Live in four weeks. Same rail, then maintained.
One methodology, scaled to scope. Six phases, clear handoffs, nothing shipped without evaluation. A typical focused build runs four weeks; multi-system platform engagements run longer, on the same rail.
Map the workflow. Baseline the metrics.
Before we build anything, we sit with the people doing the work. We map the workflow as it exists today — including the handoffs, the workarounds, the things that "everyone knows" but aren't written down. We baseline the metrics: volume, handle time, error rate, escalation rate, cost.
Without this, "automation" is just adding software to a process nobody fully understands. With it, we know what "better" actually means.
Draft the swimlane. Decide what defers to a human.
We draft the target swimlane — the sequence the agent and operators will follow once it's live. We name the approval gates, the kill-switch criteria, and what constitutes an escalation. We pick the model, the storage, the orchestration, and the boundary between automated and human-in-the-loop.
This phase ends with a written spec we both sign off on. Build doesn't start until the spec is settled.
The system, the tools, the eval harness. Shadow from day one.
We build the system in your stack — agents, tools, storage, orchestration, observability. We wire the eval harness into your CI from the first commit, not the last. Once anything is callable, we run it in shadow mode against real production traffic: every input that hits production hits the agent, but the agent's output is logged not acted on.
By the end of Build, the eval harness is green and the shadow-mode logs match operator decisions on the bulk of cases.
Shadow, supervised, live. Sign off when it runs clean.
The agent walks through three states. Shadow: outputs logged, not acted on. Supervised: outputs proposed, an operator approves before they ship. Live: outputs ship, with monitored exception paths. We move forward only when the previous state runs clean against criteria we agreed to in Phase 02.
If something doesn't run clean, we go back. The rail handles re-scoping at phase boundaries — that's the whole point of having phases.
Playbooks, runbooks, training. You own it.
Your team takes the keys. We run a training session for the operators who'll work the rail every day, walk the engineers through the eval harness and the deployment path, and transfer all credentials, secrets, and admin access. The runbooks for known failure modes ship; the debugging playbook for unknowns ships with them.
Hand-over is a real event with a real checklist, not a slow fade. By the end, you don't need us in the room.
Watch for drift. Tune. Evolve.
Maintenance is the optional sixth phase. If you take the retainer, we monitor for drift, refresh the eval set as your workflow evolves, and ship small changes on demand. If you don't, you have everything you need to do this yourself — and we're a phone call away if something gets weird.
Either way: no system runs itself. We say that up front because it's true.
What if our scope is bigger than four weeks?
Same rail, more loops. A multi-system platform engagement runs the six phases for each subsystem; we sequence them and roll new ones in as earlier ones go live. The methodology doesn't change.
What if you discover something is impossible during Build?
Phase 02's design is signed off before Build starts, so the surface area for "impossible" is small. When it does happen, we re-scope at the phase boundary — not after Build is done — and adjust fee and timeline together. No surprises at delivery.
Can we skip Maintain?
Yes. Maintenance is optional. Hand-over is real — your team can run, debug, and modify the system without us. The retainer is for teams that want a steady second pair of eyes.
How much time do we need to budget from our team?
Roughly five hours per week from an ops lead during the build, plus five hours per week from an engineer for integration questions. We do the heavy lift; we need your team for context, decisions, and access.
Do you do audits or one-off reviews?
Yes — half-day to one-day operations audits with a written report. Email hello@swimlanes.co if that's what you need.