We run ~15 automations internally. I've lost count. They fire alerts into a Slack channel none of us open, and half could break tomorrow without us noticing for a week.
I gave Grab's new AI some code to run. It ran it, then asked where I wanted dinner.
Funny from a food app. Less funny when it's the AI paying your invoices and a "supplier" emails new bank details.
Wrote the plain-English version of "AI governance": https://t.co/eYhNDeTIs8
Context engineering, RAG, all the expensive stuff. Strip the names off and it's the same job: write down what's true and put it where the machine can reach it.
That anxiety sharpens our mission at Superuser HQ: help traditional companies catch up quickly, as a partner who learns the business, builds alongside the team, and stays accountable for outcomes. Singapore has an execution gap. Time to close it.
30 mins on X tonight was enough to make me anxious. I watched people use Anthropic's Fable to compress work that used to take teams days into hours. Then I thought about Singapore.
The frontier isn't waiting for budgets, committees, or perfect roadmaps. The distance between what a tiny AI-native team can do and how most traditional companies work is widening every month.
Picking the task is half of it. The other half is time, and nobody on the floor has any, so you make it for them. A couple of hours, one person, one dull job, permission to flop. Don't marry AI. Date it first. New series for founders: https://t.co/LqeCAbjrAR
Handing the team access is table stakes now. The hard part starts when the agent can reach into your systems. Suddenly you're asking what it can see and where your data ends up, security questions most of us never had to ask before.
The work that survives is the boring work. For us, the month-end books. Mechanical, checkable, the numbers reconcile or they don't. A good first task is one nobody's reputation rides on, and one you know so well a wrong answer jumps out instantly.