Production agents that call tools, run multi-step workflows, and have a fallback story when the model is wrong — built inside your delivery process and behind your brand.
Agent pod, predictable spine
These are the conversations agency owners describe when an agent brief is in front of them and the in-house team has shipped one orchestrator script but not yet a system that survives a billing-day incident.
The video looked great. The buyer is quoting it back at you. Production wants p95 latency, retries, evals, observability, a cost ceiling, and a written fallback when the agent picks the wrong tool. We re-architect the demo into a system that survives a Monday-morning audit.
It books the wrong meeting, drafts the wrong reply, fires the wrong webhook. We add evals against a labelled task set, scoped tool permissions, and a guardrail layer — without throwing away the working autonomy.
Each new agent multiplies the prompt, eval, observability, and incident surface. We come in with an orchestration pattern and a per-agent runbook so the team can ship the next two without doubling the on-call burden.
OpenAI Agents SDK, LangGraph, Anthropic computer-use, Vercel AI SDK, custom — the trade-offs are real and the team keeps re-litigating. We pick the platform inside the architecture readout, write the trade-offs down, and move to build inside the same week.
Agent work runs eval-first. The pod doesn't ship a tool-using agent into a real workflow without a labelled task set behind it.
Working session with your delivery lead and the buyer-side stakeholder. We pressure-test the brief — what the agent must do, which tools it can call, what "wrong" looks like, and the cost ceiling. NDA and SOW signed under Salt Technologies, Inc.
Architecture readout: chosen platform (OpenAI Agents SDK, LangGraph, custom), tool surface, state model, observability, guardrails. In parallel, we build the labelled eval set every prompt and tool change runs against in CI.
Iterative agent build with weekly working review. Eval pass rate, run-time cost, and latency tracked from build one. Guardrails and human-in-the-loop checkpoints land before the agent ever touches a real system of record.
Production deployment behind a feature flag, with a percentage-rollout plan and a written runbook for incidents. Post-launch support transitions cleanly into a Support & Maintenance retainer for model drift, regression on new model versions, and v1.x feature work.
Agent work is long-arc. The two shapes below are the engagements where production-grade agent systems actually land — a new service line under your brand, or a delivery pod inside an active client SOW.
Capability lead + delivery support
Standing AI Agents up as a defensible new service line on your rate card — capability lead, pricing template, monthly retainer with quarterly review.
Agent pod inside your SOW
Active client work that has been sold and now needs end-to-end agent throughput inside your shared channel — long-arc, structured, named delivery owner.
Starting weekly rate for a single-capability AI pod. Multi-track AI pods (Agents + RAG, Agents + Integrations) and pods that mix capabilities are quoted at the highest applicable rate. Final SOW is scoped against the brief.
We work inside the agent tooling your team and your client already use — no parallel platform, no "we'll just rebuild it our way" surprise.
AI Agents shows up most often where the agency is already shipping AI work and the next brief asks for autonomy — not just a smarter chat box.
HubSpot partners shipping agentic workflows across the HubSpot stack — sales triage, support deflection, lifecycle nudges — where the standard automation surface ends and a real tool-using agent begins.
Boutique design and product studios shipping agent-driven product surfaces for clients — where the brief asks for an interface that takes actions, not a new chat tab.
Shopify partners adding agentic merchant tooling, shopper assistants that actually book and update, and back-office triage that runs without staring at a queue.
Every AI Agents 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 agent pod, a single AI specialist hire, or a freelance prompt engineer weigh the same trade-offs. These pages address them head-on.
All four AI capabilities side by side — useful when the brief sits between Agents and one of the other tracks and you want to see the routing decision laid out.
The most common shape for adding AI Agents to your rate card — capability lead, pricing template, quarterly review on whether the new line earns its place.
When an agent pod is cheaper, faster, and lower-risk than the next senior AI hire — and when staffing up is the right call.
Direct answers to the questions that come up on almost every AI Agents scoping call.
RAG grounds an answer in your client's content — the model retrieves and cites, but doesn't act. Agents take actions — they call tools, update systems of record, kick off workflows. Most agentic systems we ship combine both (an agent with a RAG tool in its toolbox), but the *primary* capability on the SOW is the one that defines the system. We make this call inside the brief & feasibility step in week one.
Three layers. First, scoped tool permissions — the agent literally can't call tools it isn't authorised for. Second, an eval suite that runs in CI on every prompt or tool change against a labelled task set the buyer signs off on. Third, human-in-the-loop checkpoints on irreversible actions (sending money, deleting data, contacting a customer). All three are part of default scope; we don't ship without them.
We don't have a single default. The choice between OpenAI Agents SDK, LangGraph, Anthropic tool use, Vercel AI SDK, and a custom orchestration sits on platform maturity, your team's existing stack, and the latency / cost envelope for the workload. The architecture readout in week one names the platform with the trade-offs written down so your team and your client can refer back.
By default the client's. We work inside their OpenAI, Anthropic, or other model-provider accounts under named seats. Where you (the agency) hold the account on behalf of the client, we operate inside yours. Either way, the account ownership and cost-attribution boundary is documented in the SOW so there's no ambiguity at month-end.
Cost ceiling per agent run, alerting on per-day spend deltas, and a circuit-breaker that pauses the agent surface above a threshold the buyer sets. Cost dashboards land before the production rollout step, not after. We've seen enough cost incidents in our previous lives to design for them up front.
Production agent systems aren't a one-week capability. The most common starting shape is a 4–6 week pod inside an Invisible Delivery Team SOW, sized around the live brief and the integration surface. For shorter agent work (a tightly scoped feature, an eval-suite stand-up, a guardrail audit), we'll quote 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 (Agents + RAG, Agents + Integrations) and pods that mix capabilities (Agents + UI/UX, Agents + Backend) are quoted at the highest applicable rate. Final SOW is scoped against the brief; the rate is the floor, not a ceiling.
Most agent systems graduate cleanly into a Support & Maintenance retainer post-launch — model drift monitoring, eval regression on new model versions, guardrail tuning, 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 code, prompts, evals, tool wrappers, 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.
Sales-side triage, support deflection, internal ops, or a client-facing agent surface. Either way, the conversation starts with the work — not with a deck.
