Engineers who have never tried explaining a bug aloud tend to hold firm opinions about it, mostly negative. The practice, often called rubber-duck debugging, involves describing a failing system, step by step, to a listener who cannot reply. It has a reputation as either a joke or a beginner’s crutch. Six common myths follow, each examined against how engineering work actually goes.
Myth 1: It only works if the listener is a real duck
The reality: The object is irrelevant. A plant, an empty chair, a blank text editor, or simply the wall will do. What matters is the discipline of producing a complete, ordered account. The duck is a prop that gives you permission to speak without self-consciousness. Some engineers prefer to type their narration into a scratch document, which works for the same reason, although speaking tends to be slower and therefore more thorough.
Myth 2: It is for juniors; experts do not need it
Experience points the other way. Experienced engineers carry large mental models, and large models contain large unexamined assumptions. A senior person debugging a deployment may be quicker to form a theory and slower to doubt it. Narration slows the theory down just enough to test it. Juniors benefit because they lack the model; seniors benefit because they have too much of one. Neither group is exempt from the habit of skipping steps that feel obvious.
Myth 3: If the problem is complex, explaining it will take forever
The reality: You rarely need to narrate the whole system. Start at the failing request or job and work backwards or forwards until you hit the first hesitation. A typical session lasts five to ten minutes, and it often ends early because the first doubtful sentence appears quickly. If a long narration finds nothing, that is informative too, since it rules out everything you described. A short spoken pass is among the cheapest diagnostics available, certainly cheaper than adding logging and redeploying.
A practical limit helps keep sessions short: agree with yourself in advance to stop after ten minutes. If nothing snagged, change tactics deliberately, by comparing the failing environment with a working one or by reproducing the issue on a laptop, rather than narrating the same path a second time and hoping it sounds different.
Myth 4: It cannot find bugs you do not already know about
This one is half right, and the half matters. Explaining aloud cannot produce a fact you never learned. It can, however, pinpoint the place where you were relying on a belief rather than a fact. Say you narrate: “The service reads its configuration at start-up, so a restart applies the change.” If you pause there, you have not discovered the bug, but you have discovered the question: does it really only read at start-up, or does a sidecar cache it? The technique turns an unknown unknown into a known unknown, which is the step before every fix.
Myth 5: Writing a clear ticket is the same thing
A well-written ticket is valuable, and the two practices overlap, but they are different activities. A ticket is composed for a reader, so it tends to hide uncertainty behind confident wording. Spoken narration is composed for nobody, so the hesitation survives. A ticket polishes; an unrehearsed account exposes. Many engineers report that their best tickets come after a spoken pass, because they now know which sentence to flag as unverified.
Myth 6: It is unprofessional, or something that sounds odd in an office
Engineering teams in many countries have adopted the practice openly, and nothing about it requires speaking loudly. Murmuring under your breath, going for a short walk, or using a meeting room for ten minutes all work. As for professionalism, the measure is whether the practice reduces wasted effort. Organisations that care about delivery quality, as discussed in the outlook piece titled the future of DevOps at home and abroad, tend to welcome anything that shortens time to a correct diagnosis without adding tooling or process.
Putting the myths to rest with a short example
A team’s nightly batch job occasionally produced duplicate invoices. Colleagues assumed a race in the scheduler and spent a day on locks. One engineer, quietly narrating, said: “The job reads unprocessed rows, sends them, then marks them processed.” Then the doubt: “Marks them processed after sending. If sending takes longer than the next run starts…” The next run overlapped the previous one and re-read the same unprocessed rows. The scheduler was fine. The order of “send” and “mark” was the issue, and the engineer found it in minutes by describing an apparently trivial step.
This kind of finding matters beyond the engineering team. When delivery problems reach customers, as the article on how delivery practices shape customer satisfaction explores, quicker diagnosis is part of how an organisation protects the people it serves.
A note on reading unfamiliar websites
The skill carries over to everyday reading. Someone arriving at a gastroenterology-focused directory site, which lays out hospital and doctor listings next to a blog and a forum, can stop and describe aloud what each part of the site is meant for before relying on any of it. The point is orientation rather than content, and it follows the same logic as narrating a system before changing it.
What to take from the six corrections
- The listener can be anything or nothing; the ordered account is what counts.
- Seniority does not protect you from untested assumptions.
- Start at the failure and stop at the first hesitation; sessions are short.
- The method finds beliefs masquerading as facts, not new facts.
- It differs from writing a ticket because nobody is there to impress.
- It can be done quietly and professionally anywhere.
If you are sceptical, run an experiment on your next stubborn issue: ten minutes of narration before any new tooling. The result will either save you the afternoon or show you, cheaply, that the cause lies elsewhere.









