What the latest Microsoft Copilot pricing change means for 12-person ops teams.
The numbers will move again. The pattern won’t. Three things for SMBs to do before the next pricing memo lands in their inbox.
Read · →
Field Notes
Working notes from the orchestra pit.
Score reports, Compose decisions, and what we’re learning as we deploy agents into Australian SMBs.
Score, Compose, Rehearse, Perform: the cycle nobody warns you about.
Four phases on the website. The version we tell each other in the office is less linear, and the gaps are where most projects die.
Read · →
What a finance agent actually does between 9am and 5pm at a 40-person SMB.
No vendor pitches. No McKinsey numbers. The actual day, hour by hour, of one finance agent we shipped late last year.
Read · →
We rebuilt an inbox triage at a Melbourne brokerage. Here’s what broke first.
Forty-person team. 800 emails a day. Five things we did not see coming, in the order we found them.
Read · →
When an agent shouldn’t be the answer.
A conductor knows which instruments to leave silent. AI deployments are the same. Three places we routinely tell teams not to build.
Read · →
The integration tax: why most AI projects die in the last mile.
The model is the easy part now. What kills SMB AI projects is the connection between the smart thing and the systems it’s supposed to change.
Read · →
A year on, where Anthropic’s Computer Use actually shipped.
The launch demos didn’t age well. The technology did. Where the agent that drives the screen earned its seat, and where it didn’t.
Read · →
Pilot purgatory: how to spot it before you’re stuck in it.
The most common AI project state in 2026 isn’t shipped and isn’t scrapped. It’s a small monthly bill that nobody’s graduating to production.
Read · →
What we audit first on every Score: the spreadsheet pretending it’s not a database.
Every operations team has at least one. By the time we see it, it’s running a meaningful slice of the business and nobody’s sure who’d be left holding the bag.
Read · →
The ‘agent’ word is doing too much.
When a word means five different things, it stops meaning any. A taxonomy we keep in Score reports to make sure clients and vendors are buying the same thing.
Read · →