tl;dr Tell your agent to be your most resepected professional Models are trained on good and bad stuff, which is why they’re not perfect Getting your model to impersonate someone who is a master pays off Story In my previous post (test coverage with agents) I mentioned that it might be wise to get the agent to write tests using TDD like Uncle Bob. That got me thinking - what if we instruct our agents to behave not like some imaginary software engineer, but like one of the pros who has a lot of published stuff.

So you’re using agents to build your project now. There’s a good chance you’ll never need to write a line of code again, and soon you won’t need to look at code again.. Where does this leave the code? Perhaps we should let our agents code in the way that’s best for them. What style of code do agents understand best? Which language do they work best with? When we ask our agents to write code perhaps we should choose the most successful ai coding language (Python) and tell our agents to code it in a way that agents can read and understand better than humans (more verbosity).

tl;dr Tell your agent to use GitHub MCP (or just the gh command line tool) Instruct it to get all the PR comments for your project, or just a selection like this: For the current branch, please apply the teachings from AGENTS.md where they make sense. Note it could be no changes are required. Thanks Now the agent has learned from your pull requests and will code more to your teams preferences, thereby making future PRs easier to accept!

Now that none of us write code any more, we’re all thinking about the best ways to get agents to do it all for us. In order to ensure our agents do a good job we need to have quality standards in place. One standard part of software development is writing tests. Back when I was a software engineer full time (now more of a devops type) I worked on many different projects where we implemented writing unit tests.

Today I’ve been working through React with some students at Manchester Digital, and I had a hard time trying to explain some of common syntax. In particular, callback functions. <button onClick={() => setCount((count) => count + 1)}> Personally, I struggle with this kind of minimal syntax until I’ve written it a few times myself. I had the same issue with PHP when they introduced all the shorthand conditionals and that kind of thing, and the same with terraform and pretty much every language under the sun.

My homelab People have shown an interest in how I run my stuff so I thought I’d open it up here. From a hardware perspective it’s very simple (and cheap) - intentionally, I have better things to spend money on like food. The software stack however is a bit more involved. I’ll go through it all here. The hardware setup Connection to the internet I have a gigabit fibre (FTTH) internet connection with a static IP address and (recently) IpV6 support.

It’s been quite a rocky year after switching jobs but one of the upshots is that I’ve managed to hopefully have finally sorted out this blog. You’ll perhaps notice all the old posts have finally come back since March. Hopefully some of it’s still of some use. Over the next few days and weeks I’ll be making some posts about what I use my new “homelab 😂” for: Personal knowledge management (PKM) How this blog is automatically updated from the PKM (fun) Media server Online book reading Webserver(s) Etc!

Earlier this week, I had an interesting problem someone asked me to solve. A custom analytics tool had been set up which was gathering lots of data at a rate of about 10 rows/second. The analytics table had 5.3 million rows, was innodb, had two fields which were nested JSON (stored as TEXT) and NO INDEXES. The customer wanted to be able to run reports on this table and was unable to as mariadb would just lock up (as expected) when trying to use any kind of clause.

john-hunt.com has been a wordpress site since the early 2000s, mostly a blog of interesting things I’ve discovered during my devops/software engineering career. I’ve recently switched over to using hugo to generate a statically hosted website rather than dealing with security updates that wordpress requires. Hopefully over the next couple of weeks I’ll migrate the old posts over as there were a couple of posts that generated quite a lot of interest (notably the external USB data recovery one.

So for a long time, I’ve been wondering why I can’t use a locally mounted directory within one of my containers - I’d just get an empty directory on the volume mount point inside the container. A lot of googling yielded nothing, but then when I ran ps -aux | grep -i docker I noticed docker was running via snap. It then occurred to me that snap has restrictions on what parts of the filesystem it allows it’s ‘snaps’ to read - this indeed was the problem so a quick uninstall of the snapped version of docker and installed via apt fixed the issue once and for all.