The Drip
Google is bringing Gemini Spark into Chrome, where it can work through logged-in accounts and saved context to handle errands such as researching flights. The capability itself is not new. The interesting part is that the AI is meeting people inside a product they already use.
That also sharpens the practical question. Just because AI can book a flight does not mean handing it the task creates a better experience. A quick tap through an airline app can easily become a long back-and-forth about dates, seats, and timing.
Inside The Bottle
AI activity is everywhere. Seats are provisioned. Prompts are counted. Tokens are consumed. Teams can point to dashboards and say people are using the tools.
That is useful information, but it is not the same as evidence that the work improved.
We have seen processes where AI was added and the execution barely changed. In some cases it got worse. The goal is not to force every project into a narrow ROI calculation. It is to step back and name what should actually be different when the work is done.
The speed of the market makes this harder. Everything around AI feels like it is moving at warp speed, which creates pressure to deploy just as quickly. But the more reliable path is usually smaller and slower: choose a bounded part of a workflow, give it the right context and rules, keep useful review checkpoints, and learn where the exceptions live.
That process builds a mental model for the humans too. A team often discovers that it does not understand its own workflow as cleanly as it thought. Unwritten exceptions, judgment calls, and handoffs appear as soon as the work is described precisely enough for an AI system to run it.
A few questions help us keep the work grounded:
| ■ | What specific part of the workflow are we changing? |
| ■ | Where does human judgment add the most value? |
| ■ | What context, rules, and review checkpoints does the system need? |
| ■ | Who owns the result and the exceptions? |
| ■ | What should become meaningfully different if this works? |
Activity can show that behavior is changing. Strategy connects that activity to the work the company actually cares about. Ownership makes it clear who is responsible for the outcome.
The technology is capable. Most of the hard work now sits around it: choosing the right boundary, making the strategy real, and giving people a clear picture of what AI is—and is not—supposed to do.
Lab Notes
| ■ | Justin's note: Getting out of the chat and thinking through the actual process is still the useful move. The question is less “can AI do this?” and more “what becomes different if it does?” |
| ■ | Kellan's note: Look for the step where your brain is most engaged. That is often where the human should stay. Automate the report pulling and CSV work, then move the saved time toward the part where judgment adds value. |
What Stopped Our Scroll
|
