Don't just add AI to the old way of working.
I sat in four developer standups a week. Forty-five to sixty minutes each. Somewhere around eighty percent of it was implementation detail that did not need a CTO in the room.
So I did what everyone does. I went looking for an AI that could summarize meetings.
That was the wrong instinct, and it took me a while to see why. A summary would have made the thing I did not need shorter. It would not have asked whether I needed to be there at all.
The better question turned out to be this: how should this work get done now that AI exists?
Those two questions sound almost the same. They are not. The first one keeps your process and bolts a tool onto it. The second one puts the process on trial.
What I found when I actually asked it
The standup was doing more than one job at once.
It was moving information, which is the part I did not need. It was also coaching, catching blockers that needed a decision from me, running demos, and giving an India-based team a reliable window to reach me. Cancel the meeting and you lose all of that to save three hours.
So the answer was not to delete it and not to summarize it. The daily development update my team already writes contains nearly everything I need. It did not need a new report. It needed something reading that update against what I actually care about, and telling me only what required me. Decisions waiting on me. Architecture and security. Initiatives that moved. Some days it tells me nothing needs my attention, which is the most valuable thing it says all week.
I kept the meeting. It is one a week now, and it is the part that was always worth showing up for.
I was not automating the standup. I was rethinking why I was in it.
The framework
That question kept working on other problems, so I wrote down how it actually goes. I am calling it RE+SHAPE.
RE finds the work worth changing.
- Reveal. Where does friction live? Recurring meetings, copy-and-paste, spreadsheets standing in for a process, work you do because somebody once asked for it.
- Examine. Why does this exist, who benefits, and what breaks if it stops? A completely valid answer here is: leave it alone.
SHAPE decides how it should change.
- Specify. Say what the outcome is without describing how it currently gets done.
- Hunt. Question every step before you add anything. Delete, reduce, preserve, elevate.
- Augment. Put intelligence between the information and the person, so attention is an escalation path and not a processing layer.
- Put AI to Work. Give it a job, not a prompt. Mission, judgment, sources, output, boundaries, what it does when it is unsure.
- Execute. Hand over initiative, with explicit limits. This is the last step, and most work never needs it.
The order matters more than any individual stage. Automate last.
Where this shows up in a normal week
You already have candidates. Here are the ones I hear most.
"My meetings are too long." Usually a meeting is bundling information transfer with something that genuinely needs humans. Split them and the long part often turns out to be the part nobody needed to attend.
"I get too much email." Filters and folders move the pile around. The question is which messages should reach you at all, and what should happen to the rest without you touching them.
"A client emailed and I missed it for two days." This one is rarely an inbox problem. It is that nothing is watching for the handful of messages where response time is worth actual money.
"Campaign research takes too long." Reading everything is the mechanism, not the goal. The goal is a defensible point of view by Thursday. Those are very different problems.
"Our weekly report takes someone a day." Ask what anyone does differently after receiving it. If nobody can answer, you have found something to delete rather than something to automate.
"Reconciliation eats two days a month." Thousands of lines match. Thirty do not. Only the thirty need a human, and only a human should resolve them.
"I spend my life scheduling calls." Once you have decided a meeting is worth having, nothing after that decision requires you.
"Onboarding a new hire takes three weeks of my time." Some of that is relationship and judgment. Some is reciting things that could be written once.
"We send the same proposal with eight fields changed." The judgment is in pricing and scope. The assembly is not judgment.
Notice what is missing from that list. None of them start with a tool.
Five answers, and none of them is the winner
Run a piece of work through this and you land in one of five places.
Leave it alone. Eliminate or redesign it. Augment it so a person and a machine do it together. Delegate it. Let it run on its own.
Nothing on that list outranks the rest. I have taken work through this process and concluded the right move was to change nothing at all, and that was a good outcome. Autonomy is earned, and most work never earns it. If every answer you reach is "automate," you are not running the process. You are decorating a decision you already made.
The goal is not more AI. The goal is better work.
Try it on something real
I built RE+SHAPE into a conversation on the site. It went live yesterday.
Bring it a piece of your work →
It asks questions. It will push back if you arrive with a solution instead of a problem, which you probably will, because I did. Bring something slow, repetitive, expensive, or just more annoying than it should be. You will leave with a written plan for that specific piece of work.
It is new and it is in limited beta. If it says something useful, I would like to know. If it says something dumb, I would much rather know that.
You can also read the framework and all thirty plays without talking to anything.
The bottom line
Most AI advice starts with the technology and works backward to a use case. That is why so much of it produces pilots nobody misses when they are cancelled.
Start with the work instead. Ask what it is for, what would actually break without it, and what it should look like now that these tools exist. Sometimes the answer involves AI. Sometimes the answer is that the work should stop. Both of those are wins, and only one of them shows up in a vendor demo.
I am going to be working through a lot more of this in the coming weeks.
Hit reply and tell me what you ran it on. The best conversations in this newsletter's history started as replies.
— Keith