Discoveries from running the work.
Notes from the people who run the work and build the runtime behind it. Opinionated on purpose.
- ·6 min read
Nobody wrote down how your company works. So we ask.
Before we run a function, we have to learn how it really works. The plan is a screenshot, the stock is a spreadsheet in a chat and the rules live in two people’s heads. This is how our discovery crew learns it in weeks, with a quote behind every line of the map.
- ·9 min read
What AI did to code, it will do to operations
Coding took its AI leap because the craft already ran on infrastructure an agent could inhabit. Sales, marketing, finance and HR still run on a Frankenstein stack of SaaS silos. The models are ready. The workplace around them is not.
- ·5 min read
Someone has to be on the hook
The blocker to running AI in a back-office function is rarely the model. It is the org chart: who is accountable when a case goes wrong, and what happens to the person whose job it used to be.
- ·5 min read
The hackathon was not the hard part
Companies that ran an AI hackathon and shipped fifteen ideas call that eighty percent done. Price what each idea is actually worth, and the number that is usually left is one.
- ·6 min read
We rebuilt our operator three times. Five things survived.
The runtime that executes Bonofici jobs ran on Cloudflare, then Docker, then Kubernetes, in three months. Most of it was thrown away. The parts that survived are the parts a client should demand from anyone touching their systems.
- ·5 min read
The work that is left is judgement
Every back-office tool is fighting to hold your knowledge. After a year of running the work ourselves, we think that is the wrong fight. The tool is the cheap part. The standard and the judgement are the product.