Lifecycle marketing's agent gap isn't the send. It's the brief. A churn-signal agent can flag 4,000 at-risk subscribers in seconds. A copy agent spins six winback variants in a minute. But the brief, the segment logic, the why-this-offer reasoning, the rejected drafts, all of it evaporates the moment the campaign goes live. Next quarter, when revenue dips and someone asks "why did we send 15% off to lapsed VIPs in February?", nobody can answer. The send is auditable. The thinking that produced it is not.
Klaviyo was built to send, segment, and report. It was never built to be the durable home for agent reasoning. Those artifacts need somewhere to live that points back at Klaviyo without replacing it.
The architecture
Klaviyo stays the system of record for profiles, flows, campaigns, and sends. Dock holds the agent output: briefs, draft variants, audience analysis, churn flags, post-send retros. Each row carries a klaviyo_profile_id or klaviyo_campaign_id pointer, so the workspace and the platform stay linked but never duplicated. Agents read fresh from the Klaviyo API on every run, so no stale state and no shadow database. Dock's consent gate is the only path that triggers a send or segment update, and it fires only after human approval.
Worked example: winback drafting
A DTC skincare brand notices repeat-purchase rates softening. The churn-signal agent runs nightly against Klaviyo profile data, scoring lapsed customers by predicted CLV, last category browsed, and discount sensitivity. It writes a flagged cohort into a Dock workspace called winback-may. The copy agent picks up the top three hypotheses and drafts four subject-line and body variants per segment, citing the churn agent's reasoning inline. The lifecycle marketer reviews, kills two, edits one. Once approved, the consent gate pushes the variant to Klaviyo as a draft campaign tied to the segment. Post-send, an analytics agent reads results back and appends them to the original brief row.
Five-step data flow
- Churn-signal agent reads Klaviyo profiles and writes scored cohorts to Dock with
klaviyo_profile_idpointers. - Copy agent reads the cohort, drafts variants, attributes reasoning in a sibling column.
- Lifecycle marketer reviews, edits, approves.
- Consent gate fires the approved variant to Klaviyo as a draft campaign tied to the segment.
- Analytics agent reads send results back from Klaviyo into the brief row.
Why this matters for Klaviyo-native brands
Email and SMS are still where DTC margin lives. Brand voice consistency is hard when six agents draft in parallel with no shared workspace. Multi-store operators running three or four Klaviyo accounts get a single Dock view across all of them. Per Klaviyo's 2026 benchmark data, flow-based emails generated nearly 41% of email revenue from just 5.3% of sends, so brief quality on flows compounds.
If you want agents drafting in a workspace humans review before anything sends, start with the ecommerce overview.
FAQ
Brand voice with multiple agents drafting? Voice guidelines live as a workspace artifact agents read on every run. Rejected drafts stay in the row history. See agent collaboration and drafted-by-agent.
Who owns the audience handoff? The Dock row is the handoff. Churn agent writes the cohort and rationale, copy agent drafts against it. Neither touches Klaviyo segments until a human approves.
A/B testing with agents? Copy agent drafts variants in the same brief row, the marketer ships an A/B, Klaviyo runs the split, the analytics agent writes results back per variant. See running an ecommerce stack with AI.
CAN-SPAM, CASL, consent compliance? Klaviyo stays the consent system of record. The consent gate enforces that no send fires without human approval, and every approval is logged. See agent audit and compliance and Dock + Shopify order pipelines.