Amplifier as Communication Infrastructure

Signal Over Noise

Encoding how a person communicates

The Idea

Amplifier can capture HOW you communicate

The my-voice bundle turns your real sessions into a Voice Profile — phrases, patterns, tone — and a Decoder Ring, so your signal survives contact with other people.

And it isn't a concept — it already ships.

Voice Profile

Captures how you communicate — phrases, patterns, tone.

Message Cleanup

Condenses verbose messages while preserving your voice (agent: message-tuner).

Decoder Ring

Helps others' Amplifiers understand your communication style.

Proof — It Ships

my-voice is a shipping bundle, not a slide

21 commits from Jan 27 to Apr 20, 2026: Voice Profile + Decoder Ring templates (151 lines each), two agents, a profiles tool, and a sync hook. Version 0.1.0.

So how does captured voice actually reach someone?

21

commits (Jan 27 → Apr 20, 2026)

302

template lines (151 + 151)

2

agents: voice-analyst, message-tuner

1

profile-sync hook

The Mechanism

The Decoder Ring is recipient-facing

It's a guide for the recipient's Amplifier: when your human receives a message, use it to extract what THEY need, transformed for how THEY consume information. The right signal reaches the right person.

Left alone, though, that signal degrades.

1A message leaves in the sender's voice.
2The recipient's Amplifier reads the Decoder Ring.
3It extracts what THEY need, transformed for how THEY consume information.

The Complication

Left alone, signal degrades into noise

my-voice's own problem framing: over-explaining drains time and creates a communication bottleneck. And generic AI cleanup adds an "AI smell" that strips the signature elements carrying the signal.

So what does it look like to encode voice at real scale?

The time drain

Over-explaining → exhaustion, reduced output, communication bottleneck.

The "AI smell"

Generic AI output loses authenticity — signature elements like "Hmmm," "w/", "let's" get stripped away.

Lost continuity

A session ending with no prior start emits an IncompleteSession signal; spikes indicate upstream event loss.

The Named Example

samclone: a persona built from a month of real work

samclone is a working persona bundle derived from analysis of real sessions — 662 interaction turns across 82 AI sessions over 30 days, plus 11 Substack essays.

That messy month becomes something reusable.

662

interaction turns

82

AI sessions

30

days of sessions

11

Substack essays

The Turn

A month of sessions compresses into 11 composable files

samclone encodes voice into a layered set of composable persona files — voice, values, session-patterns, situation-profiles, exemplars — totaling 1769 lines, with ~60 real operational quotes.

Encoded and portable — now it can be delivered.

11

composable persona files

1769

total lines

~60

real operational quotes

The Payoff

The composable bundle is the delivery vehicle

samclone packages a whole person's voice, values, and decision-lens into three reusable agents. my-voice's sync hook keeps profiles fresh across long sessions and portable to ~/.amplifier/my-voice/profiles.

Which points at a pattern worth keeping.

→sam-proxy, sam-evaluator, sam-writer — three reusable agents.
→hook-my-voice-sync keeps profiles fresh across long sessions.
→Synced to ~/.amplifier/my-voice/profiles — the signal made portable.

The Pattern to Keep

Treat voice as infrastructure

Capture how a person communicates, encode it into a composable bundle, and sync it — so your signal reaches people intact instead of collapsing into noise.

Sources

Research Methodology

Data as of: latest my-voice commit 2026-04-20; samclone extraction 2026-03-14.

Feature status: my-voice — shipping bundle v0.1.0; samclone — working persona bundle v0.2.0.

Primary sources: ramparte/amplifier-bundle-my-voice (local ~/dev/ANext/my-voice), ramparte/samclone (local ~/dev/ANext/samclone), microsoft/amplifier-foundation (included by my-voice).

Commands run:

git -C ~/dev/ANext/my-voice log --oneline | wc -l = 21 commits (16 Brian Krabach + 3 Salil Das + 2 Sam Schillace)
wc -l ~/dev/ANext/my-voice/templates/*.md = 151 + 151 = 302 template lines
grep -iA8 'What This Does' ~/dev/ANext/my-voice/README.md → Voice Profiles / Message Cleanup / Decoder Rings
sed -n '1,8p' ~/dev/ANext/my-voice/templates/DECODER_RING_TEMPLATE.md → recipient-facing transformation guide
grep -iE '662|82|30 days|11 Substack' ~/dev/ANext/samclone/bundle.md → 662 turns / 82 sessions / 30 days / 11 essays
python3 json.load session_summary.json → total_sessions 90, with_conversations 82, total_turns 662, extraction_date 2026-03-14
wc -l ~/dev/ANext/samclone/persona/*; grep -c 'quote:' persona/voice-exemplars.yaml → 1769 lines / 11 files / 66 quotes (~60 per bundle.md)
grep -iE 'IncompleteSession|event loss' ~/dev/ANext/amplifier-context-intelligence/README.md → continuity health signal (corroboration only)

Gaps: The "298-line voice profile" / "347-line decoder ring" figures could not be located locally; verified templates are 151 lines each. The 11-essay count is asserted in samclone/bundle.md but not independently re-counted. ~/.amplifier/my-voice/profiles did not exist locally.

Primary contributors: my-voice — Brian Krabach (16 of 21 commits, ~76%); samclone — Sam Schillace (2 of 2 commits).