About
Hands-on engineering leader with two decades of experience building cloud-native…
Services
Articles by Derek
-
Getting GCP Certified
Getting GCP Certified
I set out to pass both the Google Cloud Platform Associate Cloud Engineer, and Professional Cloud Architect exams this…
3 Comments
Activity
3K followers
-
Derek Herman shared thisWell… Fable 5 was good while it lasted. I was mid planning session when I lost access and the difference was immediately apparent that Opus 4.8 was a downgrade in reasoning. Kind of frustrating that the US government is meddling in what model I can use to build software with. Not loving that #anthropic https://lnkd.in/gGZ73DdJStatement on the US government directive to suspend access to Fable 5 and Mythos 5Statement on the US government directive to suspend access to Fable 5 and Mythos 5
-
Derek Herman reposted thisDerek Herman reposted thisI thought I was vibe coding. Chat GPT to Claude to Vercel - feeding myself step by step instructions on what to do next... I have pitch decks, technical decks, a landing page, and two different lackluster prototypes. On one hand, I had created things. This is cool. On the other, I lacked confidence that any of it was legitimate. How much environmental waste did I create trying to learn wtf I'm doing with these tools. No matter how much I pushed back against the "yes man" positive reinforcement my tools gave me, I still felt like they were just executing without insights. I've worked with enough designers and developers over the years to know that this just isn't how it should be working. So I built subagents to perform these roles. Similar results, the blind leading the overly capable blind. Then I met with a senior developer to gut check myself. He introduced an agent named Walter, and now, I am actually vibe coding. Walter, based on Breaking Bad and complete with all the appropriate nomenclature, is a skill that forces Claude Code to take me all the way through a full planning and scope cycle before we actually write a single line of code. It's amazing. Link in the comments. If you are adventuring with vibe coding, you need this layer. Now I've got Node.js installed, I'm operating Claude in my terminal, and I'm actually vibe coding a functional demo of my product. I can't wait to show yall. Thanks Derek Herman . Walter is amazing.
-
Derek Herman shared thisI’ve been heads down lately and not really promoting this tool yet. However, the Walter skill I created is pretty amazing if I do say so myself. Give it try and let me know your thoughts. I’m still actively working on it but it’s shipping lots of production code for me and those I told about it directly. https://walter.cooking
-
Derek Herman shared thisJust finished rebuilding my site. It is a more complete picture of what I’ve been building and thinking about lately. https://derekherman.co
-
Derek Herman posted thisI'm pretty dang excited for the Claude Code hackathon in Portland this weekend. I have a good idea of what I'm building, but still waiting to hear what the rules are and if my plan is allowed. What would you build, anyone got some ides?
-
Derek Herman shared thisAI coding tools like Claude Code are genuinely very good at writing code. That is not the problem. The problem is that they will confidently write code you do not need, do not want, and did not ask for, and then tell you it is correct. They add abstractions. They add tests. They extend the system in ways that look reasonable. If you are not already an expert, you have no reliable way to know it is wrong. It passes checks. It feels complete. Momentum builds. Until the project grows, a pivot is required, and you realize you cannot answer two basic questions: What is the blast radius of this change? Was the original path even the right one? By then, reversing course is expensive and painful. This is not a code generation problem. It is an SDLC problem. I have been working on making the SDLC executable, where changes trigger enforced processes that surface impact, assumptions, and dependencies before code is written. What do you rely on today to understand blast radius and correctness when AI is doing most of the writing? #sdlc #ai #mcp
-
Derek Herman posted thisLeadership keeps cutting staff and calling it “efficiency.” The result is skilled builders, designers, and operators pushed into the gig economy. Their best work is locked behind NDAs, yet every new project starts with “show me proof.” We need a privacy-first proof-of-impact network that lets talent share verified results without leaking IP.
