Deprecation Notice 01

Directory Assistance

Before the recognizer existed

Every listed number in a phone company’s territory, held in a switchboard operator’s head or a card index at her elbow. You dialled 411, said a name half-remembered, and a person negotiated the gap between what you said and what was listed.

That negotiation is the whole job. It’s also exactly the part a lookup table can’t do.

“The Bell System operator … could locate almost any listing from the vaguest of clues.” — from a 1970s internal Bell training manual

The pitch, as it was actually made

  • speech recognition + a structured database beats a human on cost, latency, and consistency, full stop
  • the marginal call costs a fraction of a cent instead of an operator’s wage
  • zero hold time, zero shift roster, zero attrition

Read that again. Every sentence is true. That’s what makes this week’s case useful — the pitch doesn’t need exaggerating to sound compelling.

What the pitch left out

  • ~100,000 operator jobs in North America at the peak, collapsed to near zero
  • almost none of that headcount moved sideways into telecom — it stopped existing
  • the workforce was disproportionately women, for whom this was one of the few stable office-adjacent careers open at the time

Rollback plan: none

Not “difficult.” Not “expensive.” The training pipeline, the workforce, and the exchange infrastructure that supported it were fully decommissioned. There is no paused system to restart — the occupation would have to be rebuilt from zero.

This is what an honest rollback plan looks like when the honest answer is no.

The four parts, every week

  1. Pitch — the sales case, in its own voice
  2. Migration path — how it’s actually implemented
  3. Breaking changes — economic and functional, always plain
  4. Rollback plan — what reversing it would really take

Same shape, different industry, all twelve weeks.

pnpm dev
pnpm check

Full notice text is on the lecture page — this deck is the walkthrough, the page is the record.