Two senior engineers disagree strongly about an architectural approach, and the debate has stalled for weeks. How should an architect resolve it?

The diagnosis first

Most stalled architectural debates are not about the architecture. They stall because the participants are optimising for different criteria and neither has said so, so each finds the other's reasoning incomprehensible.

The first move is to surface the criteria: what is each person optimising for, and what are they most afraid of? Frequently one fears operational complexity and the other fears being unable to scale — both legitimate, and the debate is unresolvable until that is named.

The resolution sequence

1. Agree the criteria before comparing options. Which dimensions matter, and in what order, given the business context. Half of stalled debates end here, because once the criteria are agreed the answer is obvious.

2. Identify the factual disagreement, if there is one. Sometimes the dispute is empirical — "this will not scale" versus "it will". That is testable. A time-boxed spike or a load test settles it far faster than more argument, and it is worth a week to end a month of debate.

3. Assess reversibility. If one option is easily reversible, choosing it and learning is better than deciding correctly in the abstract. Many debates deserve a decision rather than a resolution, and the cost of being wrong is the relevant number.

4. Decide, explicitly and with an owner. If the criteria are agreed and the facts are settled and disagreement remains, it is a judgement call. Someone decides, the reasoning is recorded, and the organisation moves. An undecided architecture is worse than either option.

5. Ask for disagreement and commitment. The person who did not prevail must be able to say so on the record and then support the decision. Suppressed disagreement resurfaces during the first difficulty and is far more damaging.

What not to do

• Let it run. The cost of the stall usually exceeds the difference between the options — and that arithmetic is worth stating out loud.
• Split the difference, producing a hybrid neither person believes in and nobody owns.
• Escalate immediately. It removes agency and teaches the team that disagreements are resolved by authority rather than by reasoning.
• Decide by seniority without engaging the argument, which loses the information the disagreement contains.

The pattern to notice

If the same two people stall repeatedly, the problem is not the decisions. It is unclear decision rights, or an unstated conflict about direction that keeps surfacing as technical disagreements. That requires a different conversation, and having it is more valuable than resolving any individual dispute.