AI
-
🔍 It is often more work to review something than to do it
As an experienced person, I am sometimes called on to review something. This can be a frustrating, seemingly hopeless task. With one developer, I commented that he could write code faster than I could review it. While there are some issues …
-
🔍 AI coding requires better inputs and robust verification
Some great feedback on the previous post (Impressive is not always effective): “I agree. It would be interesting, however, to compare the progression of AI code generation with that of compilers - were there star assembly-code engineers sa…
-
✨ Impressive is not always effective
When AI generates something, the result is often “impressive.” We might marvel at how quickly lots of code is generated that mostly works. Generated documentation is extremely detailed and well formatted. AI-created diagrams are colorful, u…
-
🥄 The beauty of human creativity
In this age of machine-generated perfection, something is still deeply rewarding about grabbing a stick of wood and making something by hand. This past Christmas, with the help of kids and grandkids, we made a few mixing spoons for gifts. S…
-
🚀 Platforms are not just for developers
This is an interesting post from a Marketing person creating website pages: I built 19 website pages in a single day while my team was out for US thanksgiving. Not mockups. Not placeholders. Actual, shippable pages: a full partner ecosyste…
-
🔄 Breaking the AI slop cycle
“AI Slop” is a common expression describing a familiar phenomenon. It refers to sloppy code (or other assets) generated by AI that mostly works, lacks proper review, and may be poorly architected or implemented. The primary characteristic o…
-
📝 DDPD - Document Driven Product Development
Building on TDPD and RDPD, Document-driven product development (DDPD) offers another perspective. Ideas: Document the feature from the user’s perspective before implementing anything. Keep the documentation close to the code (same repo, et…
-
🛠️ The new Artisan Age
We live in a world of powerful tools. Milwaukee battery-powered hand tools, high-end PCB design suites, cloud platforms, open-source tools, etc. And while these tools get us a good way toward the goal, they do not do everything. Additionall…
-
💻 Local First Dev
In the last couple decades, we’ve observed the success of SaaS web applications (Google Docs, Salesforce, Basecamp, Trello, etc.). These applications have unleashed collaboration using the web platform. Before this, users might have emailed…
-
🧠 Context is everything
In conversation, I find myself saying: “what is the context of your comment?” At times we find ourselves responding before we really understand something. AI does the same thing – it just spouts off, even if it does not understand your cont…
-
💡 Reflections from AI-coding a 10,000 line app
Lessons from AI-coding a 10,000 line app BRun is getting usable. Claude Code wrote almost all of it using Doc-driven development (manually write/update the docs, and have Claude write the code). Some thoughts: Claude is fast – there is no …
-
📚 Documentation never has to go out of date again
AI is really good at updating documentation after implementation. While I strongly advocate writing the documentation to drive AI, inevitably, the implementation takes some unanticipated turns, so making sure everything is in sync after imp…
-
✉️ Drafting emails in Markdown
As developers, we use Markdown all the time: in documentation, GitHub issues, etc. Pretty much every tool supports it now, except email. With email, we are still stuck with a 2005 MS Word experience. Why Markdown? Once you know the simple s…
-
🧊 Web application architecture
The UI is the part you see. The boring parts underneath determine whether you end up with a sustainable product.
-
🔩 Generating Mechanical Models with Claude Code
We recently needed a mechanical model of a USB coupler. On a whim, I suggested that the designer try using Claude Code to create it in FreeCAD from a 2D drawing - it worked! No MCP or complex setup needed. (Although there is an MCP availabl…
-
💻 Why the terminal still matters
Starting in the 1970s, the terminal, popularized by Unix, has been a mainstay in computing. Many developers today, and increasingly non-developers, are using the terminal. In previous years the terminal was defined by a physical monitor and…
-
🛠️ Tools or Processes?
What matters more – having better tools or better processes? One tends to enable, the other to restrict. Companies often split into tool-builder cultures (optimized for enabling engineers and teams) and process cultures (optimized for contr…
-
🔧 What has changed around edge build systems?