-
Derek Herman posted thisEveryone’s talking about LLMs, but most of the hard problems in AI products aren’t model problems. They’re context problems. Context now decides whether an AI product feels intelligent or painfully average. Access to powerful language models is now table stakes. What truly separates products is the context pipeline that feeds those models: what information is captured, how it is filtered and reshaped, and when it is delivered. This is more than prompt engineering; it is context engineering, a discipline that blends data architecture, memory design, and real-time orchestration. A few patterns stand out: - Dynamic relevance: context ages in minutes, not weeks, so systems need strategies for freshness and decay or they drift into irrelevance. - Selective memory: storing everything is impossible, and choosing what to forget is as important as choosing what to keep. - Signal over noise: models improve when context is structured, typed, and scoped; unfiltered text dumps rarely help. - Privacy by design: context engineering touches sensitive data, so consent and compliance must be built in from the start, not retrofitted. Most teams still cobble these pipelines together with ad-hoc scripts and manual rules. That approach will not survive the next wave of multimodal, real-time applications. The industry needs tooling and shared standards that treat context as first-class infrastructure. I’m deep in this problem space and would love to compare notes with anyone tackling the same challenge. What practices or frameworks are helping you manage context at scale?
-
Derek Herman posted thisSo many senior engineers are out of work right now. That’s not a talent problem. That’s a leadership failure. The industry’s been chasing hype, handing complex systems to undertrained devs and calling it innovation. Eventually, the bill comes due. Maybe then we’ll remember that experience isn’t optional when things actually need to work.
Derek Herman replied to a comment
2mo
Jonathan B. But they did just add games and streaks LMFAO
Derek Herman commented on a post
2mo
It is a dope song. Also, I feel like this is a better use for LinkedIn because everything else feels disingenuous or self serving these days on this platform. It’s either hiring me, I’m hiring, look at the cool stuff I or we built, or how much you hate or love AI.
-
Derek Herman liked thisDerek Herman liked thisNVIDIA has lost it 🤯 someone just made CUDA run on Apple Silicon. your MacBook can now run the same particle simulations NVIDIA H100 clusters do, using the EXACT CUDA code they wrote. for years, CUDA only speaks NVIDIA. Apple GPUs speak a different language called Metal. So they built a translator chain using get 5.6 sol that rewrites CUDA 5 Metal step by step through existing compilers. zero source code changes. they tested it on a heavy 3D fluid simulation, not a hello-world demo. this one sorts particles, builds neighbor lists, and runs atomics simultaneously. the kind of workload that breaks toy translators. results on a single laptop chip: → Apple GPU: 6 seconds → same code on CPU: 60 seconds → 10x faster, GPU fully saturated the entire run the translation layer wasted zero performance. cross-compilation usually eats 30–50% of your throughput. this eats none. every CUDA codebase in the world just became runnable on a MacBook. the NVIDIA moat everyone said was uncrackable just got a crack. 100% open source. 👉 Follow AI News World to be the first to get AI news, tools, and trends every day.
-
Derek Herman liked thisSomeone in my network will be perfect for this gig:Derek Herman liked thisNew docs/content engineering role on the team! Super excited, this is a really fun one because it spans so many products :) https://lnkd.in/gntmqsxSTechnical Documentation and Content Engineer, Claude DocsTechnical Documentation and Content Engineer, Claude Docs
-
Derek Herman liked thisDerek Herman liked thisSome personal news: I'm moving from Community Manager to Member of Technical Staff at Anthropic! 🎉 When I joined the community team, my job was to bring people together around Claude, online and in-person. Along the way, the community taught me more about what Claude Code can do than any docs ever could and I found myself building skills, writing eval rubrics, and automating everything in sight. Before I go, some things from the community chapter I'll always be proud of: * Working on Claude for Open Source — getting Claude into the hands of the maintainers whose work the whole ecosystem stands on. * Helping launch the Claude Community Ambassador program — builders hosting meetups, workshops, and hackathons in cities around the world. * Coaching community members through conference talk proposals — and watching them land the stage and deliver amazing talks at our Code with Claude conferences * Running early access programs that put features in front of community builders — and turning their feedback into real product changes ... and so many more. A huge thank you to my manager Jason Bigman. The kind of leader who hands you problems bigger than your title and then coaches and backs you completely until you find success. And to the entire community team (Martine N., Cassie Baquero Mayes, Jun Fujita): you made every launch, every event, and every day genuinely fun. Community will always have a place in my heart. I'm trading "how do we bring builders together?" for "how do we help them build?", so I won't be far! The Anthropic community is in the best hands I know. And to every ambassador, maintainer, and builder I've gotten to work with: keep thinking!
