Show us what good looks like

“Are we using enough AI?” is a question with an unlimited appetite. Give it a better question: what does good work look like here? Google’s usage research finds plenty of collaboration, while Salesforce describes the effort of giving engineers shared expectations. We like the connection. A useful AI practice needs a purpose people can recognize and a result they can explain.

Collaboration is doing plenty of the work

Google’s first AI & Economy ATLAS study classifies fewer than 10% of observed workplace interactions as fully automating a task, with most focused on collaboration and assistance.

Our take: Helping someone think through a problem deserves a place in the adoption story. A useful conversation can change the decision that follows, even when no agent clicks “submit.” We would ask people to bring examples of what became clearer or easier. A count of automated tasks misses that kind of contribution, and Google’s product sample doesn’t tell us how every workplace operates.

Salesforce gives engineers a shared definition of good

Salesforce’s engineering team describes a proficiency framework supported by hands-on practice, weekly sessions, and coaching guides, intended as a language for growth.

Our take: “Be more AI-native” gives a manager plenty of room to be vague. “Show me how you check an agent’s work” starts a useful conversation. We like the emphasis on observable behavior and support for the next step. We’d also keep room for different destinations: some work benefits from a thoughtful collaborator without ever needing a swarm of agents.

One useful move

Have an AI draft that’s technically fine but still feels wrong? Give the tool a short example of writing you do like, then ask:

My draft: {{my_draft}} Example to compare against: {{example_reference}} Compare this draft with the example. Identify three specific differences in how they communicate. Revise the draft to better match the example’s clarity and level of detail, while keeping the original meaning.

Read the revision and keep the changes that serve your reader. You now have more useful language than “make it better.”

Something fun

Ask your chatbot to invent the judging criteria for an outrageously specific village competition:

Write the rules for the annual competition to pack the world’s most emotionally supportive lunch. Give the judges three categories, one disqualifying offense, and a surprisingly bitter controversy from last year’s final.

Our money is on a sandwich that believed in you until the quarterfinals.

Built in the past two weeks

We prepared a workshop on building a personal AI toolkit, with live examples for writing reusable instructions and creating an AI sidekick for recurring work. The facilitator materials include prompts, a test scenario, and backup plans for a demo that refuses to cooperate.