SLI and SLO tracking breaks when humans copy numbers into a doc once a quarter. In Dock for DevOps, an agent reads indicators from Datadog, Honeycomb, and PagerDuty, drafts the weekly error-budget report into a Dock table, and flags policy changes for the eng lead.
Datadog, Honeycomb, and PagerDuty stay the system of record for the raw data. Dock is the system of record for what the AGENT INTERPRETS. Each Dock row carries a pointer back to the platform record, agent identity, decision, reviewer, and timestamp. The agent re-fetches platform data via fresh API reads when it needs current state.
The Dock surface: Error Budget Ledger
| Service | Window | SLO target | Burn rate | Budget remaining | Datadog SLO ID | Honeycomb query | PagerDuty incidents | Agent draft | Reviewer | Decision |
|---|---|---|---|---|---|---|---|---|---|---|
| checkout-api | 30d rolling | 99.9% avail | 4.2x | 41% | slo_8f21 | trace.duration p99 | PD-9921, PD-9947 | Freeze risky deploys 48h | priya@ | Approved 2026-05-29 |
| search-svc | 7d rolling | p95 < 250ms | 0.8x | 88% | slo_4a09 | latency_bucket | none | No action | sam@ | Confirmed |
| billing-jobs | 30d rolling | 99.5% success | 1.1x | 62% | slo_2c11 | job.status=fail | PD-9930 | Page on next breach | priya@ | Approved with note |
Every cell points back to the source query, the drafting agent, and the reviewer. The eng lead reads the table top to bottom in five minutes.
The workflow: weekly error-budget review
Monday 09:00, the SRE agent runs. It pulls SLO status from Datadog, latency distributions from Honeycomb, and the last seven days of pages from PagerDuty. For each service it computes burn rate, fills the Dock row, and writes a one-paragraph interpretation.
When burn rate exceeds 2x on a tier-1 service, the agent drafts a proposed policy change: deploy freeze, alert threshold tightening, or rollback. It does not apply the change. It routes the row to the on-call eng lead with the dangerous-ops contract attached. The lead approves, edits, or rejects. The decision writes back with the reviewer's agent or human identity, the timestamp, and a note.
If the lead approves a freeze, the agent posts to the deploy channel and flips the freeze flag in the deploy gate. Every action is a Dock row first, an external write second.
Why this matters
Error-budget reports usually live in a stale Notion page nobody opens. Moving the interpretation layer into Dock means the agent's reasoning is queryable, the reviewer is named, and the decision has a paper trail. When a new SRE asks "why did we freeze deploys on May 14?" the row answers in one click. Same pattern as agent audit and compliance, same shape as Cloud 2.0 for engineering: humans review interpretation, not raw telemetry.
The Google SRE book defines an SLI as "a carefully defined quantitative measure of some aspect of the level of service provided" (sre.google). DORA research finds operational performance separates elite teams from the rest (dora.dev). Both assume someone reads the numbers weekly. An agent plus a Dock table makes that read cheap enough to happen.
Try it
Wire your SRE agent to a Dock workspace and let it draft next Monday's error-budget review. The eng lead reviews, not writes.
FAQ
Does the agent change SLO thresholds in Datadog? No. It drafts a row. The eng lead approves, then a writer agent pushes the change. The split mirrors Dock for data analytics.
What if Honeycomb and Datadog disagree on latency? The row records both. The agent flags the discrepancy and the reviewer decides which to trust.
How is on-call rotation attached to a row? PagerDuty schedule is read at draft time, so the row points at the human who was on call during the burn.
Can the agent auto-page on burn rate? Only through the dangerous-ops contract. Auto-pages require a pre-approved policy row. Unreviewed bursts go to a draft queue.
