The Drip
If someone produces better work with AI, how do we know they're getting better at their job?
That was our question as we compared a couple OpenAI and Anthropic studies this week. We care about quality output. We also care whether the person can explain what they did without going back to ask the AI.
Inside The Bottle
We coined a new TBA term in our recording this week, the “AI Trust Bucket”.
People can move from “How can I possibly trust this thing?” to “Nobody's going to need me anymore” and back again in a day. Simply teaching the buttons in Claude Desktop doesn't prepare someone for that experience.
Our trust buckets are pretty full. We also understand what we're not going to put in them.
Justin ran into this while researching a client's public financial statements. AI spotted a trend, decided it was a big problem, and prescribed a Tiny Bottle solution. It looked plausible. A younger version of Justin might have walked straight into the meeting with it. Experience changed the next step: “Are you concerned about this?” A little more digging showed the trend was intentional and expected.
That's part of what we mean by enablement. People need to understand the tools, but they also need to know how to participate in the work. They still own the output. They need to stand behind it and explain it.
The same applies when we're putting AI into a workflow. A couple of promising examples are a starting point. We want to see it handle a reasonable body of work before we look each other in the eye and give it a thumbs-up. The trust bucket can empty quickly. Building it back takes time.
There's an organizational version of this, too. A team gets the tools, enjoys the training, and then says everything feels chaotic. All that energy needs somewhere to go: a clear vision, a clear strategy, and agreement about where AI should and shouldn't play a role.
That doesn't need to become a long exercise. It can start with an hour in a room: what do we think AI can do, where is it already being used, and who's accountable for the outputs?
And some problems still need plain process improvement. We can eliminate a step or fix a handoff without bolting AI onto it. The problem can be worth solving even when AI isn't the right next move.
Lab Notes
| ■ | Justin's note: When I need to build something like a pro forma, I can ask AI to walk me through the process and explain the choices. I still want to learn what the exercise teaches me about the business. |
| ■ | Kellan's note: My brother's camp agent got three jobs. If he finds himself adding a fourth, stop and finish the three. Meet people where they are, with the time and comfort level they have. |
More ways to listen!
You can now listen to The Fizz on Spotify and Apple Podcasts! Follow the show wherever you listen.
What Stopped Our Scroll
|
