Recent content on hank.bond
Models are loosely a compression of all of human knowledge (obviously this depends on their training corpus and parameter size etc.), so they have pretty much every software pattern stuffed in them. If you were to ask a colleage “what pattern should I apply to this problem?” they can only point you towards the patterns they are aware of. That colleage will hopefully also be able to…
This one is a bit of a ramble but I have some loose questions bouncing around rn. Who owns the Agentic capabilities you create for yourself in the course of working somewhere? When you learn something during your day job, it’s tacitly agreed upon that the knowledge in your meat machine is yours to keep. It’s not like the company has direct access to it anyway, they have to use You as…
Creative vs Assertive This is not really a hot take, but I was just thinking about how in The Before Times we already had different developer setups when we did work. I’m pretty basic and just raw dog VS Code without any meaningful extensions. Plenty of coworkers have used JetBrains flavors, Vim, EMacs, some even use Sublime Text because they like the simplicity. The human aspect of the code…
Unlike when I am prototyping a solution using tools, concepts, and frameworks that I already understand , when I am trying to work in an area where I cannot be a reliable judge, I tend to iterate much more carefully. There’s really two frames that I am under when working with Models and they are mostly mutually exclusive. I try to increase either speed of: Output artifact creation by…
The last year has been pretty wild for the software industy. As of April 2026, this year has had 33k job cuts (if you ask Challenger, Gray & Christmas, who supposedly have been providing layoff stats for quite a long time [@challenger-apr-2026].) The common theme from the CEOs here is that AI is the main motivator (or enabler) of these layoffs. If we take them at their word they are either
I was thinking about how to properly communicate the value of Software Engineering as a discipline, in the era of rapid LLM code generation Me as an individual, basically “why should you hire me specifically?” In The Before Times, the answer was vaguely “I solve your business problem using code”. This value was pretty clear to Business People, in that they would want a…
Was talking with Chat about whether or not we were maximally using HTML tags that are more semantic in nature in this site’s templates and processing. Given that this is currently a Hugo based site, and its pretty easy to improve HTML usage with automation, I thought it would good to explore this. Here are some cool (by my estimation) HTML tags that I was not aware of before. abbr is for…
To create great agents/skills: start very small and very simple with broad language telling it what you want (not how to do it), use it as-is, pay attention as it runs, add corrections for where it went wrong back into the prompt, repeat over and over and over. That’s it!
For local agentic workflows, CLI tools composed with unix pipes are leaner on tokens, faster to iterate, and more expressive than MCP. MCP is most useful for enterprises with more robust operational needs. For one developer talking to their own machine, go with the simpler CLI approach.
Everyone has their own Agent framework, SKILLs for each role in a startup, RAG system that “solves the memory problem”. The problem is that its unlikely you will have the same needs, and worse yet
Type Chrome System window labels and tiny controls.
You can ask the model nicely and it will mostly listen most of the time. “DO NOT SKIP THIS STEP!!!” is not engineering. If a regex, a parser, or a tool-call hook can enforce the rule, use it. Need a list of actions to be taken every time ? Make each a tool call <step_one> , with the loop/pipeline not finishing until the final <end_run> tool is called. At the “end” of the…
A useful session should not evaporate when the chat ends. It should produce some durable improvement, a sharper prompt, a better script, a clearer spec, a committed fix, or even just a note about what went sideways. The next agent should start ahead, not at the same blank wall with a fresh cup of amnesia.
Everyone has their own Agent framework, SKILLs for each role in a startup, RAG system that “solves the memory problem”. The problem is that its unlikely you will have the same needs, and worse yet
LLMs can handle ambiguity but are imprecise even in the best case. Your goal is to improve reliability, and so you must aim to use the most consistent tool whenever possible. When using LLMs, push as much functionality as you can into the deterministic layer by codifying functionality with actual code, not LLM instructions.