In most of my conversations right now, the question has moved past whether AI helps. What leaders are asking me is how to identify the workflows where the gain compounds across the team and the organization at large.
Most people have now had the experience of AI saving them twenty minutes on their own work. The harder question is which work, done at volume by a whole team, is worth building a system around.
The key here is how we select. Which tasks and workflows do we point AI at so the leverage shows up at the individual level, the team level, and across the organization?
I am sharing the framework I use, based on my own experience and on what I have seen published as best practices across industries.
The three-part test
The third one carries the most weight, because it is the one that lets the work leave your hands.
Start with the definition. If you can describe what a finished, correct output looks like, you can check whether you got one.
Code either runs or it does not. A support ticket is either resolved or the customer writes back again. Anyone who sat in the meeting can verify a meeting summary.
The definition becomes the standard the agent works to. Verifying it once proves the standard holds, and that is the moment the work becomes something you can automate.
You are answering one question about the work itself: can an output be recognized as correct? Once the answer is yes, the workflow can be built, and you spot-check it the way you would review anyone's work.
Define it, verify it, and an agent can run it. Work you cannot describe well enough to grade stays work you have to keep doing yourself.
One caution before you commit to a candidate: verifying has to be faster than doing. If checking the output takes as long as producing it yourself, the workflow buys you nothing.
So how do you build the fluency to spot this kind of work and set it up?
By doing it. And you do not have to do it alone.
I hear the same thing from leaders constantly. They want the benefits and they cannot find the time to learn, so the videos get saved and the prompts get bookmarked to try later, and the time never quite opens up.
That is why I built a course that runs on Saturdays. Ninety minutes a week for four weeks, and you build one working system end to end: an AI chief of staff running on your own calendar and inbox. Once you have built one, you can do it again for anything else on your list.
The next cohort starts Saturday, August 29. Attend all four sessions, do the setup between them, and if your brief is not running by September 19, I keep working with you until it is.
The research behind it
In April 2026, Andreessen Horowitz published an analysis of where enterprise AI has actually converted from pilot to paid deployment. Key findings below.
Three use cases dominate: coding, support, and search. Coding leads the other two by close to an order of magnitude.
Why coding leads, in their words: it is data dense, text-based, "precise and unambiguous, with strict syntax and predictable outcomes," and "crucially, it is verifiable: anyone can run it and know if it works, creating tight feedback loops for models to learn from and improve."
What the fastest-adopting industries share: they "are text-based, involve rote and repetitive work, have natural human-in-the-loop involvement to inject human judgment, limited regulation, and have clearly verifiable end outputs (e.g., code that runs, a resolved support ticket)."
What has not adopted: industries that "deal with the physical world, rely heavily on interpersonal relationships, have clear coordination costs across many stakeholders, impose regulatory or compliance hurdles, or lack verifiable results."
Source: Kimberly Tan, Where Enterprises are Actually Adopting AI, Andreessen Horowitz, April 8, 2026.
Examples of where to start
The three below show up in almost every organization, and each one clears all three questions above. They are also the ones I see set up as agents most often, running on their own without anyone triggering them. That is where the leverage we are after comes from.
1. Meeting synthesis and action items
Every meeting produces the same thing: what was decided, what is still open, who owes what by when.
It clears all three. High volume, all text, and the standard is easy to say out loud: every decision captured, every action item attached to a name and a date. You were in the room, so you can check it in seconds.
Set it up once, and every meeting record in the company comes out the same, whoever ran the meeting. That consistency is worth as much as the time saved.
2. Meeting and call prep
A short brief before every external conversation: who they are, what we discussed last time, what is still open, what to ask.
It clears all three. High volume on any client-facing team, all text, and the room tells you within thirty minutes whether the brief was right.
3. Recurring status reports
The weekly update, the project readout, the pipeline summary, the monthly board pack.
It clears all three, and the third question is already answered for you. The format has been fixed for years, so somebody defined good a long time ago and an agent can work to that standard from the first run.
This is the highest volume of the three, and usually the one people are most relieved to hand over.
All three can be set up as agents and put on a schedule. Build them once, and they arrive without anyone asking.
Where else this shows up
If you look across a company for functions where all three show up together, sales, business development, legal, and account management are the clearest.
On the revenue side, the volume sits in proposals, RFP responses, pricing sheets, quarterly business reviews, and renewal paperwork. All text, all repeated constantly, and the definition of good is already written down in the template and the last version that went out.
Legal has the same shape. NDAs, vendor agreements, and standard clause reviews run off templates and precedent, so the standard is documented and a reviewer can check a redline quickly. It is also one of the industries a16z found moving fastest, which fits.
Want help finding yours?
If you want to identify the workflows and high-value tasks your organization could hand to AI, book a 30-minute intro call. We will look at what your team repeats every week, score it against the three questions, and pick the one worth building first.
💡The rule
Point AI at the work you produce at volume, in words, where you can say what good looks like.
Define it, verify it, and an agent can run it.
Found this useful? Forward it to one person who needs it. Getting this for the first time? Subscribe free here.
Think Better. Adopt AI Responsibly.
See you next week - Purti
Clarity Prompts - The weekly AI operating system for sharper leaders.
