When Options Are Free
Some organisations have never learned to choose. They have only learned to run out of capacity.
The question at the end of the meeting
There is a familiar moment in a portfolio committee.
A proposal has been presented. The sponsor is credible. The numbers are respectable. It aligns to at least one strategic priority and probably three. Nobody can point to a fatal flaw.
Then somebody asks what we are going to stop in order to do it.
That is usually when the conversation becomes real.
Until that point, the committee has been discussing the quality of an initiative. Now it has to confront the existence of a portfolio.
For most of my career, scarcity made that confrontation unavoidable. Capital was limited. Delivery capacity was limited. Senior attention was limited. But another form of scarcity mattered too, and I do not think we gave it enough credit: the ability to examine an option properly was expensive.
Turning a hunch into something a serious committee could consider took weeks of human effort. Someone had to research it, test the market logic, find the dependencies, estimate the benefits, build the business case, negotiate the assumptions and persuade a sponsor to spend political capital on it.
That effort was not just administrative friction.
It was a hidden gate.
Before an initiative reached the formal portfolio process, somebody had already answered a prior question: is this worth enough to work up?
We rarely called that portfolio management. But it was doing portfolio work.
The gate we forgot was there
Think about what that hidden gate accomplished.
It killed weak ideas before they became propositions. It stopped marginal opportunities demanding executive attention. It forced sponsors to ration their own effort. Most importantly, it made the organisation reveal something about its priorities before any scoring model appeared.
The formal funnel might show eight propositions competing for three places. But perhaps one hundred ideas existed upstream. The first ninety-two were filtered by cost, attention, sponsorship and the simple fact that nobody could afford to elaborate everything.
That is easy to dismiss as inefficiency.
AI makes the dismissal tempting.
Stanford’s 2025 AI Index reported that the inference cost of a system performing at roughly GPT-3.5 level fell more than 280-fold between November 2022 and October 2024. That does not tell us directly what a business case will cost to produce. It does tell us that the model layer capable of supporting analysis, drafting and comparison is becoming dramatically cheaper.
A half-formed idea can increasingly become a market scan, customer proposition, benefits case, risk analysis and implementation narrative before anyone has had to invest very much in it.
The obvious response is positive: good. We can explore more.
I agree.
The mistake is assuming that nothing else changes when the hidden gate disappears.
The old system bundled three things together: exploration, persistence and commitment. Expensive exploration rationed all three. AI begins to separate them.
That separation is the real portfolio problem.
When “why not?” becomes normal
Today an advocate usually carries the first burden of proof.
Why is this worth examining?
Why should scarce people spend time on it?
Why should it enter the funnel?
As elaboration gets cheaper, that burden weakens. If a credible first case can be produced quickly, “we have not looked at it” begins to sound less like focus and more like neglect.
The question changes.
Can we explore it?
Probably.
Can we prototype it?
Increasingly.
Can we run a limited test without asking for a large team?
Often.
So why not?
That phrase matters more than it appears to.
The old question demanded a reason to proceed. The new one demands a reason to refuse.
Yes becomes easier to justify because yes can be made small. No remains absolute.
A sponsor no longer needs to win approval for the full proposition. They only need approval for another experiment, another iteration, another three months, another small exception. Each step is individually reasonable.
Portfolio functions are not designed for that burden reversal. Most have developed elaborate ways of asking an initiative to justify its existence. Far fewer have developed equally legitimate ways of saying: this is plausible, affordable and still not something we want.
That requires direction in a harder form than a strategy deck usually provides.
The old machinery has an afterlife
There is an adoption problem here as well as a capability problem.
AI can make elaboration technically cheap long before an organisation redesigns its governance around that fact.
So the first phase may look strangely familiar. The same business cases. The same gates. The same scoring models. The same quarterly forum.
Just much more material flowing through them.
This is where I expect some of the worst behaviour.
A scarcity-era process will be asked to govern an abundance-era input stream. The rational response will be to automate more of the process: AI-assisted business cases, AI challenge, automated scoring, richer scenario analysis, continuous portfolio refresh.
Some of that will be useful.
Some of it will also create an odd circularity.
Take an ordinary scoring model: strategic alignment, financial value, customer impact, deliverability, risk. Now give those criteria to the machine helping the sponsor construct the proposition. Ask it to improve the case against the rubric. Ask another agent to challenge the result. Repeat.
The framework has stopped being only a filter.
It has become part of the prompt.
Every initiative learns to speak strategy. Every weak proposition becomes a pilot. Every dependency becomes manageable. Every benefit acquires a confidence range and a narrative.
This does not require anyone to cheat. The proposal may genuinely improve.
That is precisely why the problem is difficult.
In April 2026, Nature published a warning about a structurally similar risk in research funding. AI agents can generate, review and submit grant applications, while the authors warned that rising volumes of high-quality AI-assisted proposals could make it much harder for funding systems to discriminate among them.
Enterprise portfolios should recognise the shape of that problem.
The funnel does not vanish. It develops a strange afterlife: more sophisticated, more automated, and less able to rely on the friction that once narrowed what arrived.
The strongest objection changes the argument
There is an obvious objection, and it is a strong one.
If exploration becomes cheap, why manufacture scarcity at all?
Why not generate hundreds of possibilities, cluster them with AI, test them cheaply, and let evidence eliminate the weak ones? More exploration could expose neglected opportunities, reduce executive bias and make strategy less dependent on whichever sponsor happened to have the resources to develop a case.
I would not want to rebuild the old friction simply because we are nostalgic for a smaller funnel.
The objection is right about exploration.
It changes where I put the boundary.
The future portfolio should not make possibilities artificially scarce. It should make commitment scarce.
“The future portfolio should make exploration abundant and commitment scarce.”
That distinction sounds tidy until an organisation tries to live it.
