Bill Weinman

9 interesting observations from my conversation with Mitchell Hashimoto (creator of Ghostty, founder of HashiCorp): 1. Vargant was created because dev environment setup was an unbillable time sink at a consultancy. At the Ruby on Rails shop where Mitchell worked, jumping onto another client’s project could waste half a day. This inspired building Vargant. 2. Terraform won, despite being 7th to market. Terraform won through relentless conference presence, community building, and a better developer experience — not timing. 3. HashiCorp had no real business for four years and their first commercial product was a full-on failure. The initial product, Atlas, required customers to adopt the entire HashiCorp stack. It was a hard sell. HashiCorp pivoted to selling individual services like Vault, and this approach proved to be a winner. 4. VMware almost bought HashiCorp for ~$100M, and Terraform would have not happened if it did. VMWare took took the offer to their board, where they rejected to buy with a single vote. Mitchell said that Terraform probably never would’ve existed if the VMWare purchase went through. 5. Mitchell’s new rule for building software: always have an agent running in the background doing something. He kicks off tasks before leaving the house — research, edge-case analysis, library comparisons — so work progresses while he drives or is away. 6. Open source is moving from “default trust” to “default deny” — and Mitchell thinks that’s how it should be. This is because AI makes it trivial to create plausible-looking but incorrect and low-quality contributions. As he put it: “open source has always been a system of trust. Before, we’ve had default trust. Now it’s just default deny.” 7. Git and GitHub may not survive the agentic era in their current form. Agents cause so much churn that merge queues become untenable, branches proliferate, and repos balloon. Mitchell compares the needed shift to Gmail’s revolution for email: “We’re at the Gmail moment for version control... never delete, archive everything.” 8. The best engineers Mitchell ever hired had boring, invisible backgrounds. No GitHub contributions, no public profiles, companies you’ve never heard of. “Every moment you spend on social media is taking away from something else... the best engineers are the ones that context-switch the least.” 9. Mitchell’s advice for AI-skeptical engineers: start by reproducing your research, not your code. As he puts it: “There’s a lot of people like, ‘I don’t want it to write code for me.’ But just delegate some of the research part.” He uses agents for library comparisons, edge-case analysis, and deep research — not just code generation. “You don’t need to pick up on the ‘it must replace you as a person’ kind of propaganda.” Watch the full episode here: https://lnkd.in/ecdgWKbY Other platforms and transcript: https://lnkd.in/drgXiipV

617

42 Comments

What's one thing you wish existed in your community tech stack? Not a feature request — I mean something that feels fundamentally missing. A type of connection, visibility, or capability you keep wishing someone would build. Tell me. I pay attention to these answers. 👇 #CommunityTech #BehindTheBuild #BuildingInPublic

2

Take it from a reformed software process engineer: folks pursuing 100% autonomous "agentic" software development are making all the same mistakes we did. 100% is an asymptote. A fool's errand. https://lnkd.in/eyhZQaAY

59

15 Comments

Most rework isn't a design error—it's a coordination hole that was never claimed. Can you trace your last three RFIs back to a single unresolved coordination item? Run the audit this week. Comment 'LOOP' and I'll email you the three-question rework loop diagnostic.

1

Do you buying technical debt? If your agency just takes your project and passes it to someone else, you’re buying tech debt you’ll pay for later. Real developers don’t just code, they solve problems, debug, and care about your business. Middlemen? They sell promises, not quality. Your app might work for 10 users… but crash when you hit 1,000 ( like as shows in this image ) Want a product that scales, works under pressure, and actually solves problems? Go straight to developers who care. Stop paying for people who sell projects. Start paying for people who deliver solutions 🌟 . #SaaS #MVP #Startup #Founders #ProductDevelopment

1

When someone says that vibe coded products "become an unmaintainable mess", I'm asking - at which threshold? The threshold at which it becomes a mess depends on one's capacity for technical understanding. Someone who is good at system design, have built large systems before, can reason about them, debug, have meaningful conversations about it - has a very high threshold and can accommodate a lot. They can guide codings agent to produce reasonable results. A junior dev, on the other hand, can get overwhelmed by Claude Code from that first generated website. I still don't have an answer how new engineers should gain that required architectural understanding.

