David Golden

One of the skills of a great software developer is being able to read code you’ve just written with the eyes of someone who isn't intimately familiar with it.

The work of software dev has switched from mainly "how do I make it work" to mainly "how will I know it works?" and I am here for it. This part of the problem is hard and juicy.

@emollick Features is one thing. Scaling operations is something else and I'm not impressed. And there are important enterprise features deep in their roadmap while they chase the public hotness. Not a lot of companies in the economy are in such demand to pull that off sustainably.

The ability of the Claude team to learn from things like OpenClaw and implement features like this on a daily basis is a very strong argument that, for AI-powered coding teams, a very different software development process is possible, with large strategic implications.

Using Claude Code, I deployed another distributed systems algorithm visualization. Interactive single-page Paxos consensus tutorial game. Players try (and fail) to break the safety invariant by killing and delaying nodes. What is your high score?

@catehall FWIW I started allergy shots a year ago. Bad timing already in spring but it seems to have helped some by the fall. Real test coming this season. Worth exploring for the longer term if you haven't.

My allergies have been terrible this year & I'm worried what they're doing to my vocal cords will make my audiobook bad. I'm doing all the things to address but want to try a course of prednisone as a backup--anyone know of a telehealth provider that might be open to prescribing?

I've used em-dashes my whole life — they add rhythm and grace to writing. But now they're an AI tell. Can we get a grandfather clause for those of us who were fluent in em-dashes before ChatGPT launched in November 2022?

This article explains clearly why the proposed wealth tax would be so damaging to California. It's not an accident. It's *designed* to be damaging. astralcodexten.com/p/seiu-delenda…

@staysaasy Just wait.

Your late 30s is when you find yourself in a room of peers and think jeeeze these people look old.

@TheZvi Good for? Replacing Haiku in Claude Code so Opus stops kneecapping itself delegating to a toy model.

Sonnet 4.6 reaction thread. What is this good for? What is it not good for?

Inspired by @ramit's Briefcase Technique youtube.com/watch?v=NViqHC…

I've been asked for job hunting advice lately, so wrote up my thoughts on interviewing in particular. xdg.me/hacking-the-jo…

@emollick I think it's emblematic of what the pressure to ship faster does. They aren't stopping to think about what they should be shipping, just consuming tokens as fast as they can in a Red Queen's race. A lesson for everyone else.

I am sure they are optimizing/growth hacking or something, but there just is no way to onboard yourself if you are a newcomer trying to understand these tools. And it is like the companies are suggesting you ignore them, when some are vital (Canvas on Gemini) or Deep Research

If you haven't hit the little plus button on your favorite chatbot recently, it is a complete hodgepodge mess that nobody not on X would understand: Canvas and Web search and Learning and confusing icons. I am not even calling out any of the Big Three directly, its all of them.

@js4drew I think that could work for some domains, but for others, I wouldn't trust it to be deterministic. Challenge I see is knowing in advance whether such a detailed spec is less time than just coding it by hand.

@xdg my framework for AI coding is giving such clear instructions the agents output is deterministic… and religious use of version control

This echoes my experience and is the art vs science aspect of AI coding. By intentionally not looking at side project code and seeing where I hit the wall, I'm getting a better feel for how to manage it. But it leaves me skeptical of the Dark Factory pattern for many domains.

@TheZvi If it's not too late, others in company Slack has similar reactions: "it tries to frontload a LOT of thinking and tries really hard to one-shot codegen", "feels like a completely different and less agentic model", "I have seen it spin the wheels on the tiniest of changes"

@TheZvi Feels off somehow. Great in chat but in the CLI it gets off track in ways that 4.5 didn't. Can't tell if it's the model itself or the way it offloads work to weaker models. I'm tempted to give Codex or Amp a try, which I never was before.

Going through the reactions now, so bumping this for last minute entries.

I think agentic AI would work much better if people took lessons from organizational theory, which has actually spent a lot of time understanding how to deal with complex hierarchies, information limits, and spans of control. Right now most agentic AI systems seem to pretend that models have basically unlimited ability to manage subagents when that is clearly not true. We need measures of spans of control for AI. A human tops out at less than 10 direct reports. I am pretty sure that 100 subagents is too much for an orchestrator agent - suspect we need middle management agents (yes, I get it, insert middle management joke here). Similarly, we need more attention to boundary objects. These are what is handed between groups (marketing to IT to sales) in organizations to convey meaning as a project crosses group boundaries, like a prototype or a user story. Right now agents pass raw text & maybe code back and forth. Structured boundary objects that multiple agents of different ability levels can read and write to would solve a huge number of coordination failures & reduce token use. I also think aboht coupling, which is how tightly units inside organizations are bound. Most agentic systems are either too tightly coupled (every step needs approval) or too loose (Moltbook). This tradeoff is well-studied in organizations, I bet a lot would apply to agents. Other known issues like bounded rationality also apply, I suspect. Everyone is rushing towards the (terribly named) agent swarm, but the issue won’t just be how good the model is, it will be org design choices. I am not sure the labs see this, but we definitely need a lot more experiments with organizing agents done by people who understand real coordination issues.

This is a petition to get @AnthropicAI / @claudeai code to support the .agents/skills standard, just as these have: - amp - codex cli - factory/droid - gemini cli - gh copilot - kimi cli - opencode - vscode - etc Like this post if you want them to support .agents/skills ✊

@umachingunde "Dear boss, when you Slack from your vacation about non-crisis matters, it sends an unhealthy signal that you value face time over work life balance. This encourages a burn out culture that will hurt retention."

I hate the Slack message ‘Hey! We love seeing you, but aren’t you supposed to be offline’ … it’s the most useless and passive aggressive notification …

OK this story was from Wealthsimple, I can now confirm with a name. Their CTO pulled the trigger, and it's going great so far. Respect. Deepdive on them (and 9 other companies in how they choose new tools): newsletter.pragmaticengineer.com/p/measuring-ai…