Every diagnostic produces the same artifact: a shortlist of agents the operation could ship, ranked. The ranking function is where portfolios are won or lost, and the wrong function is the popular one: excitement.
Excitement ranks the demo that will impress the board, the use case the CEO mentioned, the technology the team wants to try. Payback ranks the agent that returns its cost fastest in a number the finance function already tracks. The two lists rarely agree, and only one of them buys you the right to ship the next agent.
Why the first agent carries the portfolio
The first agent is not really a project. It is the proof that changes the budget conversation. When it pays for itself inside two quarters, the second agent is not a pitch; it is a line item with a reference. When it stalls, the entire programme inherits the stall. This asymmetry is why we refuse to build more than one agent at a time, and why the one we pick is the boring one with the eleven-week payback rather than the thrilling one with the eighteen-month story.
The fastest way to ten agents is one that works.
Sequencing rules we hold
Rank by payback period, using the client's own numbers, verified against history. Prefer queues with volume, because volume feeds evaluation and shortens the proof cycle. Prefer functions with a willing owner over functions with a better business case and a reluctant one. And bank the win before scaling: the playbook from agent one, from data access to security review, is most of the cost of agents two through ten.
A portfolio your board can underwrite is not a vision document. It is a sequence of paybacks, each one funding the next. Get the order right and the programme finances itself.