The Herald With the Sealed Scroll
In the old days, when the king asked his sage a question, the herald would carry the question to the sage's hut, wait through the long composition, and return with a sealed scroll only when the answer was complete. The king would sit on his throne, drumming his fingers, sometimes for hours, never knowing whether the sage had begun, was mid-sentence, or had fallen asleep. The court hated the silence almost more than they hated the wait.
“A long silence is not the same as no answer — but to the listener it feels exactly the same.”
The River That Spoke Aloud
A wandering yogi suggested a different way. "Let the herald not wait for the scroll. Let him stand at the sage's door, and as each syllable leaves the sage's lips, let the herald repeat it aloud, one syllable at a time, all the way back to the throne." The king tried it. The first syllable arrived in a heartbeat. Then another. Then another. The composition was no faster than before — but the king felt as though it were, because he was hearing it form. The court ceased its drumming.
“The first syllable matters more than the last. Show that you have begun, and the wait shortens itself.”
The Aqueduct That Hoarded
But between the sage's hut and the palace ran a clever new aqueduct, lately built by a proud engineer. The aqueduct, wishing to be efficient, would gather water in small pools and release them in great gushes. So the king would hear nothing for ten heartbeats, then a flood of syllables, then nothing again. The illusion of conversation was lost. The yogi inspected the aqueduct, found the hoarding pools, and ordered them sealed. "Let the river flow as it flows. The aqueduct's job is to carry, not to gather."
“A pipeline that buffers a river breaks the river. Disable the proxy buffer.”
The Listener Who Raised a Hand
One morning the sage began an answer about the southern campaign, and three syllables in, the king realised the question itself had been wrong. In the old days he would have had no choice but to wait for the sealed scroll. Now he raised a hand. The herald, seeing the gesture, ran back and tapped the sage on the shoulder. The sage stopped mid-syllable. No more breath was wasted; no more ink was spent. The court learned that listening could be cancelled — and that this freedom was almost as precious as the speed itself.
“The right to interrupt is the other half of the right to listen.”
When the River Broke Its Banks
Once, midway through a great composition, lightning struck the aqueduct and the herald lost his footing. The sage continued to speak; no one was carrying the syllables. The king was furious. The yogi, the next day, taught the herald a new trick: at every twentieth syllable, the herald would mark his place. If the line broke, the next herald could pick up from the last marked syllable rather than asking the sage to begin again. The sage was old and tired easily. The mark saved him much breath.
“A long stream needs anchor points, so a broken connection is not a lost composition.”
The River That Fed the Fields
In time the practice spread. Heralds learned to whisper the syllables to scribes who could begin formatting the scroll while the sage was still speaking. Cooks in the royal kitchen learned to begin preparing the feast that the sage was, syllable by syllable, ordering. The river of speech no longer ended at the king's ear; it irrigated the whole court as it flowed. And the old sealed-scroll method, when occasionally tried again, felt to everyone like trying to drink from a stone.
🪔 Deepak — the lamp of meaning · what this fable means in code
The herald with the sealed scroll is the old request-response API: silence until completion. The river of syllables is token streaming — each token sent the moment it is generated, collapsing perceived latency even when total latency is unchanged. The hoarding aqueduct is a buffering proxy (Nginx without `X-Accel-Buffering: no`) that defeats streaming silently; the fix is to disable proxy buffering end-to-end. The listener's raised hand is AbortSignal — cancellation propagated upstream, saving tokens and dollars when the user no longer wants the answer. The marked twentieth syllable is the resumable stream pattern (Server-Sent Events with `Last-Event-ID` or chunked checkpoints) that lets a broken connection recover without restarting. And the irrigated court is the deeper architectural shift: streams enable downstream consumers — UIs, code-formatters, tool-calling middleware — to begin work before the answer is finished.