# Reading Benmore and Agent Factory together

Reviewed 14 September 2026. This guide informs the independent FDE curriculum; it is neither an endorsement nor an adaptation of either author's complete book. Exercises in the [field workbook](field-workbook.md) are editorial applications.

Use the [paired reading companion](reading-companion.md) for all twelve sessions and the [material review record](material-review.md) for the exact scope of this revision.

## Source roles

| Source | Useful contribution | How the guide applies it |
| --- | --- | --- |
| Richard Buehling, *Forward Deployed* (Benmore, February 2026) | Discovery before commitment; explicit specifications; prototype gaps; capable receiving teams | Evidence ledger, conditional scope, release-gap register, and witnessed handoff |
| [Agent Factory: About](https://agentfactory.panaversity.org/docs/about) | Human direction and review around agent execution; session work versus ongoing workers | Direction-and-review lesson and worker operating contract |
| [Spec-Driven Development](https://agentfactory.panaversity.org/docs/spec-driven-development-crash-course) | Written intent and acceptance maintained through change | Delivery specification with behavioral examples |
| [The FDE AF Model](https://agentfactory.panaversity.org/docs/ecosystem/fde-af-model) | Governed domain knowledge and reusable delivery assets | Source register and a reviewed reuse proposal |
| [Designing the Vertical System of Record](https://agentfactory.panaversity.org/docs/ecosystem/designing-the-vertical-sor) | Examine inherited steps against outcomes and obligations | Keep/change/remove record with an accountable reviewer |
| [Eval-Driven Development](https://agentfactory.panaversity.org/docs/eval-driven-development-crash-course) | Evaluation as part of iterative worker development; graders also need checking | Version comparison, explicit failure categories, and human calibration |

## Benmore chapter guide

The supplied file is **Forward_Deployed_Benmore.pdf**. Its title page names Richard Buehling and February 2026. Page references below are the book's printed numbers; PDF viewer page numbers are four higher. The user-supplied original was inspected locally and is not included in the published website. No public download or redistribution license was established.

| Read | Printed pages | Question to bring to it |
| --- | --- | --- |
| Chapters 2–3 | 8–17 | Which assumptions have become commitments without validation? |
| Chapters 4–6 | 18–35 | Do goals, scope, risk, time, budget, and the working relationship align? |
| Chapter 7 | 36–50 | What must be learned, decided, and specified before delivery? |
| Chapter 8 | 51–56 | What would justify pausing or ending this engagement? |
| Chapter 10 | 63–71 | Which gaps separate the prototype from accepted, operable software? |
| Chapters 11–12 | 72–76 | How does embedded delivery change day-to-day decisions? |
| Chapter 13 | 77–81 | Can the receiving team recover and change the system? |
| Chapters 14–17 | 82–90 | Who maintains the result, and where does support stop? |

## Differences to retain

Benmore describes one consulting firm's delivery practice. Agent Factory presents an educational and business architecture for AI workers. Their vocabulary overlaps without making their definitions universal. An FDE may work on deterministic software, data platforms, or AI; the role is not defined solely by manufacturing Digital FTEs.

Use direct observation to discover what the existing work protects, then examine which steps can change. These are complementary questions: copying the workflow preserves waste, while redesigning without evidence can discard a necessary control.

A specification records agreed intent, but it can still encode a mistaken assumption. Test the underlying problem and the resulting implementation. Keep security, tests, and review active during development; do not interpret a later production-readiness phase as permission to postpone them.

The site's practice gates and templates are its own synthesis. No particular model, orchestration framework, context-reset rule, cloud platform, or meeting cadence is required to pass them. Choose tools that meet the customer's needs and can be maintained by the receiver.

## Claims not adopted as evidence

- Agent Factory's 10-80-10 rule is a teaching heuristic, not a verified time ratio or productivity benchmark.
- Digital FTE is the source's term for an AI worker; it does not demonstrate equivalence to a human employee's responsibilities or hours.
- Benmore's prototype percentages, support percentage, and commercial examples describe that author's practice. They do not establish market pricing or universal acceptance criteria.
- Claims about AI replacing work, earnings, salaries, or enterprise pilot failure require their own evidence. Neither source was used to revise the website's job counts or salary tables.
- Descriptions of named tutors, building partners, connectors, or infrastructure do not establish current availability, security, or operating cost. The review did not test the Agent Factory ecosystem.
- The workbook teaches workflow design, not professional advice for a particular legal, medical, or financial decision.

## How to study

Begin with the chapter or lesson matching your current uncertainty. Build its artifact in the workbook, ask a peer to challenge one assumption, and keep the failed example. Continue only when you can explain the decision and demonstrate the corresponding evidence.