Beyond the Chat Box
Start where the work
already lives.
The cheapest AI win on most teams needs no new tool and no engineer. It needs someone to decide what the AI can see and who checks what it says.
Most AI conversations I walk into start with a shopping list. Which tool, which plan, which one everybody else bought. I start somewhere else: where does your team's work already live? Help that shows up anywhere else gets ignored by Wednesday.
This year I set up a shared AI assistant for a team that talks in Slack, tracks work in Jira, keeps customer records in HubSpot and keeps its hard-won know-how in internal documents. For several months, multiple people on that staff asked it the kind of questions that used to take three open tabs and a coworker's memory.
The human part
I didn't write code. I made decisions.
The plumbing is solved. The judgment is the job.
I want to be exact about this, because it's the whole point of this newsletter. The assistant ran on connectors, the official hooks those platforms offer so an AI model can read from them. Anybody can switch those on.
The work that mattered was everything around the switch. What should the assistant be allowed to see? How would people actually phrase their questions on a busy Tuesday? Which answers needed a second look before anyone acted on them? Those are leadership calls, and they're the reason a tool gets used or gets abandoned.
Where the line sits.
What the plumbing handles
Pulling the ticket, the Slack thread and the customer record into one answer. If your team still copies and pastes between three tabs to answer a routine question, that's the cheapest win you have right now, and it doesn't take an engineer.
What still needs a person
Connectors fetch anything they're allowed to touch. They can't tell which document is out of date, which customer note was written on a bad day or which thread was a joke. Every connected source needs a name next to it: the person who knows whether it can be trusted.
Same rule for anything that leaves the building. An assistant drafting a reply to a customer is a fine use. An assistant sending one without a person reading it first is a decision your team should make on purpose, out loud, and probably not in month one.
What I can't tell you
I don't have a clean count of how many people used it or how often, so I'm not going to invent one. What I can tell you is that it connected those four sources and that several people relied on it for months. If a vendor hands you a confident adoption number three weeks into a pilot, ask them how they counted. Honest evidence beats an impressive slide every time.
Build your first pilot in sixty seconds.
Try it this week.
Pick the one question that means checking two systems. Give an assistant read access to just those two. Have the person who usually answers that question check every reply for a week.
The question
What's the question your team answers by hand every single week?
Ciao, Jason
AI amplifies my strengths and minimizes my weaknesses. It can do the same for a team, as long as somebody decides which weaknesses it's allowed to cover. Reply with your weekly question; the good ones shape the next edition.
Jason Padgett is the No-Code Chief AI Officer at Chakra AI Solutions, co-founder of Wooden Boy Industries and host of the AGI Podcast Network. Previous edition: Surfing the Gap.