The Bram binary now embeds SQLite with its FTS5 fulltext indexer and search engine. When you run Bram in a local GitHub (or GitLab) repo, here is what it indexes: – the JSONL session files written by Claude Code and/or Codex – the worklist items written by Bram – git commits – issues posted to … Continue reading Make agent memory searchable

Agents like Claude Codex and Codex run in your terminal. From that strategic vantage point they can wield system tools (awk, bash, curl, git, python) as well as MCP tools. Here’s how that looks in the terminal. ⏺ Bash(awk -F’T’ ‘/2026-07-23T05:/’ /Users/jonudell/bram/resources/bram-traces/bram-trace.log | grep -E “describe-patch|heartbeat-batch|reason\”:\”subscribe” | tail -12 | cut -c1-195) In What is … Continue reading Agents that narrate their work are the best team players

Handwriting was always problematic for me. My fifth-grade teacher, Mrs. Cloud, placed a high value on well-formed cursive strokes that she flowed smoothly onto the blackboard. At my desk I struggled to copy her examples and failed miserably. In middle school, with no one judging my handwriting, I abandoned cursive in favor of printing my … Continue reading Talking to Claude Code and Codex

In this snapshot of the Bloomington calendar you can see that events are neatly categorized. This was an intractable problem a dozen years ago. Should the county fair land in community / social or family/kids? It’s not a critical choice, and as a user of the calendar you’d accept either. For the calendar’s curator, though, … Continue reading Small models can solve big problems

Agents are hardwired to be prolific writers and readers of code. As my work on Bram progressed I found that their code-first instinct wasn’t serving me well. So I began pushing them to be, also, prolific writers and readers of logs. Bram is a Tauri app, so it’s written in Rust. But it’s also a … Continue reading Don’t infer behavior from code, observe it in logs

In his keynote talk at the first Perl conference, Larry Wall couldn’t get the Windows computer on the podium to behave. So he SSH’d into his own machine and said, with relief and joy: “Home sweet home”. Three decades on, software developers still live in the terminal, now more than ever as coding agents dethrone … Continue reading “What is the terminal?”

I recently attended a talk, by an engineer at a large software company, on the topic of unreviewable PRs. The problem? When agents raise PRs with thousands of lines of LLM-written adds/deletes/edits, people can’t make sense of them. The solution? Throw more agents at the problem: reviewer agents that scan what coding agents have produced, … Continue reading “Doctor, it hurts when agents create unreviewable PRs.” “Don’t do that.”

In Working With Intelligent Machines, written at the beginning of my AI-assisted coding journey, I quoted from Garry Kasparov’s The Chess Master and the Computer. The winner was revealed to be not a grandmaster with a state-of-the-art PC but a pair of amateur American chess players using three computers at the same time. Their skill … Continue reading Vibe coding as a team sport

I’m working on a new tool whose tagline is the title of this post: Make best use of git and GitHub for AI-assisted software development. Called Bram (“Bram runs agents mindfully”), the tool runs as a Tauri desktop app with three panes: a terminal where you use Claude Code and/or Codex, an agent pane that … Continue reading How to make best use of git and GitHub for AI-assisted software development

I had an idea about 15 years ago that I wound up pursuing a lot longer than I should have. Near the end of that era I read an essay by Seth Godin called The Dip, about that low point when an idea you are convinced is worthy just isn’t taking hold. How do you … Continue reading Beyond The Dip

I once wrote an article with the optimistic title GitHub for the rest of us. The idea was that everyone who works with others on collections of shared documents needs a powerful and easy way to see and manage change. The foundational tool for software version control — git — is incredibly powerful, and the … Continue reading How LLMs make Git and GitHub easier to use and learn

Tools built to generate vast amounts of code can, paradoxically, help us write less of it: How To Use LLMs for Continuous, Creative Code Refactoring LLM series at The New Stack

How to manage a team of AI assistants in a virtuous cycle of improvement. The LLM flywheel effect LLM series at The New Stack

Tuscon’s Museum of Miniatures features hundreds of exhibits like this one. “Artist Madelyn Cook spent over 3 years planning and constructing Lagniappe, which includes two separate wings and 25 individual rooms.” People have been making these for hundreds of years, but in recent decades practioners have become more precise about measurement and scale. Many of … Continue reading Release the Kraken!

