Follow the real workflow
Not the documented one. Including the exports, the reformatting, and the workarounds people stopped mentioning years ago.
guest@make-directory:~$ cat ./capabilities/automation-development.md
make-directory:~/capabilities/automation
Automation Development
Remove the steps a person should not be doing.
Integrations, scheduled work, and AI-assisted processing that take the repetitive parts of an operation and make them run without anyone remembering to do them.
guest@make-directory:~$ ls ./how-we-do-it
how we approach it
What the work involves.
Cost each manual step
Volume times effort times error rate. Some manual steps are not worth automating, and knowing which is most of the value.
Connect what you already own
Most operations already pay for tools that cannot talk to each other, which is why the spreadsheet in the middle exists.
Automate the repetition, keep the judgment
Rules and thresholds belong in code. Decisions with consequences belong with a person, with the system doing the preparation.
AI where it earns its place
Unstructured documents, extraction, classification. Deterministic code is cheaper to run and easier to trust for everything else.
Make it observable
What ran, what failed, and what needs attention. An automation nobody can see becomes the next thing nobody owns.
guest@make-directory:~$ cat ./what-we-think.md
so you know how we decide
Where we will push back.
A fragile automation is worse than a manual step
A person notices when something looks wrong. A silent job does not. Anything we automate has to fail loudly.
Sometimes the answer is to change the process
Automating a bad workflow makes it faster and harder to fix. Occasionally the right recommendation involves no software at all.
guest@make-directory:~$ cat ./what-this-is-for.md
where this gets used
What we get called in to solve with it.
If one of these describes your situation more closely than the technology does, start there instead.
guest@make-directory:~$ mkdir ./what-needs-to-change
next step
Tell us what you are trying to change.
Describe it however makes sense to you. If this turns out not to be what you need, we will point you at whatever is.
Would you rather look at code?
We build and operate our own software in the open — infrastructure tooling, a modernization toolchain, and a research arm. All of it is public, so you can see how we work before you talk to us.
What we build