Internal automations that take a manual, repeated workflow off your client's plate — built on n8n, Make, Zapier, or first-party serverless, with AI in the loop where it earns its keep, and the run-book and monitoring to keep them alive in month two.
Automation pod, predictable spine
These are the conversations agency owners describe when an automation brief is in front of them and the in-house team has wired up a Zap or an n8n flow but not yet a system that an ops team can rely on at 9am on a Monday.
The automations were built in a sprint, never instrumented, and have been quietly dropping records for weeks. We re-build the critical flows on a real platform, add observability and alerting, and put one named owner on the runbook so failure stops being invisible.
Half the steps are deterministic, half need a judgment call. We design the flow once — deterministic where it should be, AI-assisted where the judgment is — and ship it as a production automation with retries, idempotency, and a clean human-in-the-loop checkpoint where the call genuinely needs a person.
An LLM node went into a working pipeline without structured output validation, retries on bad responses, or a cost ceiling — and now the flow fails on shape, runs up the bill, or both. We retrofit the AI nodes with the production scaffolding they should have shipped with.
CRM in, helpdesk out, billing system updated, Slack notification, audit log written — with conditional branches, retries, idempotency, and a fallback if any one system is down. Past the point Zapier handles cleanly. We re-platform onto n8n or first-party serverless and ship it with the operational discipline a real workflow needs.
Automation work runs reliability-first. The pod doesn't ship a flow into a real system of record without retries, observability, and a documented failure mode behind it.
Working session with your delivery lead and the buyer-side stakeholder. We map the workflow, the systems involved, the cost-of-failure for each step, and where AI genuinely earns its place. NDA and SOW signed under Salt Technologies, Inc.
Architecture readout: chosen platform (n8n / Make / Zapier / first-party serverless), AI placement with structured output contracts, retries and idempotency strategy, observability and alerting plan. The flow map is signed off before any node gets wired.
Iterative build with weekly working review. Flows wired with retries, dead-letter queues, idempotency keys, and per-node logging. AI nodes get structured output validation and per-run cost ceilings. Observability lands before production access, not after.
Gradual rollout against real volume, dashboards live for ops, and a written runbook your client's team can act on without a Slack ping. Post-launch transitions cleanly into a Support & Maintenance retainer for drift, third-party API changes, and v1.x feature work.
Automation work tends to come in two shapes — a defensible new service line on your rate card, or a queue of bounded, time-boxed automation jobs delivered through an Overflow Pod. Pick by the shape of the demand, not by the brand of the platform.
Capability lead + delivery support
Standing AI Workflow Automation up as a defensible new service line on your rate card — capability lead, pricing template, monthly retainer with quarterly review.
Time-boxed automation pod
A queue of bounded, scoped automation builds across one or more clients — short windows, named delivery owner, predictable cadence inside your shared channel.
Starting weekly rate for a single-capability AI pod. Multi-track AI pods (Automation + RAG, Automation + Integrations) and pods that mix capabilities are quoted at the highest applicable rate. Final SOW is scoped against the brief.
We work inside the automation tooling your team and your client already use — no parallel platform, no "we'll just rebuild it our way" surprise.
AI Workflow Automation shows up most often where the agency is already shipping ops-side work and the next brief asks for AI-assisted reliability — not just a smarter prompt.
HubSpot partners shipping cross-system automation across the HubSpot stack — sales triage, lifecycle nudges, support routing, billing handoffs — where the standard HubSpot workflow surface ends and a real cross-tool automation begins.
Boutique design and product studios shipping internal-ops automations for clients — content review pipelines, asset enrichment, brand-safety checks — where AI nodes accelerate the manual judgment without breaking the pipeline.
Shopify partners adding merchant-side automations — order triage, fraud-signal review, content enrichment, support deflection — where the platform's native automation surface stops and a real cross-system flow needs to start.
Every AI Workflow Automation engagement runs on the same operating spine that protects long-arc retainers and Dedicated Partner Pods — contracted through Salt Technologies, Inc.
We don't email your client, join their calls, or appear in the proposal — unless you explicitly white-list a named engineer in the SOW.
We work in your GitHub, your model-provider accounts, your hosting, and your shared channel under aliases that fit your team's naming.
Mutual non-solicitation written into every MSA, with a defined window after the engagement ends. Same clause across every track.