Exploration needs somewhere to end.
A test has to expire.
A pilot that has not crossed its evidence threshold has to stop rather than becoming a cheap permanent resident.
A reversible experiment has to face a different standard when it wants access to shared data, customers, architecture, brand, regulatory permission or enduring operating support.
In other words, the portfolio has to become much better at distinguishing three states that the old world often allowed scarcity to blur together:
- worth exploring
- allowed to persist
- chosen as a commitment
AI can make the first state enormous.
That is probably desirable.
The second and third cannot be allowed to expand automatically with it.
Cheap things are difficult to kill
This is where the argument becomes less comfortable.
Consider a modest internal AI service. It once would have required a team large enough to attract scrutiny. Now it needs two people, some model access and a tolerable cloud bill.
The evidence after three months is mixed.
Do we kill it?
The sponsor has an easy answer: why? It barely costs anything.
Give it another quarter.
Then another initiative receives the same treatment. And another.
Nothing is individually large enough to force a decision.
The portfolio begins to fill with inexpensive undead work: pilots, agents, automations, services and almost-products that never became important enough to commit to and never became expensive enough to stop.
The direct cost remains small. The accumulated cost does not.
Each thing introduces some combination of data dependencies, security obligations, architecture choices, customer surfaces, controls, integrations, exceptions and management attention. Each adds another fact the organisation has to remember about itself.
The scarce resource has moved.
It is coherence.
That is the second-order consequence of cheap execution. Lower cost does not only allow more experiments. It weakens one of the signals that used to tell us when an experiment had overstayed its welcome.
Cheap experiments leave expensive residue.
What strategy has to do now
This is where the argument needs a qualification.
It is too crude to say that AI makes options abundant and therefore strategy must manufacture scarcity.
Scarcity of what?
Certainly not curiosity. Not analysis. Not the ability to test a reversible idea.
The scarcity that matters is permission to persist and permission to commit.
That is a much tougher form of strategy than deciding which proposition scores 78 and which scores 71.
It means saying:
We are willing to explore this, but not integrate it.
We are willing to test this, but it expires on this date unless a stated condition is met.
We are willing to learn about this market, but not enter it.
We are willing to run ten experiments in this space, but only two may become enduring products.
We will not create a third customer-facing AI layer, however attractive the case, unless leadership explicitly changes the architectural direction.
Those are not prioritisation scores.
They are constraints.
And constraints expose something that scarcity previously allowed us to hide.
Some organisations have never learned to choose. They have only learned to run out of capacity.
A company with limited delivery capacity looks selective because it cannot do everything. A company with limited analytical capacity looks focused because it cannot examine everything. A company with limited capital eventually discovers a funding line.
When those limits loosen, apparent discipline can disappear remarkably quickly.
That is the wager underneath this argument: capacity constraints have been doing more strategic work than many organisations realise.
The portfolio function moves to the boundary
If that is right, the portfolio function does not disappear.
But its centre of gravity moves.
Machines should become increasingly good at the middle of the problem: generating possibilities, comparing them, exposing duplicates, modelling dependencies, challenging assumptions, finding combinations and identifying where the evidence is weak.
The interesting human and institutional work moves to the boundaries around that machinery.
Who may explore what?
What must be true for an experiment to persist?
What changes when a reversible test touches an irreversible asset?
What evidence ends the work automatically?
Who can grant an exception?
When does a collection of individually sensible initiatives become strategically incoherent?
Executive leadership still owns direction. It has to decide what kind of organisation it is trying to build and what kinds of opportunity it is prepared to leave behind.
The portfolio function becomes the place where those choices acquire operational force.
Its most revealing artefacts may therefore look less like rankings and more like boundaries: explicit exclusions, appetite ceilings, expiry rules, kill conditions and an exception ledger showing when leadership chose to breach its own direction.
That last one interests me particularly.
A scoring model tells us what the organisation says it values.
An exception ledger tells us what it values when a persuasive opportunity arrives.
I would read both.
What would make this wrong?
The argument should expose itself to failure.
I am staking four claims.
- As the cost of elaboration falls, substantially more plausible propositions will reach the point where an organisation feels obliged to consider them.
- Organisations will initially absorb that abundance through existing portfolio machinery, producing more automation and more formal analysis before they redesign the underlying decision rights.
- The more important portfolio distinction will shift from which options should we explore? towards which explored options are allowed to persist and become commitments?
- Organisations with explicit expiry rules, strategic exclusions and commitment thresholds will maintain greater portfolio coherence than organisations relying mainly on comparative scoring, even if the latter use more sophisticated AI.
The argument weakens if cheap exploration produces more variety without more persistence; if inexpensive initiatives terminate naturally without stronger kill mechanisms; if scoring systems continue to discriminate cleanly when propositions are constructed against them; or if organisations can absorb a much larger number of enduring initiatives without losing coherence.
I would be pleased by that result.
It would mean AI gave us more optionality without weakening the mechanism of choice.
But if the sequence runs the other way, the next portfolio committee may face a problem that sounds almost absurd by today’s standards.
It may have the analytical capacity to examine everything.
It may have the technical capacity to try almost anything.
It may even have enough delivery capacity to keep far more alive than before.
Then somebody will still have to ask what should stop.
Only this time, “we do not have the capacity” will no longer be much of an answer.
That is when we find out whether the organisation had a strategy, or merely a shortage.
— The Fringe makes claims. The Lab tests them. Verdict to follow.
Sources
Stanford Institute for Human-Centered Artificial Intelligence — The 2025 AI Index Report — 7 April 2025 — https://hai.stanford.edu/ai-index/2025-ai-index-report
Nature — Could agentic AI topple grant-funding systems? — 27 April 2026 — https://www.nature.com/articles/d41586-026-01297-y