The engagement model.
Tengwar is a boutique advisory firm. We embed forward-deployed engineers inside private equity portfolio companies and put frontier AI models into the operations that actually move enterprise value. The posture of the firm is the product: clean services revenue, no software resale, no model licensing markup, no vendor incentive dressed up as advice.
The firm exists because of a specific gap. Two large model-provider joint ventures are forming in 2026 — one anchored by Anthropic with Blackstone, Hellman & Friedman, and General Atlantic; one anchored by OpenAI with TPG, Bain Capital, Advent, and Brookfield. Both are single-vendor by construction. Both rely on the global systems integrators for delivery. Both put a sponsor on the cap table of the entity doing the work. That structure forecloses a set of engagements neither JV can honestly accept.
The firm PE sponsors call when they do not want the model vendor in the room is a different firm. It has to be model-neutral so that a portco CIO who wants Gemini for multimodal, Llama for on-prem, or Claude Opus for long-context coding can be served without an argument. It has to be sponsor-neutral so that a Blackstone operating partner and an Apollo operating partner can both retain the same team without flinching. It has to move faster than a six-month Big Four discovery phase because that is how operating partners actually measure value. And it has to stay through the first earn-out cycle, because that is when the operating committee finally reads the numbers.
What we sell is not a deck. It is two to ten engineers, a time-boxed plan, a production system, and a handoff document that reads like something a portco CFO can defend. Everything else is supporting scaffolding.
A four-column read across the credible alternatives an operating partner already considers. Footnote markers point to the thesis.
| Tengwar | In-house hire | Big Four / strategy AI | Model-vendor JV | |
|---|---|---|---|---|
| Speed to deploy | 4-week diagnostic, production use case inside a quarter. | Quarters to hire, longer to onboard. Dependent on the labor market for senior AI engineers. | Six-month discovery phases are common before code is written. | Delivery routed through the Claude Partner Network of global SIs — the same rate-card-heavy firms PE already complains about. |
| Model flexibility | Model-neutral by charter. Claude, GPT, Gemini, Llama — whichever wins the use case. | Flexible in principle, in practice constrained by the platform the first engineer happened to know. | Flexible in principle. In practice tied to platform alliances. | Single-vendor by construction. The Anthropic JV sells Claude; the OpenAI JV sells Frontier. |
| Sponsor conflict | No sponsor on the cap table. We can work across every portfolio without the appearance of competitive intelligence leakage. | None. | None at the firm level, though individual partners carry relationships. | The Anthropic JV has Blackstone, Hellman & Friedman, and General Atlantic on the cap table.[1] The OpenAI JV has TPG, Bain Capital, Advent, and Brookfield.[2] Every other sponsor becomes a second-class citizen. |
| Cost structure | Services revenue only. Gross margins earned on the work, not on a bundled software resale. | Fully loaded headcount plus ramp time. | Blended rate card, large teams, utilization pressure. | Services subsidized by a Claude Enterprise subscription underneath. Cheap on paper, locked in in practice. |
| Knowledge transfer | Embedded through the first earn-out cycle. Handoff to internal owners is a deliverable, not a courtesy. | Full retention — if the person stays. | Limited. Utilization targets pull consultants off the project at go-live. | Unclear. Neither JV has publicly addressed what happens after go-live. |
[1], [2] Publicly reported facts sourced and cited in the thesis.
Diagnose
- Four to six weeks on site.
- AI value-creation thesis, three to five use cases.
- Tech stack assessment and data readiness review.
- Ninety-day plan readable in one sitting.
- Sponsor-ready operating committee memo.
Deploy
- Three to five engineers embedded in the portfolio company.
- One or two production use cases live inside a quarter.
- Measured KPI lift against a pre-agreed baseline.
- Evaluation pipelines wired from day one.
- Handoff documentation written for internal owners.
Govern
- Quarterly cadence through the first earn-out cycle.
- Model drift monitoring and eval refresh.
- Sponsor reporting translated into operating-committee language.
- Value attribution tracked to the KPI that was agreed in writing.
- Renewal, expansion, or graceful exit — the portco chooses.