16

12 Comments

If you've used Fiverr in the relative recent past, you might want to review what information of yours is on there. things like your tax forms, auth, PII, etc. https://lnkd.in/eNFRX5yd

3

Stop paying for $3,000 "RAG" bootcamps. Qdrant just put a full, production-grade vector search course on YouTube. For free. This isn't a demo. It's a 7-day sprint where the final project is to ship a complete, production-ready documentation search engine. The full curriculum for real engineers: ➡️ Day 1 • Get on Qdrant Cloud & build your first basic vector search. ➡️ Day 2 • Master Points, Vectors, Payloads, & Chunking. • Project: Build a Semantic Movie Search. ➡️ Day 3 • Learn HNSW Indexing fundamentals. • Project: Benchmark HNSW for actual recall vs. latency. ➡️ Day 4: • Master Hybrid Search (sparse + dense) with score fusion. • Project: Build a Hybrid Search Engine that actually finds keywords. ➡️ Day 5: • Learn Vector Quantization to slash memory costs. • Master high-throughput ingestion & accuracy with rescoring. • Project: Quantization Performance Optimization. ➡️ Day 6: • Use Multivectors for advanced reranking. • Learn the Universal Query API. • Project: Build a Recommendation System. ➡️ Day 7: • Final Project: Synthesize all 6 days to ship a production-ready doc search. ➡️ Bonus: ➕ Full integration guides for LlamaIndex, Tensorlake, camelAI, Jina AI, Unstructured(dot)io, and more. This is the syllabus that separates the "demo builder" from the "production engineer." This is how you build RAG that actually scales. (I will put the playlist in the comments.)

2

You Don’t Have To Move To A No Income Tax State To Win →💰 What This Looks Like In Real Life 🪙 $600,000 Income Baseline CA Effective Tax: 43–45% Tax ≈ $260K+ With proper stack: • R&D credit ($60K–$90K) • Retirement deduction ($80K–$100K impact) • Entity optimization ($20K+) → Possible savings: $120K–$170K → Effective rate may drop into low 20s 💰 $800,000 Income Baseline Tax ≈ $350K+ Strategic stack: • R&D ($90K–$120K) • Retirement ($100K–$130K impact) • Real estate depreciation ($50K–$80K impact) • Entity tuning → Possible savings: $200K+ → Effective rate may fall near mid-20s 👇🏽 🤑 $1,000,000 Income Baseline Tax ≈ $440K+ Strategic stack: • R&D ($100K–$150K) • Defined benefit plan (~$200K deduction impact) • Real estate acceleration • Multi-entity structuring → Potential savings: $230K–$300K+ → Effective rate may compress toward 22–25% 👇🏽 💵 🤑 💰 $2,000,000 Income Baseline Tax ≈ $900K+ Advanced architecture: • R&D ($160K–$230K) • Large defined benefit ($300K+ deduction) • Real estate acceleration • Strategic capital gains positioning • Entity layering → Potential annual savings: $400K–$500K+ → Effective rate may compress toward mid-20s 👇🏽 ⚠️ This is not about zero tax. It’s about intelligent compression. And here’s the part no one says out loud: If you’re building AI, robotics, software, automation, healthcare systems, or advanced manufacturing… You may already qualify. You just haven’t structured it properly. Stop asking: “Which state should I move to?” Start asking: “How should I design?” 🧠 #Alpha #AlphaWife #MarriageIntelligence #HelpYourHusbandSucced #NavyOfficer #PentagonsFinest #InvestorClass #HighIncome #HNW #FamilyOffice #FamilyBusiness #ApprovedAmerica #GRACIQ #PrivilegeCareHomes #PrivilegeHealthSystems #BuildRobotsWithMe #BeautyIQInstitute #BeautyIQPro #WifeyBeauty #WifeyUniversity #MarriageSecurityCouncil #CIAYachtCharters #TaxStrategy #WealthArchitecture #RDTaxCredit #EntrepreneurLife #AIBuilder #RoboticsFounder #WealthEngineering #BusinessOptimization #ScaleSmart #TaxIntelligence #FounderMindset #MillionaireMoves #FamilyOfficeThinking

