For FDEs, applied AI engineers, technical leaders, and operators responsible for the handover.
source boundaryauthority ladderclient handover
01the work after the demo
Applied AI engineersSolution architectsTechnical leadersOperators of the handoverApplied AI engineersSolution architectsTechnical leadersOperators of the handover
The gap
A correct answer is not yet a working system.
The model can be right while the operator still cannot use it.
Real work brings records, permissions, exceptions, approval paths, and somebody who owns the outcome. This is where generative AI has to earn its place.
Choose a workflow worth changing.
Make the value path, authority, and evidence visible.
Leave the client able to run, inspect, correct, and pause it.
01 / VALUE
Start with the work
Map the decision, user, and outcome before choosing a model.
02 / CONTEXT
Build a source boundary
Show what the system can trust, when it goes stale, and who may access it.
03 / AUTHORITY
Bind the action
Make assist, recommend, and act explicit—and recoverable.
04 / EVIDENCE
Evaluate the path
Test workflow fit, failure, adoption, economics, and edge cases.
05 / OWNERSHIP
Rehearse the handover
Leave operators able to run, inspect, correct, pause, and retire it.
Scope, data access, authority, acceptance criteria, ownership, and stop conditions.
02
Agent Authority Ladder
Decide whether the system should assist, recommend, or act—and bind the boundary in policy, identity, tools, and recovery.
03
FDE Deployment Loop
Observe the work, shape the boundary, build the smallest useful path, evaluate it, operate the release, and feed evidence forward.
The FDE deployment loop
Make evidence change the next decision.
The book’s practical route for enterprise GenAI deployment is not a straight line from prompt to production. It is a connected loop that keeps workflow, authority, evidence, operations, and ownership in the same conversation.
01 / DISCOVER
Watch the work
Name the outcome, workflow, systems, authority, affected people, baseline, and constraints.
02 / BOUND
Set the boundary
Record what the release may see, prepare, recommend, commit, and trigger—plus its fallback and exit.
03 / BUILD
Make the smallest useful test
Build a bounded work unit in the real environment, with human control surfaces kept visible.
04 / EVALUATE
Test the edges
Use representative, boundary, stale-source, conflicting-source, authority, adoption, and operational cases.
05 / DEPLOY
Release with a fallback
Name operators, support, approval, evidence, and a manual path before the live boundary widens.
06 / OPERATE
Learn from production
Observe corrections, incidents, source changes, workload, cost, and capability atrophy—not only infrastructure.
07 / TRANSFER
Prove the exit
Rehearse routine work, correction, incident response, and retirement with the client team.
Every loop ends in a decision: continue, narrow, pause, stop, or scale.
Start here, free
Five lessons for the work after the demo.
A short, ungated field course drawn from the final release. Each lesson has its own page, exercise, and next step. Use it for a first working pass on the method; use the book for full cases, artifacts, evaluation, operations, economics, and handover.
20 minutes · 5 lessons · no sign-up
A real course, with a page for each lesson.
Follow the method in sequence, or jump to the decision you need. Each lesson stands alone and ends with a field exercise; the full field guide carries the work into complete cases, tools, production operations, and handover.
The course is ungated. The complete artifacts, extended case route, and field reference remain in the book.
Inside the book
Fourteen chapters. Five parts. One handover question.
Forward-Deployed Engineering follows an ambitious enterprise AI request from discovery and economics through context, agents, evaluation, security, integration, operations, and the client-owned exit.
PART I / THE WORK
See the real job
What an FDE does, how to draw the value path, and why reconnaissance starts before solutioning.
PART II / SHAPING THE MISSION
Bound the mission
The Deployment Contract, the CDEF cycle, and context that survives contact with work.
PART III / ENGINEERING TRUST
Earn wider action
Human–AI work units, evaluation, safety, security, and the move from system to workflow.
PART IV / LEARNING IN PRODUCTION
Operate and transfer
Operations, capability atrophy, handoff, and the evidence required for a client-owned exit.
PART V / SCALING PRACTICE
Scale with judgment
Field-to-product economics, team design, careers, the market, and the FDE Code.
Choose your route through the field guide.
Start with the problem in front of you, then return to the full route when the deployment earns the next decision.
Shipping a systemFollow Chapters 1–12 with the Meridian files at their decision points.
New to FDEStart with Chapters 1–5, then use Chapters 9–11 as an orientation route.
Architects and managersRead the value path, Deployment Contract, context, and operating chapters together.
Executives and budget ownersStart with the value path, business lens, Meridian files, and field-to-product economics.
The field guide
Make the handover part of the engineering.
Use the book as a delivery guide, a leadership review route, or a shared language in the client room. It pairs a field method with artifacts you can take into the next deployment.
01
Define the workValue-path canvas and reconnaissance maps
02
Set the boundaryDeployment Contract and authority matrix
03
Test the pathEvaluation plan and technical decision labs
04
Transfer ownershipOperations checklist and handover rehearsal
The model is one component. The deployment is the system.
Forward-Deployed Engineering
Evidence and boundaries
Practical, vendor-neutral, honest about proof.
Authored by Max Zhang and CTOL Digital Solutions, this book is a field guide for the decisions that sit between an AI prototype and a capability people can operate, inspect, correct, measure, and retire.
It draws on first-party practice context, public sources, and clearly labeled fictional or composite material. The methods are practical syntheses—not universal laws or independent customer-outcome claims.
Public exampleWhat a named source supports.
CompositeA recurring pattern, not a named customer.
HypotheticalA decision exercise to test judgment.
Fictional simulationA teaching device, not client evidence.
Field notes
Questions before the handover.
What is forward-deployed engineering for GenAI?
It connects a business outcome to a real workflow, approved context, tools, authority, evidence, operations, and client ownership. The FDE carries an ambiguous request through a bounded deployment and client-owned exit.
Who is this book for?
The primary readers are applied AI engineers and technical deployment leads accountable for putting generative AI into enterprise work. Architects, managers, product leaders, executives, founders, and career entrants have shorter routes through the same material.
Does the book advocate autonomous AI agents?
No. It separates agent capability from agent authority and recommends choosing the least autonomous shape that fits the work. Wider authority is earned through evidence about the whole work unit.
What is the empty-chair test?
It asks whether the relevant client roles can operate, inspect, correct, evaluate, pause, and retire the capability when the external builder is out of the room.
What is a minimum viable deployment?
An MVD is the smallest bounded live release that can change the next investment or operating decision. It is not proof that the whole future business case is true.
What can a reader use immediately?
Start with the value-path canvas, five-map reconnaissance, Deployment Contract, authority matrix, evaluation plan, deployment checklist, or client-owned handover rehearsal.
Is the mini-course the full book?
No. The mini-course is a short, ungated orientation to the method. The full book carries the work through Meridian decision files, field tools, playbooks, engineering blueprints, technical decision labs, economics, and client-owned handover.
Start with the work
Turn the next AI demo into a decision you can defend.
Start with the work, make the authority visible, and leave a path somebody else can operate.
