Writing
Notes from work I've done myself, with the numbers I measured and the parts that went wrong.
-
Tuning GLM-5.1 on eight H200s: three speculative tokens beat two by 29%
Two speculative-decoding settings and a CUDA graphs run, measured on one stream. Three draft tokens won by 29%, and the acceptance rates I didn't save are the gap in the data.
-
Running 38,150 coding-agent benchmark attempts on Kubernetes
Most of what I learned running coding-agent benchmarks at scale was about bookkeeping, not GPUs or prompts.
-
Our dev database backups were zero bytes for four days
The job ran on time, the files had the right names, and none of them could be restored.
-
Git hooks change PATH, and our gate cache missed 97% of the time
A validation cache keyed on the environment looked fine when I tested it by hand. Inside a Git hook, the environment was different.
-
Spreading CI across four k3s workers took a 51-minute run to 17
Parallelism bought the headline number. Putting jobs where their cache lives and never repeating a passed validation is what keeps runs fast after that.
-
One ARG in the wrong place cost 226 seconds a build
Docker reruns every RUN after a changed ARG. Put the volatile one last, and stop building unchanged images one at a time.
-
Stop letting agents poll: 765 status checks a day cost 1.8 hours
An agent that checks status every few seconds pays for a model turn every time. One blocking call per condition costs one.
-
How I fence the coding agents that work on my production systems
Agents get a place to run, a short list of things they can reach, and no way to change production except through review.
-
Review the delta, not the branch
Second reviews in my agent fleet were no faster than first ones, because every reviewer re-read everything. The fix was to give them a better question.
Get in touch
If something here sounds like your problem, send me a few lines about what you're running.