AgentReady prepares and retrofits agentic coding "best practices" onto existing codebases, so that agents have clean and accurate context. This is something everyone has to do, as context engineering has surfaced as one of the most important aspects of LLM usage. Very important for models to generate shippable code.
This tool does a lot of stuff. I have a bunch of cold-start prompts to add a few more features. Much of the rules are subjective and pluggable so if you don't like them, vibe delete them :-)
Jeremy, I tried to comment in the article as to not pollute HN with the usual font complain, but the comment was rejected saying I had to be logged in.
Your text is really hard to read! The "f" looks like a "j"... my brain was all the time "a j is coming" and it made no sense in the word... that's because it's a "f". I had to parse "efficiency" 3 times in the first paragraph because it just reads like "ejjiciency"!
Hi Brian, thanks for that feedback; we are definitely going to improve our documentation and will take this into account. I hope you're able to test out InstructLab and let us know how it goes.
OP here, thanks to everyone for all the comments on this post. They were interesting to read.
I appreciate the positive comments and can also certainly understand why some are not so positive.
For some context ... the blog itself was scoped to a very small portion of what I intend to write as time goes on (probably won't be submitting those to HN as I agree in hindsight that it doesn't match the preferred content for HN readers).
Someone else also commented "Are there any votes you cast that you’re not proud of? What kind of person do those votes tell you that you wish to become?" Yes, I'll be writing that and expanding beyond work scope because as a family oriented person (which doesn't come across in the post, nor did I intend it to), I'm glad to work at a place that advocates for a healthy work-life balance.
Burn out is a very real thing, and I 100% agree with the "Life needs to be built and enjoyed together, through your whole life." that's spot on!
Hey sure, I should have been more specific. The enhanced networking in AWS requires ixgbevf at least version 2.14.2. In addition we had random hardlocks with ext4 under the centos7 kernel. Also we had a lot of difficulties running docker on EBS disks with the device manager driver - hard lock ups randomly. That's all I recall off the top of my head. All these issues disappeared upon moving to Ubuntu 16.04 on a later kernel. Centos7 was great for us otherwise, it just didn't seem to work out so well with Docker + AWS.
...depends on the gear. SMI used to be (4-5 years ago) a much larger problem than they are now.
I agree with you in that context, and it's why so few systems are certified for Red Hat's Realtime kernel. They are simply not all created equal.
But I'd encourage you to review the results of any of the 25+ benchmarks we did with STAC over the last few years.
We didn't see much (if any) SMI interference on the gear we had, which was off the shelf regular servers, with WSM, SNB, IVB and HSW. All the hardware, software and config is disclosed within those benchmark write-ups.
There is some tooling called hwlat that can detect and report SMIs. It's in the rt-tests package.
Although some of the tuning there-in can be used for improving performance on any workload (handling NUMA, for example), it's probably not necessary for the majority of environments because it involves intimate knowledge of hardware, software and application stack.
It also talks about disabling a bunch of power management which is really only necessary when you're chasing microseconds.
BTW if you're interested in NUMA and memory management on RHEL7, my team mate Bill Gray wrote an awesome whitepaper:
Hi guys, thanks for the feedback. I updated the blog post to include a table comparing different versions of RHEL, along with tick behavior you should expect (keys off nr_running).
Hi Jeremy. The articles on your blog are great. I wandered across this one while trying to figure out the best practice for isolating a core to reduce variability when benchmarking CPU intensive tasks. Your tickless explanation helped a lot. Hope you stick around HN. Thanks!
This tool does a lot of stuff. I have a bunch of cold-start prompts to add a few more features. Much of the rules are subjective and pluggable so if you don't like them, vibe delete them :-)
https://github.com/ambient-code/agentready