Three moves, one engagement. First decide whether to build the ladder or wrap an incumbent. Then deliver through the Edge product suite. Then operate it through the pod, the five-stage spine, and the gates that turn signed scope into measured EBITDA.
Rebuilding what already works is the single most expensive thing a client can do: it re-opens compliance approvals, freezes the operating cycle, and produces throwaway work. The disciplined alternative is the strangler-fig: bolt a thin, controllable layer onto the incumbent's last mile, run both in parallel, keep it reversible.
| Posture | When it applies | What BOD builds | Economics |
|---|---|---|---|
| Build the ladder | Greenfield, or the existing foundation is genuinely weak / ungoverned | Foundation-up, L1 to L5 | The L1-L5 investment bands |
| Wrap the incumbent | A packaged system already owns governed data, compliance, and workflows; the gap is a locked capability layer | Only the missing domains, typically the orchestration harness (D6), evals (D8), guardrails (D9), at the last mile | An order of magnitude smaller: thousands to tens-of-thousands, weeks |
Wrapping is valid only when you have verified the incumbent's data foundation actually is sound (the same D1-D2 check). If the foundation isn't good, a wrap just automates the error faster, and you're back to building the ladder. "The data is fine, the output layer is locked" is an assumption to test in Discover, not to accept.
Best when: you need control plus a fast output fix without disruption.
Trade-off: two systems to keep in sync as workflows grow.
Best when: many more workflows are coming and you want one home.
Trade-off: new-platform approval and migration effort up front.
Best when: a multi-year mandate with a funded platform team.
Trade-off: most flexibility, but more to secure and approve.
The wrapper plus sub-agent routing: how work is decomposed and fired. Optimized for control and inspectability. We change the harness without re-platforming.
Where the agent runs and how it's governed. Optimized for governance, identity, data proximity. In-tenant for regulated clients. We change the platform without rewriting the harness.
It delivers control and the output fix now, at the lowest cost, and the tool you wrap with is the same one you'd grow into for B or C, so every future option stays open. A migration, if it ever comes, stops being an event: you shift workflows in parallel until one day it has already happened.
A growth-equity portfolio company runs Harvey at ~20 seats. Quarterly reporting is already down to ~1 hour per analyst; the irreducible hour is qualitative commentary that's too generic. Diagnosis: not a data problem, a locked harness (Harvey's Agent Builder is fixed on the OpenAI Agents SDK). Recommendation: exhaust Agent Builder first, then wrap the residual last mile (CrewAI lean; Databricks if they want the platform). Roughly $15-45K for the first workflow in 4-6 weeks, Harvey contract unchanged.
A specific workflow can skip the intermediate levels and go straight to an L4 agent. Each level transition forces the organization through its change-management gauntlet, so dragging a team through L2, then L3, then L4 pays that adoption tax three times. For the right workflow, do it once, at the level that delivers the outcome.
Skip levels, not layers. A governed data foundation and governance are never skippable. If they aren't real, leaping to L4 just automates the error at higher autonomy, worse not faster. Wrapping an incumbent is the most common leapfrog: the packaged system supplies the governed foundation, and BOD bolts an L4 agent onto the last mile without the organization ever climbing L2 or L3.
Edge is how engagements compound: orchestration-agnostic, cloud-portable, built for middle-market PE. Assess diagnoses the tier and writes the 90-day plan. Deploy ships the L2 substrate and the first L3 agents. Scale is the control plane we're building for L3–L5. Advisor repeats the playbook across the portfolio.
Read-only assessment on AWS, Azure, GCP, Snowflake, or Databricks. A foundation-maturity score across six dimensions: data integrity, governance, security, observability, identity, AI-readiness. The same diagnostic doubles as a free read for PE deal teams on a target. Credentials deleted at engagement close.
Engagement: 2-hour to 2-week session. List $25K, waived for partnership portfolios; free sponsor-side on a target.
Ships the L2 substrate and the first production L3 workflow agents into a function in 90 days, finance variance, sales SDR, support tier-1, AP, recruiting, product research. Each wired to the warehouse, shipped with an eval harness, gated and observable.
Engagement: 60–90 day phased build per function. Outcome-based pricing where appropriate.
