How I think about AI and search and the way teams work
Four longer pieces. Each one sets out how I approach a problem and why and what I count. They are the methods behind the research and the points of view on this site and they are here so that anyone can use them or argue with them.
Discovery in search and AI answers
People now ask AI the questions they once typed into a search box. The answer names a few brands and leaves the rest out. This piece sets out the three questions a brand should ask and the user's journey the research follows and why the fix is a plan rather than an edit.
Discovery in search and AI answers →Where AI should do the work
Most teams have been handed AI tools and no plan for using them. This piece sets out the time audit I start with and how to choose the three tasks that pay back first and why the only measure that matters is the count at day 30.
Where AI should do the work →How I build agents
An agent is an AI model set up to do one job from start to finish. This piece sets out how I agree the job and write the instructions and run it beside the person who does the job today and what I count when it is done.
How I build agents →Define a measure once
Every AI a business uses answers only as well as the definitions it reads. This piece sets out why the layer of meaning over a company's data is now a feature of the platforms it already owns and what the connectors still cannot carry and why the work is writing the definitions down.
Define a measure once →What runs through all four
Be on the front foot with AI. A business should have a plan for AI and know what to do next rather than wonder. Wondering is the expensive option because the work does not wait.
Close the gap one project at a time. The gap between what AI could do for a business and what the business does with it only closes through real work done one piece at a time and started small. A strategy document on its own has never closed it.
The way we work has changed. Resource is expensive and AI can now do most of the work with one person leading it. That is how I do my own research and it is what most businesses have not caught up with yet.
Avoid waste. Count what a team still uses a month later and stop what it dropped. Every figure should show how it was counted and say plainly what could not be checked.
AI and platform agnostic. Nothing should depend on one supplier. The calculations belong outside the AI wherever that is possible and where a model's judgement is used it should be said.
Questions people ask me
Can I use the Barley Labs methods in my own team?
Yes. That is why they are written down. The time audit and the day 30 count and the journey questions need no special tools. If you try one I would like to hear what happened.
Where should a team with no plan for AI start?
With a time audit. It takes a week and costs the team little and shows exactly where the repeated work is. The plan writes itself from there and the piece on where AI should do the work shows one audit drawn three ways.