AI agent retention signal

AI 에이전트 리텐션 신호

Compare first-session artifact saves with later return behavior.

···
html
<div class="stage"><div class="label">FIRST SESSION → LATER RETURN · EXAMPLE</div><div class="legend"><span><i></i>SAVED OUTPUT</span><span><i></i>FAILURE SIGNAL</span></div><svg viewBox="0 0 300 120" preserveAspectRatio="none"><path class="axis" d="M15 5 V105 H295"/><path class="save" d="M15 8 L70 26 L125 39 L180 47 L235 54 L290 59"/><path class="fail" d="M15 8 L70 57 L125 76 L180 86 L235 93 L290 98"/></svg><div class="hint">ASSOCIATION, NOT PROOF OF CAUSE</div></div>
css
.stage{position:relative;width:min(94vw,760px);height:min(88vh,310px);font:800 clamp(10px,max(3.4vmin,1vw),15px)/1.25 sans-serif}.stage *{box-sizing:border-box}.label{font-size:clamp(10px,max(3.4vmin,1vw),12px);letter-spacing:.1em;color:var(--muted)}.accent{color:var(--accent)}.pill{border:1.5px solid var(--fg);border-radius:999px;background:var(--surface);padding:.5em .8em}.hint{position:absolute;right:2%;bottom:0;font-size:clamp(10px,max(3.4vmin,1vw),12px);color:var(--muted)}.legend{position:absolute;right:2%;top:16%;display:flex;gap:clamp(6px,2vmin,20px);font-size:clamp(10px,max(3.4vmin,1vw),12px)}.legend i{display:inline-block;width:9px;height:9px;margin-right:3px;background:var(--accent)}.legend span:last-child i{background:var(--fg)}.stage svg{position:absolute;left:5%;top:28%;width:90%;height:58%}.axis{fill:none;stroke:var(--line);stroke-width:2}.save,.fail{fill:none;stroke-width:5;stroke-linecap:round;stroke-dasharray:450;animation:line 4s infinite alternate}.save{stroke:var(--accent)}.fail{stroke:var(--fg);animation-delay:.3s}@keyframes line{from{stroke-dashoffset:450}to{stroke-dashoffset:0}}

In an analysis of its own agent, Amplitude found higher later retention among first-time users who had a positive session and saved output than among those with failure signals. This is an observation from one product, not a universal multiplier.

The demo compares illustrative retention curves by first-session save behavior. Forced saves could improve the metric without improving value, so inspect actual reuse and failure experiences too.

When to use

Use it to connect first agent experience quality with later return.

Open as page ↗