6

1 Comment

Five Next.js codebases. Same parser. Coverage ranged from 100% on cal.com down to 71% on dub. I've been building a tool that derives a user-journey map from source code, and ran it against nine real OSS repos to test the premise. When the parser reads cal.com, it finds all ten of the journeys a PM there would name out loud. Each one lives in its own directory. The code organizes the product by feature area, and the journeys come back clean. When it reads dub, it finds five out of seven. The two it misses both live inside the same 113-route workspace directory. The structure isn't bad. It's optimizing for something else. The journeys are there in the product; they just aren't legible in the code the way they are in cal.com's. The shape of the code is a measurement of how the team thinks about the product. Teams that think in journeys end up with code that reads like journeys. Teams that think in components, or domains, or some other axis, end up with code that hides the journey behind whatever structure they chose. Cal.com's engineering team wrote about exactly this recently, how team structure and architectural standards end up shaping the codebase: https://lnkd.in/eY-w2b4P. Read it back-to-back with what the parser pulled out of their repo and the mapping is striking. That's information. Not a bug. Full nine-repo data, framework breakdown, honest gaps: https://lnkd.in/eFz_hQn6

2

Have too much fun on your podcast, you do. Yes, co-host Bradley Shimmin and I have no problem mixing fun, tech, cultural references, and commentary into every episode of the Agents of Dev podcast. Ep 8 is no exception. At the risk of being left behind, languishing in the shadows of the open-molt-clawdbot controversies and the run on Apple Mac Mini inventories, we delve into why Clawdbot, then Moltbot, and now OpenClaw lit up the AI, agent, and software development airwaves in a matter of a few days. Yes, it's open source. Yes, it shows how powerful a vibe coding tool can be in the hands of one person with a creative idea. And yes, it shows both how easy it is to create software that is entire unsecure but will spread across the internet anyway, passwords, prompt injection, and a dozen other security issues be-damned. We also talk about the issue of agents and memory, and the interesting, yet simple, ways that OpenClaw handles this. This one open source project has many, many lessons we are already learning from it, and I predict OpenClaw will be teaching lessons about AI agents, good and bad, for a long time to come. That's why you should listen to this episode. (Yes, I buried the lead!) Links to Agents of Dev Ep 8 are in the comments below. Thanks for listening, watching, following, and sharing with your friends. The Futurum Group Techstrong.ai

22

1 Comment

Cloud dev environments tried for years to move development off your local machine. They never quite landed because editors were designed around the assumption that your machine was both the steering wheel and the engine. Coding agents took a different approach. They didn't try to make the editor work remotely. They made the editor optional. That is a fundamentally different move. You prompt, you wait, you review, you iterate. The editor becomes a place to steer from, not where the action happens. Once you accept that separation, a lot of things fall into place. Your laptop is no longer a ceiling on how much work you can parallelize. The constraint moves from hardware to orchestration. Development becomes headless. And when development is headless, the companies that control the agent layer are in the best position to become the next git forges. They already manage the compute. They already run the code. The collaboration layer is just the natural surface that forms on top. I think we will see movement here before the end of the year. https://lnkd.in/drjn22SW

6

Anyone who has worked on one of my dev teams knows I'm obsessed with tooling. You can write magical *transcendant* code in C - heck, you can write it directly to the processor - and if it's still talking to a slow database, or connecting to a slow API, it's... all a wash. So I always start with tooling. But tooling in LIFE is also important. And there are a lot of modern tech influencers - the kind of folks who want to MOG your CLAUDE Cowork setup, or optimize your Ollama models to your exact kit (Macbook M5 by the way - SINGS) - these guys (they're usually guys) always seem to be so into excessive tooling on their own workflows and it just cracks me up. I mention this because nine out of ten of these tech influencers are into Obsidian - a kind of notetaker that exposes meta-data to itself so you can cluster your ideas if you so wish. This isn't wholly useless and I'll admit that it has its place - but it's not really any better in like 95% of anyone's workflows. But you watch them load it up on their little TikToks or YouTube videos like they're all part of the cool club and I just want to write a comment like, I SEE YOU BRO. Stop. It's not just Obsidian, of course - it's everything from some of the kit they all seem to have to have (like they all have the same lav mics?) to the uniformity with which they jumped on and then off of things like OpenClaw. And none of this is a diss. It's fine - and there is a totally valid argument here that these techies are just all doing the latest and greatest and so it just happens to sync up. But I think it's just a kind of subtle gatekeeping. What they're doing is a less insidious version of what the fashion or the looksmaxxing or the political influencers are doing - they're saying, Hey kid, you'd better like and subscribe or else you'll miss something really important and you'll be left behind! What, you did't install Obsidian? Here's a video on why you should have by now! And in 3 weeks, here's another one about why it's old news! This is all fine - and there's room in the ecosystem for this. But as a tech leader all I can add to the conversation is this: Chasing trends is no way to build stable workflows. The latest isn't always the greatest. Tech influencers are very seldom dealing with the same vectors and variables we are in the enterprise space. We have to do what works for us without getting caught up in what's cool with the influencers.