-
Derek Herman liked thisDerek Herman liked thisIntroducing Claude Sonnet 5, our most agentic Sonnet yet. It makes plans, uses tools like browsers and terminals, and runs autonomously at a level that, just a few months ago, required larger and more expensive models. Sonnet 5 is a substantial improvement over its predecessor, Sonnet 4.6, on reasoning, tool use, coding, and knowledge work. Its performance is close to Opus 4.8, at lower prices. Sonnet 5 is now the default on the Free and Pro plans, and available to Max, Team, and Enterprise users. It's available everywhere today with introductory pricing through August 31.
-
Derek Herman liked thisTruly excited to be partnering with GitHub and specifically, John Maeda, Mario Rodriguez and team to bring quality design and Impeccable's steering layer to everyone using Github Copilot (starting today, prebundled in the GitHub Copilot app). The GitHub team cares about quality and craft the same way I do. Time to raise the baseline and then amplify human creativity 💪Derek Herman liked thisImpeccable by Renaissance Geek earned its way to 40k+ stars on GitHub by delivering exceptional value. Today, we are making design and quality a built-in layer for all creators: Impeccable is now a built-in skill in the GitHub Copilot app. AI raised the floor. We’re here to help raise the ceiling.
-
Derek Herman liked thisDerek Herman liked thisToday I’m launching my new company: Renaissance Geek, backed by Andreessen Horowitz, building AI that amplifies human creativity and craft. I built Impeccable as a multiple-weekend project to clean up my own AI slop. It’s now past 40,000 GitHub stars and hundreds of thousands of installs, quietly wired into how solopreneurs and Fortune 500 teams alike ship. Today it becomes Renaissance Geek’s first product, in a round led by my GP, Anish Acharya. I’m also announcing a first partnership with GitHub: Impeccable will ship as a built-in skill in the GitHub Copilot app, so a design and quality layer is there by default for everyone who opens it. Why now: AI made a passable first draft of almost anything free. That’s fantastic, but passable is now the baseline, and passable all looks the same. The work that makes someone actually stop in awe comes from a human with impeccable judgment, taste and craft, who iterates relentlessly on the last 20% that make creative work great. Building AI tools smart enough to integrate creatives in that last mile is the gap I’ll close with Renaissance Geek. And I’m hiring the people I named the company after (renaissance geeks: intensely curious, driven T-shaped generalists): a chief of staff, design advocates, and design engineers who sweat the last 10% more than is strictly healthy. AI raised the floor. Time to raise the ceiling.
-
Derek Herman liked thisHugely excited to share eve. Building agents has always involved a lot of plumbing or getting locked into proprietary services or figuring out complex abstractions. Now it's declarative - only a matter of writing a few files (as little as 1!), on top of an OSS framework.Derek Herman liked thisIntroducing eve, an agent framework. 𝚊𝚐𝚎𝚗𝚝/ 𝚊𝚐𝚎𝚗𝚝.𝚝𝚜 𝚒𝚗𝚜𝚝𝚛𝚞𝚌𝚝𝚒𝚘𝚗𝚜.𝚖𝚍 𝚝𝚘𝚘𝚕𝚜/ 𝚜𝚔𝚒𝚕𝚕𝚜/ 𝚜𝚊𝚗𝚍𝚋𝚘𝚡/ 𝚜𝚌𝚑𝚎𝚍𝚞𝚕𝚎𝚜/ Like Next.js, for agents. https://lnkd.in/gcQquQiB
-
Derek Herman liked thisDerek Herman liked thisWe are excited to welcome Brandon Ross as Head of Partnerships. Welcome, Brandon!
Experience & Education
-
Jack
View Derek’s full experience
See their title, tenure and more.
or
Licenses & Certifications
Honors & Awards
-
Recent Rockstar
WordPress
■ Following the August 2015 release of WordPress 4.3, was added to the "Recent Rockstar" section on the WordPress credits screen for my contributions to the WordPress.org open source project
-
Recent Rockstar
WordPress
■ Following the April 2015 release of WordPress 4.2, was added to the "Recent Rockstar" section on the WordPress credits screen for my contributions to the WordPress.org open source project
Recommendations received
5 people have recommended Derek
Join now to viewOther similar profiles
Explore more posts
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top content