The importance of looking after your LLM's context window

When you send a chat message to a large language model LLM the text you enter gets converted into a list of numbers each one representing a fragment of text The LLM then generates a response as a list of numbers which then get converted back to each nu...

Is it worth learning a new programming language now that AI is here?

Cost of writing code with AI is now practically zero If a coding agent can write code for you in any language do we really need to learn a new language What would be the point In order to validate the output of AI we have to know what we are looking at...

Think about the feedback loop

Coding agents work best when there is a feedback loop that allows them to check their results In software engineering this typically means some sort of testing When working with your agent to add a new feature also work on the feedback loop the testing...

The importance of automating the right things

The danger of automation is that if you are not careful you can end up automating things that just don t matter as well as the things you do The by product of this is that you can end up creating a lot of work for yourself that doesn t contribute to yo...

One day at a time

Today I am reminded that everything that can be achieved can only ever be achieved one day at a time The advice isn t mine it comes from the AA https www aa org and it is something I regularly come back to and reflect upon when life gets difficult When...

The trust ladder: how much do you trust your AI agents?

There is a lot of buzz about AI software factories at the moment One of the big questions in this area is how can I trust the code is good when no human has reviewed it There is an interview between Lauren Tan aka Poteto and Matt Pocock that has some g...

Work on your workflow, not in it

Michael Gerber wrote in his book The E Myth Revisited Work on your business not in your business The basic idea is that you should build your business as a system and then get other people to run that system for you He said you should build your busine...

Put your tickets with your code

One of the core ideas of my skills repo https github com jodysalt skills is that all the tickets the type that would normally get added to Jira or GitHub Issues now live in your git repo as simple markdown files This is done for a couple of reasons 1 T...

The 80/20 principle in software

According to Vilfredo Pareto https en wikipedia org wiki Vilfredo_Pareto 80 percent or more of outputs come from 20 percent or less of inputs This is known as the 80 20 principle sometimes called the 80 20 rule The universe is as Richard Koch https ric...

Begin with the end in mind

Inspired by Tim Ferriss https tim blog and his approach to product validation I have added a new skill to my skills repo https github com jodysalt skills to help me micro test the market before actually building the product add landing page Here is an ...

Give your coding agent a glossary

Mixing domain language can cause confusion and muddy the picture Here is a good example Imagine you are a software engineer and I ask you to build a DB transfer management system Your first thoughts might be that you need to build some sort of database...

You might not need MCP

Today I have been tuning a spam filter on Blognami https blognami com and I wanted to make it easy for Claude Code to analyse spam that has slipped through the net My initial instinct was to bolt on a Model Context Protocol https en wikipedia org wiki ...

Give your coding agent a sandbox

Today I have added a new run in sandbox https github com jodysalt skills blob main skills run in sandbox SKILL md skill to my skills repo https github com jodysalt skills Its primary use case is to make it super simple to run tests potentially in paral...

We live in strange times

We live in very strange and uncertain times Within the last year I have gone from just tinkering with AI to using it as a core part of my dev workflow Large language models have gotten really good at doing a version of what you tell them The challenge ...

Challenging assumptions in AI

Coding agents such as Claude Code are very good at producing a version of what you want When coming up with a design for a solution the underlying large language model LLM is going to make a bunch of assumptions derived from the collective opinion of a...

What doing burpees teaches me about systems

For about a year and a half now I have been doing burpees as the core of my fitness regime For most of that time I have been following the Busy Dad Training https busydadtraining com program though latterly I now do a modified version Here is an overvi...

Inline-executed metaprompting

I want to share an idea I have been playing with There isn t a standard term for it but for now I am calling it inline executed metaprompting I am a big fan of metaprogramming https en wikipedia org wiki Metaprogramming I love languages like Ruby and J...

Tactical vs. strategic programming

I have been thinking a lot latterly about what we as software engineers should be doing going forwards in a world with AI In his book A Philosophy of Software Design John Ousterhout describes two types of programmers 1 The tactical programmer someone w...

Chain-of-thought prompting

You can make your AI model produce much better results by forcing it to show its intermediate reasoning before giving an answer much like asking a maths student to show their workings This is known as chain of thought prompting You can read a paper abo...

3-paragraph challenge

I have been thinking lately that I am getting far too caught up with trying to make my posts perfect to the point where it has been stopping me from posting It is like each post must encompass everything all at once and be worthy of some self imposed i...