5

If you're looking for a new docs framework for your dev tool, I can't recommend Docusaurus enough. 1. You can change the appearance of any component—no matter how big or small—by ejecting its raw React code into your source tree. I've ejected almost all of its major components and restyled them with Tailwind. 2. It supports content written with MDX, which is like Markdown but better. MDX lets you write JSX tags so you can render React components (like badges and banners) directly inline your content. No more messy templating. 3. It bundles Prism for code snippet highlighting. So you if you want to change how each and every token of your language gets styled—which I do—you can override it with a simple centralized map—which I have. 4. You can swap out its on-site search engine with Algolia, now widely considered the best option for developer sites. It is, in other words, the perfect tool for people like me who are obsessed with every pixel.

118

16 Comments

Today many will think I’m killing my company. From now on, at Pullpo, we’re allowing our clients to purchase the full Pullpo codebase. Every line of code. As a one-time purchase. Why we are doing this: DevEx platforms make a ton more sense when they’re fully customized to every aspect of a team: their workflows, their internal tools, their automations and the very specific metrics that matter to them. That’s the perfect DevEx platform. But as a SaaS, it’s impossible for us to deliver that level of customization for everyone. Every client has different pains and priorities, and as a company we can’t satisfy all of them quickly. Some requests are so specific that they don’t make sense to consider in a generic product. This is changing today. Just two months ago, it would have made zero sense for a small/medium team to spend precious engineering time maintaining and evolving its own DevEx platform. You’d rather ship product than build internal tooling. The cost, the complexity, and the ongoing maintenance burden were simply too high. But the equation has changed. Building in-house is getting faster and easier every month with AI. Integrations are simpler. Infrastructure is more plug-and-play. And more and more “new features” look less like multi-week projects and more like a well-structured prompt plus a thin layer of engineering. That’s why we’re providing teams with a SOC 2–audited, secure, optimized, production-proven base. So they can build their perfect DevEx platform on top of it. And we won’t just hand over the code and disappear. We’ll provide support, advice, and even forward-deployed engineers to help teams ship the custom features they need. Reach out if you want the perfect DevEx platform for your team.

41

5 Comments

Almost every big company I've worked for has a clear policy about using GPL code: don't touch it, don't even look at it. It's not a stand against open source overall. The reason seems to be a fear that if virally-licensed GPL code was ever found inside a proprietary product, not only could the product's source code be forced open under GPL, but any related patents too. IANAL, but that's what I understood to be the case. So with AI coding assistants, are people concerned enough that they might accidentally use a bit of code that is too similar to GPL code -- assuming the training set didn't exclude GPL code up front, so any similarities could be considered coincidence or common practice? FWIW, of all the ethical considerations of AI in the last 3 years, I find coding assistance to be less worrying, mostly because so much of the training set was intentionally released in public to help other programmers save time, solve problems, and write better code. OTOH, StackExchange was the previous winner of the "most helpful" badge in that regard, and that's kind of been usurped by AI. Whether it's fully legal or not I can't say. What I think is still missing from the LLM-style AI world is proper attribution tracking for source material. If a third party company can find correlations between public sources and final AI outputs ad hoc, it seems like the AI companies can better figure out how to carry attribution metadata through the transformer models to build a list of credits at the end, down to some minimal level of contribution, below which nobody cares. I see some third party services that scan codebases for similarities to GPL code to mitigate licensing risk specifically around AI coding, where in the past it was more about employees who maybe didn't follow company policies.

22

7 Comments