Exactly one hundred and fifty years ago John Muir walked around in the same grove of giant sequoia trees that I walked around in today, and stood next to the same two thousand ton behemoth that had been growing for two and a half millenia. It has only been known as the General Sherman tree … Continue reading A day in Sequoia National Park

The Volts podcast continues to be my favorite listen. Climate change will wreak ever more havoc on the world, that’s just baked in. But the transition to clean energy is also now baked in. David Roberts delivers a steady stream of hopeful news on that front: plummeting prices for solar panels and batteries, “reconductoring” to … Continue reading Reimagining car culture

I’m visiting with American friends who are staying in a rural farmhouse in France’s Dordogne valley. The house, which might be several hundred years old, provides faster internet access than my fiberoptic setup at home. The cars we are piloting along these ancient byways have touchscreens that control Bluetooth and satellite connections. It feels like … Continue reading Maps old and new

Although autonomous LLMs are inherently unreliable, there’s a long software tradition of building reliable layers on top of unreliable layers. That applies here too. We can’t guarantee that you’ll never be led astray when building an XMLUI app with the help of agents that use the XMLUI MCP server to extract patterns from docs, sources, … Continue reading Context engineering anchors AI agents to ground truth

In the mid-1990s you could create useful software without being an ace coder. You had Visual Basic, you had a rich ecosystem of components, you could wire them together to create apps, standing on the shoulders of the coders who built those components. If you’re younger than 45 you may not know what that was … Continue reading Introducing XMLUI

We mostly don’t want to read the docs, but we do want to converse with them. When we build search interfaces for our docs, we have always tried to anticipate search intentions. People aren’t just looking for words; they need to use the material to solve problems and get things done. When you create an … Continue reading MCP is RSS for AI

The great adventure of my birth family was the fifteen months we lived in New Delhi, from June of 1961, on a USAID-sponsored educational mission. So the destruction of USAID feels personal. I’m only now realizing that we were there at the very beginning of USAID, during what Jackie Kennedy later mythologized as the Camelot … Continue reading The Musk Massacre

This Fresh Air interview with Hanif Kureishi had me riveted from the beginning, for one reason, and then at the end for a different reason. Kureishi is best known as the author of the 1985 British rom-com My Beautiful Laundrette. During an illness in 2022 he fainted, fell on his face, broke his neck, and … Continue reading Who will take care of you in your time of need?

A few years ago I abandoned Twitter in favor of Mastodon. Recent events validate that choice and underscore the strategic importance of a decentralized fediverse that can’t be owned by a single corporate or state actor. But while Mastodon meets my needs, much of the Twitter diaspora has gone to Bluesky. That’s fine for now … Continue reading Making the Fediverse More Accessible With Claude 3.7 Sonnet

The JavaScript industrial complex won’t crumble anytime soon. But the stage is set for a return to an ecosystem of reusable components accessible to business developers, only this time based on the universal web platform and its core standards. How To Build Web Components Using ChatGPT LLM series at The New Stack

Perhaps, even though they are not themselves explainable, AIs can help us engineer explainable systems. But I’m not optimistic. It feels like we’re on a path to keep making systems harder for humans to configure, and we keep expanding our reliance on superhuman intelligence to do that for us. The Configuration Crisis and Developer Dependency … Continue reading The Configuration Crisis

The first time I heard a critique of mediated experience, the critic was my dad. He was an avid photographer who, during our family’s year in India, when I was a young child, used his 35mm Exacta to capture thousands of photos that became carousels of color slides we viewed for many years thereafter. It … Continue reading The social cost of mediated experience

Earlier efforts to diagram software with LLM assistance weren’t fruitful, but this time around things went really well. I ended up with exactly what I needed to explain the architecture of a browser extension, and along the way I learned a lot about a couple of formats — Mermaid and Graphviz — as well as … Continue reading How To Create Software Diagrams With ChatGPT and Claude

“If you work with these cloud platforms every day, you have doubtless forgotten that you ever had questions like these. But every newcomer does. And on a continuing basis, we are all newcomers to various aspects of applications and services. In so many ways, the experience boils down to: I am here, what do I … Continue reading What Claude and ChatGPT can see on your screen

There are plenty of ways to use LLMs ineffectively. For best results, lean into your own intelligence, experience, and creativity. Delegate the boring and routine stuff to closely supervised assistants whose work you can easily check. Mix Human Expertise With LLM Assistance for Easier Coding Part of the LLM series at The New Stack.

I was aware of The Geysers, a geothermal field about 35 miles north of my home in Santa Rosa, but I never gave it much thought until my first bike ride through the area. Then I learned a number of interesting things. It’s the world’s largest geothermal field, producing more than 700 megawatts. It accounts … Continue reading Geothermal power in the North Bay