BOD's proprietary Agent Ops control plane: a product-in-build, phasing through 2026. Governance, audit, cost attribution, RBAC, SSO, connector catalog, and a managed tier targeting 99.9% SLA. Orchestration-agnostic; runs on LangGraph, CrewAI, Bedrock, Mosaic, or Cortex.
Engagement: Annual managed service. The recurring-revenue centerpiece.
The standing partnership across a portfolio: repeating the Assess → Deploy → Scale playbook company by company, and reporting EBITDA back to the fund. An operating-partner-grade view of AI maturity across the cohort.
Engagement: Ongoing retainer at the fund level.
A six-question survey to every portfolio CEO, then a 2-4 week per-company engagement. Delivers an exec summary, opportunity register, 90-day plan, and durability flags. List $25K. Waived for partnership portfolios.
T&M against Deploy blueprints, or outcome-based pricing with fees triggered by agreed EBITDA, revenue, or efficiency. Operating-partner visibility on every proposal.
Recurring patterns become portable scaffolding (IaC, K8s, AI coding scaffolding, reusable workflows). Each company opts in. None is coerced.
Urgency at the portfolio level. A BD signal the operating partner can use, companies ready to move, named not inferred. Durability risks flagged early: companies whose AI posture is a real exit problem show up before the buyer sees them. A shared, apples-to-apples maturity map across the fund.
The standard delivery unit. No engagement is staffed without all three named, even when the Implementor count is one.
Single point of contact for the operating partner and portco leadership. Stakeholder alignment, change management, board reporting, and the EBITDA tracker.
Technical vision, architecture, quality-gate ownership. Workflow audit, data model, system design, governance. Code-review approver, success-metric definition.
Production code, automations, integrations, model deployment. Testing, observability, runbooks. Pair-programming with portco engineers.
Lead with the number. Use the operator's language (money, time, throughput, risk), not architectures. End with a decision the operating partner or sponsor must make in the next 30 days.
Every engagement moves through five stages; if the exit criterion isn't met, the stage repeats. The level is read in Discover, the stack chosen in Define, proven in Validate, shipped in Build, compounded in Run.
| Stage | Key exit criterion | What the POV contributes |
|---|---|---|
| 1 · Discover | Workflow inventory with cost/time on the top 10 candidates; current-state diagram corrected by portco IT. | Place the company on L1-L5 using the capability spine: which domains are core vs. absent. |
| 2 · Define | Top 5-7 opportunities scored with ROI; architecture review signed; sponsor + OP sign the 90-day plan. | Set the target level (current+1, or a leapfrog to L4); choose the platform lane on data gravity. |
| 3 · Validate | MVP demonstrated end-to-end; measured results within the modeled range; no critical assumption broken. | Stand up the eval harness (golden dataset + regression), so "measured results" is provable. |
| 4 · Build | Gates signed; 30 days in production with no P1; KPI movement measured; knowledge transfer complete. | Ship on the level's prescriptive stack; the control plane wires cost attribution and audit. No over-build. |
| 5 · Run | Portco team handled 3+ incidents; docs refreshed; KPI tracker self-service; run-handoff plan signed. | Identify the next rung; expansion moves the client up a level via the transition triggers. |
Each gate is signed in writing by a named approver, and each carries an AI-specific criterion on top of the standard check.
How BOD engages is a deliberate decision, not a default, orthogonal to the maturity level.
The control plane run as a managed service, target 99.9% SLA, recurring revenue. Viable when the orchestrator is hostable multi-tenant and the client values continuous outcome ownership.
Co-build + knowledge transfer + partial handoff. The default for most engagements.
Deliver and fully hand over. For one-shot scopes where the client has internal engineering muscle.
Point the client at a low-code tool (n8n/Zapier) they run themselves. Reserved for genuinely simple glue needs.
custom planning or decision logic dominates the value; audit and replay must be perfect (regulated); or the packaged framework's opinions actively misfit the problem.
"Anyone can throw information into Claude and get a research brief that says they're behind. But once they're talking to you guys, the pressure is on the table."
Sep Taghavi, Lead EdgeDiscover the level, find the highest-conviction use case, decide build-vs-wrap, and leave with a 90-day plan an investment committee can underwrite. Two hours, read-only, credentials deleted at close.
