Instrumenting agent work

AI agents can produce and send far more work than the person directing them could send alone. In one of our projects, agents research a recipient, build a piece of work specifically for that recipient, and send it with a short note. The dominant outcome, as with any outreach, is silence: most of what gets sent receives no reply, and this stays true even when the work is good.

The common way to handle silence is to treat it as one fact, meaning no. Under that reading the only available lever is volume, because if some fixed fraction of sends produces a reply, more replies require more sends. This is how agent-scale outreach turns into spam, and it also throws information away, because a process that reads every silence the same way cannot tell which of its own parts is failing.

Decomposing the silence

The alternative is to measure where each silence happened. Every piece of work we send out carries its own unique signed token, which is a link parameter that identifies exactly one recipient. Because the token is unique, the path from sending to reply decomposes into stages that can be measured separately: the message was sent, the link was opened, the work was started, the work was finished, a reply arrived. The early stages are measured by beacons, small requests that fire when the page loads and when the work starts playing, and the last stage is logged by hand, because replies arrive in language and carry sentiment a sensor cannot read. Each transition is recorded as a typed event against that one recipient, in the same append-only log the rest of the project runs on, so the funnel report is a query rather than a spreadsheet someone maintains. Once the stages are measured, a silence stops being one signal: it points at whichever stage leaked, and each stage has its own fix.

The stages fail for different reasons and respond to different adjustments:

• Sent but never opened. The work was never judged at all. The failure belongs to the first sentence of the note, the channel, the send time, or the sender’s standing on that channel, and improving the work itself would change nothing.
• Opened but never started. The page carrying the work failed, which is rare and usually mechanical: load weight, a broken embed.
• Started and abandoned early. The work’s opening failed, and the adjustment is to move the strongest material earlier or make the whole thing shorter.
• Finished with no reply. The work succeeded and the ask failed. This is the only case where a follow-up is justified, and it goes only to the people who finished, with a single concrete question, because chasing the people who never opened anything on the same channel is how sender standing gets destroyed.

Sending in waves

Sending continuously blurs every reading, so we send in waves and treat each wave as a sampling period. Waves are sized for information rather than volume. The first wave's job is to calibrate the stage conversion rates, so it deliberately mixes recipient types, channels, and work variants rather than concentrating on the likeliest targets. After a fixed window we read the funnel, find the one stage with a disproportionate drop, fix that stage, and send the next wave with the fix in place. We make one change per wave, because two simultaneous changes make the next reading unattributable.

Across waves the targeting corrects itself as well. Stage rates broken down by recipient segment feed the next target list, so if one segment opens everything and never finishes, and another finishes everything it starts, the next list shifts toward the second segment. The system adjusts who it approaches as well as what it sends, and both adjustments come out of the same event log.

The instrumentation also changes what a non-converting recipient is worth. A recipient who finishes the work, comes back to parts of it twice, and never replies has still delivered a reading on the work itself, separated from any reading on the ask. When the work offers choices, a choice made is a preference labeled, and a choice declined with a stated reason is worth more than most replies. A process that only counts conversions writes these recipients off, which discards most of what the wave measured, because the next wave's work is built from what the silent majority observably did.

Beyond outreach

Nothing in this pattern is specific to outreach. Any agent-driven process that produces something and then waits for the world to respond has the same anatomy: a job posting waiting for applicants, an article waiting for readers, a support answer waiting for confirmation, a proposal waiting for a decision. In every case the response path decomposes into stages, each stage can carry its own measurement, and the same two rules apply: every artifact gets its own token, so that events attach to one recipient rather than averaging over a campaign, and only one thing changes per wave, so that the next reading means something.

A system without per-stage measurement can only vary its volume, and volume is the one adjustment that makes the channel worse for everyone who uses it. A system with per-stage measurement runs every wave as an experiment. The silences, which are most of what any outreach ever hears, become the readings the next wave is built from.