Reflecting on Embedded Linux/Edge build systems, several things have changed: ARM and RISC-V hardware is fast. Modern ARM boards and cloud instances (AWS Graviton, Hetzner CAX) build at speeds that make cross-compilation unnecessary for mo…
-
🔧 Why Go Works Well for Building CLIs
The last post discussed the importance of command line interfaces (CLIs) in the AI agent stack. The next question is — what language works best for building CLIs? Consider the requirements: Often a CLI needs to interact with APIs. Easy to …
-
🔌 API + CLI + Skills = the Future of AI Agents
DHH’s recent post offers a glimpse into the future: “So while we keep cooking on actually-useful native AI features in Basecamp, we’re launching a fully agent-accessible version today. We’ve revamped our API, created a brand-new CLI, and w…
-
👥 The release is a language everyone understands
A release is a good way for developers to communicate with everyone else: What got done What changed How to use or build the latest version Where to get support or provide feedback A regular cadence for communication A standard format ever…
-
🐛Kernighan's Law
Kernighan’s Law: Debugging is twice as hard as writing the code in the first place. Therefore, if you write the code as cleverly as possible, you are, by definition, not smart enough to debug it. Cliff’s Law: In three months, it will be thr…
-
📝 Document "why" not "how"
Be kind to others (and your future self) and document why you did something. AI is pretty good at generating the “how” any time you need it, but only you know the “why” because you have that context.
-
🔄 Close the loop with AI
There is a fascinating statement in a recent MIT study that claims 95% of organizations are getting zero return from GenAI investments. The core barrier to scaling is not infrastructure, regulation, or talent. It is learning. Most GenAI sy…
-
💰 Ways to reduce your AI costs
AI is not free, as there are significant costs associated with training and hosting modern AI models. Some ideas on how to reduce costs with tools like Claude Code: If you do something that is somewhat repetitive, have AI generate scripts …
-
📁 The file dilemma
There is this movement against files in most disciplines. Files are hard to manage, track, version, and secure (unless you use Git). Thus, a lot of tools have moved to SaaS (software-as-a-service) applications in the cloud. Jon Hirschtick a…
-
🔄 AI and documentation - close the loop
Starting with good documentation is often a better way to use AI. However, you can also close the loop and have AI update the documentation when it is finished. It can fill in details, clean things up, etc. This closes the loop. Then repeat…
-
🤖 AI is the new compiler
There are two ways to use AI: Start with a vague prompt and keep prompting until we get the result we want. Point the AI at documentation, and keep updating the documentation until it is clear enough that AI can do the right thing. Even t…
-
✨ Anyone can write scripts now
With Claude Code (and likely many other AI technologies), anyone can write scripts now. Instead of writing in a technical language like Python or Bash, you can now write a sequence of commands with some rules in plain English, and Claude tr…
-
🤖 How I use Claude Code to help publish this blog
I have been using Claude Code to help with the following tasks on this blog: proofread (including detecting passive voice) add tags add an emoji to the title These are not rocket science, but they also take time and are something AI does …
-
🔭 Text based formats and command-line tools are the future
For the past three decades, we have been told that graphical user interfaces (GUIs) will revolutionize computer use and even programming. In many ways, they have made computers very accessible for the average person – especially phone UIs. …
-
🔄 Flipping the playbook for AI coding
A lot of focus these days is on “good AI prompts.” Some of the best prompts may be documentation and tests. Write your CHANGELOG.md entries first, then update the documentation, then write the tests, then tell the AI to update the code to m…
-
🧑💻 The first step in AI coding
✔️ Should be “How does this work?” ❌ NOT “Implement this feature.” Understanding should precede any work, especially in complex systems.
-
🤖 The biggest problem with AI
“It ain’t what you don’t know that gets you into trouble. It’s what you know for sure that just ain’t so.” - Mark Twain (maybe) AI is really good at summarizing existing information. It is less good at thinking, although it can do some ama…
-
🤖 What is AI good for?
AI is really good at summarizing existing information – Perplexity Pro mostly has replaced Google/Duckduckgo, etc., and is worth every penny. AI is really good at finding API documentation – much quicker than searching through docs, and oft…
-
🧑💻 Will AI cause coding to disappear?
AI is not a binary off/on thing, like many of the industry commentators like to think. Many of us are already using AI to improve our productivity. For me, Perplexity has largely replaced Google search. Technology has always brought product…
-
🤖 What is the role of AI in your platform?
We hear much today about how AI is going to do our jobs better than us. AI is powerful and has access to a lot of information and does a lot of things humans can’t do. It is really great at writing shell scripts, boilerplate in programs, fi…