so I've found myself saying yes to more of them. No one person or company has all the answers right now. We're all building, experimenting, and figuring out different pieces of this at the same time. Getting in a room and comparing notes has become one of the fastest ways I know to learn. That's the spirit we're bringing to the CircleCI office for Tech Week. On October 6, I'm hosting a session around a problem we've been working through ourselves: AI-generated code is going into production, so how do you actually know whether it's good enough to ship? I'll share what we've been learning, then we'll build together. If you're around, come join us. I'm sure I'll leave with a few new ideas of my own. (And thank you to Dusan Omercevic for the recent pic from Infobip Shift Conference.)
-
Rob Zuber shared thisIf you’re optimizing a CLI for agent use, check out these thoughts from 🚴🏻 Pete Steyert-Woods on the surprising behaviors of agents.Rob Zuber shared thisOur CLI docs were too long, so Claude stopped reading them. After the recent agent-centric CircleCI CLI rewrite I was expecting Claude to easily understand how to use it. The help text was specifically designed to tell an agent how to use it. That’s when I noticed Claude was running `circleci run trigger --help | head -40` — only looking at the first 40 lines! Read the article to see what I did to deal with it, and make sure it didn’t come back. https://lnkd.in/deS3WyuwYour CLI docs are too long. Claude stopped reading them.Your CLI docs are too long. Claude stopped reading them.
-
Rob Zuber shared thisWe spent the last decade consolidating the software development stack. I’m wondering if we’re about to spend the next one pulling it apart again. The forge became the center of gravity for a reason. Source code, identity, pull requests, review, CI, security, deployment. Putting more of the development lifecycle in one place made human-driven software development easy to coordinate. Now development itself is spreading out. Code can start with an engineer locally, an IDE agent, a cloud agent in an ephemeral environment, scheduled automation, a review agent, or something hitting an API. And we're starting to see new Git hosts and distributed Git networks emerge alongside those workflows. That makes the tradeoff behind consolidation harder to ignore. Tying critical delivery infrastructure to a single forge has always meant accepting more lock-in and concentrating more risk in one platform. As development spreads across more actors and environments, and teams have more choice about where their code lives, that coupling makes even less sense. CI needs to be able to follow the work, wherever the work happens. We're still early here, but I think we're going to spend a lot more time reconsidering which parts of the development stack actually need to live together.
-
Rob Zuber reposted thisRob Zuber reposted thisSoftware delivery is changing fast, so we made a place to compare notes. 📝 Confident Commit is now on Substack, where we’re sharing: 🧪 Field experiments from CircleCI engineers, including what they tried, what happened, and what they learned along the way 🎙️ Conversations with engineering leaders & practitioners about how teams are building, validating, and shipping software 📊 Data from millions of software delivery workflows, and what it can tell us about how great teams ship Come compare notes on what’s happening in software delivery: https://circle.ci/4xXWNF9
-
Rob Zuber shared thisEngineering teams aren't going to solve AI sticker shock by using less AI. A few months ago, Gergely Orosz wrote about a shift he was seeing across engineering orgs: after pushing hard for AI adoption, leaders were suddenly looking at the bill and asking how to rein it in. Uber just published a pretty good answer to what comes next. Their agentic requests grew 9.4x from February to August while total spend held roughly flat. Getting there meant becoming more deliberate about benchmark-driven model selection, smarter context, and routing the right workload to the right model. Almost every engineering leader I talk to is somewhere in this arc right now, and we're not going backwards. As models commoditize, we need to get better at the systems around them: routing, context, and knowing which capability fits which job. This is what industrialization looks like. The usage is here. Now the systems around it have to mature. How is your team thinking about cost as you scale agent usage?
-
Rob Zuber shared thisMy first job out of college was in a factory. I worked at Celestica in Toronto, managing the process of building hardware. Most of my time was spent looking at data, finding problems in the manufacturing process, and figuring out how to make that process more efficient. 30 years later, I'm still thinking about factories. We're shifting the job of a software engineer from writing code to designing systems that write code, then building the systems around them that make the output useful and trustworthy. At CircleCI, my team is now called the Factory team (maybe a coincidence, maybe not). We're trying to figure out what software delivery looks like when machines can take human intent and carry more of that work all the way through delivery. What's interesting is how familiar some of the questions feel. Where's the bottleneck? What's slowing down production? Where do we need better feedback? What can we automate without losing confidence? The product is different, but the engineering problem is surprisingly similar.
