Packages
A published ladder — not an open-ended pilot
Every engagement starts fixed-price and fixed-scope, ends in a working artifact, and rolls its fee forward when you continue. You own the code, the runbooks, and the tooling — at every rung.
The ladder
Step 01·2–3 weeks
Readiness & Pilot-Rescue Audit
Start here — starting fresh or restarting.
For teams asking what to automate first — and for teams with a stalled pilot or an internal RAG build that never reached production. A preliminary 2025 MIT NANDA report estimated that about 95% of pilots in its limited, self-reported interview sample showed no measurable P&L impact. That is a sample estimate, not an industry-wide failure rate; the audit is designed to reduce the deployment risks behind stalled pilots.
from $15,000
100% credited toward a build started within 90 days
- Readiness scorecard across data, systems, permissions, and governance
- Failure-mode analysis and an eval harness for anything you've already built
- Prioritized workflow shortlist with cost-to-build estimates and a named metric per workflow
- A working agent slice built against one real workflow — deployed in your environment where access allows, otherwise sandboxed
- Platform cost & licensing model — what your agents will actually cost to run, across model tokens and platform credit meters
- A 90-day roadmap and an executive readout
Step 02·4–8 weeks
Agent Build Sprint
One workflow, one metric, live in production.
The anti-PoC. We take one back-office workflow — invoice-to-pay, service-desk deflection, onboarding — and ship it to production on your systems of record, behind loop-and-verify guardrails, with the business metric written into the contract.
from $75,000
Fixed fee, scoped to one workflow and one metric
- Production-track from day one: sandboxed, verified, then promoted
- An eval harness for the contracted metric — regression evals gate every release
- Tracing and observability wired from day one, with human-in-the-loop escalation and a full audit trail
- Your team shadows the build — enablement is not an add-on
Step 03·Quarter scale
Production Build
From one workflow to an operating layer.
Additional workflows roll onto the foundation the sprint proved: shared orchestration, shared governance, shared integrations. Once production data exists, outcome-linked terms are on the table — we price against results you can meter, not against promises.
Scoped per engagement
Fixed implementation fee with performance adjustments + monthly minimum — never open-ended T&M
- Fixed implementation fee per added workflow
- Shared governance layer: one audit trail, one permission model
- Outcome-linked pricing available once the metric has a baseline
- Model and API costs pass through at cost — never marked up
Ongoing·Monthly retainer
Managed Agent Operations
Agents degrade without operations. Ours don't.
Monitoring, eval regression, drift and model updates, cost optimization, and a monthly report against the contracted metric. Gartner projects that more than 40% of agentic-AI projects will be cancelled by the end of 2027, citing escalating costs, unclear business value, or inadequate risk controls. The retainer is designed to manage those operational risks.
Per maintained workflow
Business-hours SLA · three-month minimum, thirty days' notice · retainer slots deliberately capped
- Continuous evals run against every model and prompt change
- Drift detection between documentation, data, and agent behavior
- Cost tracking across model tokens and platform credit meters — sized so production agents never get switched off at an overage
- A monthly report against the contracted metric — plus one scoped improvement each month: a new integration, an expanded eval suite, or a workflow extension
What comes standard
However you engage, three things are always true.
An embedded pod, when it matters
For deeper work, one to three senior engineers embed directly in your stack and your standups, shipping production agent code against your systems of record. We run one pod at a time — deliberately. If the slot is taken, we'll tell you when it opens.
The hand-back guarantee
Every engagement ends with your team able to own what we shipped. You hold the IP, the runbooks, the eval suites, and the tooling, and you can take operations in-house at any time. No lock-in is a contract term, not a promise.
Model-neutral, by contract
Frontier AI labs now operate or back enterprise deployment ventures, and those ventures are naturally strongest on their sponsors' stacks. We work the same forward-deployed way, inside your org and against your real systems, but model-neutral by contract: we pick the model per workflow and swap it when the numbers say so. Paired with the hand-back guarantee, nothing we build deepens your dependence on any single vendor.
Agent interfaces we ship disclose AI interaction by default, in line with EU AI Act Article 50 transparency obligations applicable from 2 August 2026. Our documentation practices are designed to support Annex III high-risk requirements where relevant under the current implementation timeline, scheduled from 2 December 2027. Engineering practice, not legal advice.
Not sure which rung?
A thirty-minute conversation is enough to point you at the right entry — or to tell you honestly that you don't need us yet.