The data loop for coding agents

Let your coding agent make decisions with data.

Telemetry gives agents a feedback loop: instrument what matters, query real outcomes, and decide what to improve next—with SQL you can inspect.

Create free workspace See the setup

Real outcomes · Inspectable SQL · Human-reviewable decisions

agent · telemetry-monorepo

running

›

Improve checkout completion without hurting latency. Instrument the flow first, deploy, use Telemetry to find an opportunity, fix it, then verify the change helped.

Agent activity

step 1 of 8

●

PlanInspecting the checkout flow and defining success…

Plan

Measure before changing behavior.

1. Record completion, duration, step, and release
2. Deploy the event contract in v42
3. Compare outcomes before making a fix

Prompt

Instrument

Deploy

Query

Improve

Verify

The decision loop

From prompt to verified improvement.

The agent instruments before it optimizes. Then it deploys, queries the outcome, finds an opportunity, makes the fix, redeploys, and confirms the result actually improved.

Define what better means

Give the agent a goal and guardrails: improve checkout completion without increasing latency or errors.

Ship the measurement first

It instruments the outcome, tests delivery, and deploys the event contract before changing behavior.

Improve, then verify

It queries the results, finds an opportunity, ships a fix, and confirms the outcome improved after redeploy.

Decision proof

Ask what changed—not just what happened.

In this loop, the agent compares checkout releases, finds a slow path, ships the fix, and checks that completion improved while p95 fell. The same pattern works for growth, reliability, AI quality, and cost.

See agent observability Inspect the SQL

Live

Agent question

Did v43 improve checkout without slowing p95?

EVIDENCE

Completion

68.4%

p95

1.1s

Abandoned

8.7%

Checkout completion by release

v43: +9.2%

Better decisions

The agent sees the tradeoff, not one metric.

A faster run is not better if acceptance falls. A cheaper model is not better if handoffs rise. Telemetry lets an agent reason across the outcome and the cost of producing it.

Connect changes to outcomes

Compare releases, prompt versions, models, routes, and workflows against the result users actually received.

Make tradeoffs explicit

Keep quality, success, latency, and cost in the same query so an optimization cannot hide its downside.

Keep the reasoning auditable

The event contract, SQL, result, and caveats stay visible to the human responsible for the decision.

Decisions worth measuring

Give the agent a question that changes the next move.

Start where guessing is expensive. Each workflow becomes a repeatable evidence loop, not another dashboard to maintain.

View all use cases

A good decision starts with a bounded event.

Give the agent stable outcome, release, duration, and workflow fields it can compare before and after a change. Keep secrets and raw payloads out so the evidence stays safe and legible.

Completed outcomeStable dimensionsNo raw prompts

Read the docs Send an event

checkout.ts · completion boundary

Ingesting

telemetry.log("checkout_attempt_completed", {
  attempt_id,
  status: "completed",
  duration_ms,
  step,
  release,
});

JSONEvent

AutoSchema

SQLReview

Humans in the loop

The evidence is shared, not hidden in the agent.

Teach the loop

Give your agent better analytical context.

All resources

Give your agent a feedback loop.

Start with sample outcomes. Review the query. Connect real traffic when the decision is useful.

Get started for free Read skill.md

Let your coding agent make decisions with data.

From prompt to verified improvement.

Define what better means

Ship the measurement first

Improve, then verify

Ask what changed—not just what happened.

The agent sees the tradeoff, not one metric.

Connect changes to outcomes

Make tradeoffs explicit

Keep the reasoning auditable

Give the agent a question that changes the next move.

AI agents

Model costs

API reliability

Background jobs

Product analytics

Security events

A good decision starts with a bounded event.

The evidence is shared, not hidden in the agent.

Give your agent better analytical context.

Agent setup

Event schemas

SQL recipes

Integrations

Give your agent a feedback loop.