Rob Zuber commented on a post
1w
The full convo we had: https://www.confidentcommit.com/p/junior-developers-have-one-big-advantage-a5e
Rob Zuber commented on a post
1mo
Full convo with Andrew Zigler: https://www.youtube.com/watch?v=-ihKWEiBEog
-
Rob Zuber liked thisRob Zuber liked thisPull up the dashboard you actually look at each week. Star every metric you added in the last year because of AI. Now defend each one at your table. That was an exercise I ran last week in my Director+ plenary workshop at LeadDev's LDX3 in New York: Judgment is the new bottleneck. The core idea: code got cheap, judgment got expensive. Our software delivery pipelines are well-lit in the middle — speed, quality — and dark at both ends, where precision and impact live. And every metric you keep is an instruction to your org about what to optimize for. Measure adoption, and adoption is exactly what you'll get. Thank you to Ruth Yarnit, Mariana Valette, and the LeadDev team for the invitation. To Suzan Bond and Ameya Kanitkar, whose talks the day before set this up better than I could have planned. And to everyone who did the exercise honestly — including the ones who admitted their dashboard is a mess. Mine was too. That's the point. I'm turning this into a workshop I run with engineering leadership teams directly. If your org is drowning in AI output and can't tell motion from progress, let's talk. Full write-up in the comments.
-
Rob Zuber liked thisRob Zuber liked thisWe’ve been doing a lot of testing to understand what actually makes a product work better for coding agents, and some of the results have surprised us: 🧹 Removing ~99% of a build log cut wrong-approach fixes from 48% to 18%. 🛠️ Giving agents more tools didn’t necessarily make them better. In one test, an agent had 138 MCP tools available and consistently relied on 10. ✅ Even when ~90% of fixes made CI pass, only ~77% met our bar for something we’d actually want to merge. Read on in the latest Confident Commit ⬇️What we've learned (so far) making CircleCI work for agentsWhat we've learned (so far) making CircleCI work for agents
-
Rob Zuber liked thisRob Zuber liked thisTravelling back from the #Shift Conference in #Zadar, and, as usual, it was a beautiful event in a beautiful place! 🌊☀️ I was very happy to present my favorite AI framework, #SpringAI, and our open-source integration with Amazon Bedrock #AgentCore in a workshop. A huge thank you to all the workshop participants for your motivation, interest, questions, engaging discussions, and for helping us hunt down and solve the challenges! 🙌 Many thanks to Filip Srnec for extending the workshop with the #Infobip #MCP_Message_Server, presenting alongside me, and helping resolve issues along the way. It was a real pleasure! It was also a really nice surprise and a great pleasure to meet Nikita Koval from the #AppGlass team at #JetBrains at the conference. We had a great time discussing millions of technical and non-technical topics! 😄 Infobip team — Vassilis Tsantiotis, Ervin Jagatic, Josip Antoliš, Marko Mudrinić, and many others: It was a pleasure to see you all! Some of my conference highlights were: 🔹 “From AI-assisted Coding to Agent-led Execution” by Luis Velasco from OpenAI 🔹 “It’s Giving Insecure Vibes: Secure Coding Literacy for Vibe Coders” by Betta Lyon Delsordo from AWS 🔹 “Building the Software Factory You Actually Need” by Rob Zuber from CircleCI Great talks and plenty of food for thought! And I even had a chance to go jogging and go for a swim in the sea between the conference days. 🌊🏃♂️🏊♂️ It’s hard to ask for a better conference location — absolutely amazing! Finally, a big thank you to the organizers and all the participants for creating such a warm, friendly, and family-like atmosphere. That's what makes #Shift so special! See you next year, Zadar and Shift! ❤️🇭🇷
Experience & Education
Recommendations received
5 people have recommended Rob
Join now to viewOther similar profiles
-
Jeffrey Nelson
Jeffrey Nelson
Mr. Nelson invented Google Chrome OS while working at Google in 2006. He has 19 years of experience in cloud computing and Big Data analytics. He has written two books and many magazine articles on Java and cloud computing. He has extensive experience in search, SEO, and eCommerce industries, building highly scalable web services, and leading world class engineering teams at such companies as Google and eBay. He holds a Masters Degree in Applied Mathematics.<br><br>Specialties: * Big Data, Machine Learning, Cloud Computing<br><br>* Hadoop, Java, C++, Javascript, SQL<br><br>* Leading engineering teams of up to 150 through full life cycle development.<br><br>* Software, Java, C++, Javascript, AJAX, XML, WebSphere, and SQL on Windows or Unix/Solaris platforms.
Explore more posts
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top contentOthers named Rob Zuber
39 others named Rob Zuber are on LinkedIn
See others named Rob Zuber