Phone, web voice and web chat are variations on one problem: a synchronous conversation with different clock speeds and different output affordances. An email thread is a different problem wearing the same clothes.
| Phone | Web chat | ||
|---|---|---|---|
| Gap between turns | Under a second | Seconds | Hours to weeks |
| Context freshness | Guaranteed | Usually fine | Frequently stale |
| Identity evidence | Caller ID - weak | Session - weak | Address - moderate, and stable |
| Correction possible | Immediately, by speaking | Immediately | Only by sending another message |
| Turn length | One thought | One thought | Several questions at once |
Why email is the hard channel
Three things break at once. Context that was correct when the thread paused may be wrong when it resumes - the appointment happened, the price changed, the order shipped. A single email often contains several questions that a synchronous agent would have handled as separate turns. And there is no interruption: a wrong answer is fully delivered before anyone can react.
That last property changes the risk profile enough that we treat email as a channel requiring review rather than autonomy, and we say so rather than counting it as shipped.
Shared context across synchronous channels is a prototype, not a shipped capability. In the production system built on this research - bitpull.ai - email currently appears as conversation summaries and notifications, not as an agent-operated channel. Anything else would be describing a roadmap as a product.
What does transfer cleanly
Knowledge, tools, business logic and guardrails share across all of them without difficulty. It is memory - and specifically the identity evidence that gates memory - where the channels genuinely differ.