MSA, NDA, and engagement SOW are issued by Salt — the Delaware C-Corp behind AgencyRelay.
The same operating spine sits underneath every AgencyRelay capability. Read the no-poach and confidentiality page for the contractual instruments behind these defaults.
Most agency owners deciding between an automation pod, a single ops-engineer hire, or a freelance Zapier consultant weigh the same trade-offs. These pages address them head-on.
All four AI capabilities side by side — useful when the brief sits between Automation and one of the other tracks and you want to see the routing decision laid out.
The most common shape for time-boxed automation work — short windows, queueable jobs, named delivery owner inside your shared channel.
When an automation pod is cheaper, faster, and lower-risk than the next ops-engineer hire — and when staffing up is the right call.
Direct answers to the questions that come up on almost every AI Workflow Automation scoping call.
Workflow Automation is a deterministic flow with AI nodes — the path is mostly fixed, AI helps with the judgment-call steps (classify, enrich, summarise, route), and the system runs on a schedule or a trigger. Agents are open-ended — the model decides which tool to call next, in what order, with state and retries across multiple steps. If the workflow's path is mostly known, it's automation; if the path has to be reasoned out at run-time, it's an agent. Most briefs pick one cleanly; we route the ones that don't in week one.
Per flow, not per project. Zapier is fastest when the flow is short, the connectors exist, and the buyer's team owns operating the result. Make handles richer logic and visual debugging without going custom. n8n (self-hosted) is the right call when the flow needs custom code nodes, sits inside the client's network, or needs costs decoupled from a per-task billing model. First-party serverless (Lambda, Cloud Functions, Vercel Functions, Inngest, Temporal) is the right call when reliability, durable execution, or vendor-lock concerns dominate. The architecture readout in week one names the platform per flow with the trade-offs written down.
The honest places: classification (route this ticket / record / lead), enrichment (extract structured data from unstructured text), summarisation (collapse a thread into a short note for a human reviewer), draft generation (a first-pass reply, summary, or follow-up that a human approves), and routing decisions (which branch of the flow to take). AI nodes always come with a structured output contract (JSON schema or function-calling), retries on bad responses, and a per-node cost ceiling. We don't add AI nodes for novelty.
Three layers. First, every flow has one named owner inside the client's team — written into the runbook. Second, observability ships before production access — success / error dashboards, alerting on silent failure, and a per-flow heartbeat. Third, dead-letter queues and idempotency keys so a partial failure can be replayed safely. Most graveyards we inherit fail on layer two — they were built with no monitoring at all.
Per-node cost ceilings, alerting on per-day spend deltas, and a circuit-breaker that pauses the flow above a threshold the buyer sets. Cost dashboards land before the production rollout step, not after. Most cost incidents we've seen come from a model node inside a high-volume flow with no per-run ceiling — that's the failure mode we design out first.
Production-grade automation work isn't a one-week capability. The most common starting shape is a 2–4 week scoped engagement — either inside an Overflow Pods SOW (a queue of bounded automation jobs) or as the first quarter of a Capability Expansion engagement (standing AI Workflow Automation up as a new service line). Single-flow audits and Zapier-graveyard rescues are quoted as a tighter window against the same weekly rate.
The starting weekly rate for a single-capability AI pod is $3,200 per week. Multi-track AI pods (Automation + RAG, Automation + Agents) and pods that mix capabilities (Automation + Backend, Automation + UI/UX) are quoted at the highest applicable rate. Final SOW is scoped against the brief; the rate is the floor, not a ceiling.
Most automations graduate cleanly into a Support & Maintenance retainer post-launch — third-party API changes (the connectors that quietly deprecate), model drift on AI nodes, schema changes upstream, observability upkeep, and v1.x feature work inside a monthly envelope. Either the same pod or a smaller maintenance crew carries it on the same MSA, no second sales cycle.
Yes. IP ownership and assignment on delivered flows, code, prompts, runbooks, and supporting artefacts is written into the MSA — the work belongs to your agency (and onwards to your client per your own client contract) on payment of the relevant invoice. The Salt Technologies templates are counsel-reviewed and shared before signing.
A new internal automation, an AI bolt-on to an existing flow, or a Zapier-graveyard rescue. Either way, the conversation starts with the work — not with a deck.
