When the agent is cheaper than the meeting about the ticket
Software economics has always been discussed in terms of engineering time, which was the expensive input. Once generation is nearly free, the expensive input turns out to be coordination, and almost nothing in how we work is organised around that.
Here is an arithmetic exercise that changes how a planning meeting feels once you have done it.
Six people spend forty minutes deciding whether to build a small feature. Loaded cost of an engineering hour, call it a hundred and twenty currency units, so the meeting costs somewhere around five hundred. Refining the ticket afterwards takes a product manager another half hour. Total, call it six hundred, before anything is built.
The feature itself, once specified, is perhaps ninety minutes of agent work and human review. In token terms it costs a few units. In human terms, maybe eighty.
The deliberation cost seven times the delivery.
This is not a new phenomenon; coordination has always been expensive. What is new is the ratio, and the ratio has crossed a threshold where it changes what the correct decision procedure is.
Why the ratio changed
For most of the industry’s history, building was the expensive part, and every process we have was designed on that assumption.
Requirements documents existed because building the wrong thing wasted months. Estimation existed because commitments were expensive to revise. Prioritisation was rigorous because capacity was severely constrained and each choice foreclosed others. Design review existed because rework was brutal.
All of these are rational responses to expensive construction, and all of them are themselves expensive. They were worth it because they were cheap relative to the thing they protected against.
When building was expensive
- Deliberate carefully before committing
- Estimate, because revision is costly
- Prioritise hard, capacity is scarce
- Review the design, rework is brutal
- Deciding is cheap relative to doing
When building is cheap
- Deliberation is the dominant cost
- Estimation costs more than the variance it removes
- Prioritisation still matters, for a different reason
- Building a version can be cheaper than discussing one
- Deciding is expensive relative to doing
The move this licenses
The practical consequence: for a certain class of question, building the thing is now a cheaper way to answer it than discussing it.
The class is specific. Questions about how something would work, or feel, or perform, or fit into the existing system. Questions where the disagreement in the room is factual rather than about values, and where the fact could be established by trying. Questions where the meeting is going in circles because the participants have different mental models and no way to reconcile them.
For those, the correct response to “we should discuss this properly next week” is often to have a working version before the meeting and cancel it.
I want to be careful about how far this goes, because there is an obvious bad version. Not every question is answerable by building. “Should we be in this market” is not resolved by a prototype. “Does this violate our data retention commitments” is not a build-and-see question. The class this applies to is narrower than enthusiasts claim and much broader than current practice assumes.
Build instead of discussing
How would this feel. Does this integrate cleanly. Is this fast enough. Which of two designs reads better in practice.
Discuss, do not build
Should we serve this market. Is this consistent with our commitments. What are we choosing not to do. Who owns this afterwards.
Build to inform the discussion
Where the disagreement is factual but the decision is strategic. The prototype settles the fact, humans still make the call.
Neither, decline it
The largest and least used category. Cheap to build is not a reason to build.
What estimation is for now
Estimation deserves a specific mention because it is the ritual with the worst remaining cost-to-value ratio.
Estimation was always partly theatre, and the honest defenders of it usually argued for the conversation rather than the number: the value was in surfacing hidden complexity and disagreement about scope, not in the points.
That defence is still valid, and it is also now an argument for doing much less of it. If the conversation is the value, have the conversation and skip the number. And if the work is small enough that building it takes less time than a planning poker round, the estimate is pure overhead.
Where estimation retains value is at a coarser grain: is this a week or a quarter, does this need a team or a person, will this require a change to how something fundamental works. Those distinctions still matter because they inform genuinely expensive commitments. Estimating individual tickets in a world where a ticket is ninety minutes of work is measuring with an instrument more expensive than the thing being measured.
The prioritisation point that survives
One thing does not go away, and it gets misunderstood in the general enthusiasm.
Prioritisation used to be about capacity: we can do six things, here are twenty, choose. That version weakens.
But the reason to prioritise was never only capacity. It was that every feature built is a feature maintained, operated, secured, documented, and eventually deprecated. The construction cost fell. The ownership cost did not move at all.
So prioritisation shifts from a capacity question to a portfolio question. Not “can we afford to build this” but “do we want to own this for five years.” That is a harder question and a more important one, and organisations that hear “building is cheap” and relax their prioritisation will discover the answer expensively in about three years.
Where this breaks down
Cheap to build is a powerful incentive to build unnecessary things. This is the dominant risk and it is not hypothetical. A backlog full of small, plausible, individually cheap features that collectively make the product incoherent is a real and common outcome, and the low construction cost is precisely what causes it.
Prototypes have a way of shipping. Build-to-decide only works if the artefact can be thrown away, and organisational gravity is strongly against that. A working demo that a stakeholder has seen is very hard to delete. Anything built to answer a question needs to be visibly disposable, ideally in a way that makes shipping it inconvenient.
Some meetings are not about the decision. A number of recurring meetings exist to align people, distribute context, or let someone feel consulted. Cancelling them because the technical question could be answered by building will produce a political problem that costs more than the meeting did.
The arithmetic assumes specification is free, and it is not. The ninety-minute figure at the top of this piece assumes someone already wrote a clear specification. Writing it well is the expensive part and it is human work. The cost did not vanish; it moved upstream, which is the theme of half this series.
Coordination cost is not purely waste. Some deliberation is how an organisation builds shared understanding, and a team that builds instead of discussing will move fast and gradually lose the ability to agree on anything. There is a floor here and I do not know where it is.
The takeaway
The cost ratio between deciding and doing has inverted for small work, and none of our processes have been recalibrated. A forty-minute meeting to decide on a ninety-minute change is now the expensive part of that change by a wide margin.
For factual questions, build rather than discuss. Estimate at a coarse grain or not at all. And hold prioritisation firmly, on the grounds that ownership cost never fell even though construction cost did.
If you take one thing into next week: price your next planning meeting, honestly, against the work it is planning. If the meeting costs more, that is not an argument against meetings. It is an argument for knowing which of the two you are actually deciding about.
The next piece closes the economics series with the question every one of these arguments eventually has to face: how to measure the return without fooling yourself.