Architecture reviews exist to catch problems before they become expensive. In practice, most reviews catch a different set of problems from the ones they were designed to find, and miss a different set from the ones that will eventually cause trouble.
This is not because the reviewers lack competence. It is because most architecture reviews are not designed to produce understanding. They are designed to produce alignment and distribute accountability. Once that is the function, the outcome is predictable.
A review that produces endorsement rather than understanding
An architecture review is a room full of people with different stakes in the design.
The architect has a professional investment in it. Managers have already made commitments that depend on it. Engineers have opinions about feasibility that may conflict with timelines. Security has concerns that may require rework that nobody has planned for.
These are not neutral positions. They shape what can be said.
Most reviews ask for candid challenge while making that challenge professionally costly. The result is a process that records attendance, produces action items on safe topics, and quietly converges on endorsement. The substantive concerns are present, but they are not expressed in a way that forces resolution.
What happens instead of honest engagement
Under these conditions, people do what the system allows.
The technical participant produces a precise and increasingly detailed account of the design that never quite addresses the concern that was raised. Questions about delivery risk are answered with explanations of technical correctness. These are not the same thing, but the latter is easier to defend and harder to challenge directly.
The reviewer who raises a concern sees it reflected in the next version of the design. The wording acknowledges the issue. The design does not change in the way intended. In the next review, the concern is marked as addressed. Challenging that status risks being seen as obstructive.
The manager agrees that the concern matters. Nothing changes. Addressing it would require revisiting a prior commitment, and the incentives in the room are not aligned with that outcome.
These are not individual failures of honesty. They are the expected output of a process that rewards agreement and penalises disruption.
What the review misses as a result
The concerns most likely to cause problems are also the ones least likely to survive this process.
They are uncomfortable because they require changing something significant. They are expressed indirectly because direct expression carries cost. They are acknowledged and deferred rather than resolved.
The review closes. The design proceeds. The problem appears later, during implementation, in production, or in an incident. What was deferred becomes expensive.
This is not an exception. It is the normal outcome of a review that functions as a governance checkpoint rather than a genuine examination of whether the design will work.
The function most reviews actually serve
Many architecture reviews exist to distribute accountability rather than to improve the design.
A room full of people has seen the proposal. Questions have been asked. Concerns have been noted. Decisions have been minuted. When problems emerge later, they are no longer owned by a single person.
If this is the real function, the behaviour in the room makes sense. The goal is not to surface the most difficult problems, but to avoid being the person who blocked progress or took sole responsibility.
Any attempt to improve reviews that ignores this underlying function will produce better looking meetings with the same outcomes.
What changes when the conditions change
A review that is intended to find problems requires different conditions.
The most important is separation between review and approval. If a review is expected to endorse a design that has already been agreed, it cannot also function as a mechanism for challenging that design. Where organisations cannot or will not separate these activities, the limits of the review should be made explicit. Otherwise, it becomes a ritual that implies scrutiny without enabling it.
Concerns need to be raised without being treated as obstruction or as a personal challenge. This is not achieved by stating a norm once. It requires consistent demonstration that raising a difficult issue is expected and useful.
Architects need to be able to acknowledge that a concern is valid and that the design needs to change without this being treated as failure. Architectural error is normal. Error caught in review is cheap. Error suppressed because the conditions did not allow it is expensive.
These are not process tweaks. They change what the review is for.
The facilitator’s role
A review with an external facilitator who has no direct stake in the design behaves differently.
A facilitator can ask the questions participants tend to avoid: what concern has not been said, what would need to be true for this design to fail, what is most uncertain here. These are ordinary questions, but they are structurally difficult for participants with stakes in the outcome to ask of each other.
A facilitator can improve the conditions in the room. They cannot override the incentives around it. If the organisation does not want certain answers, those answers will remain difficult to surface regardless of who asks the question.
The system around the room
The review does not exist in isolation.
Timelines, funding decisions, and prior commitments shape what is possible long before the review takes place. By the time a design reaches formal review, some decisions are already effectively locked in. Expecting the review to surface and resolve those constraints is unrealistic.
If the upstream system rewards speed, certainty, and adherence to plan, the review will reflect those priorities. It will not compensate for them.
What this means in practice
Most organisations already have architecture reviews that perform exactly the function they are set up to perform.
They create alignment, distribute accountability, and allow work to proceed. They do not reliably surface the problems that matter most.
The question is not whether the review looks rigorous. It is whether it is designed to make difficult truths visible and actionable.
If it is not, the risks are still there. They are simply deferred, waiting for a more expensive moment to appear.
For a deeper look at how technical, emotional, and political factors shape architecture work, see The human layers