Forward-Deployed Engineering03
Two or three engineers,in your repository,a quarter at a time.
A pod joins your standups, your review queue and your pipeline, and ships workflows to production against goals written at the start of the quarter. Renewal is a decision made on what shipped.
- week 01 · pod in the repository, quarter goals signed
- week 02 · first pull requests merged from all three
- week 05 · first workflow cut over to real traffic
- Term
- One quarter at a time
- Team
- Two to three engineers, one lead
- Cadence
- Your standups, your reviewers
- Ends with
- A cadence your team keeps
FDE / 01The problem
A backlog is not a cadence.
Most AI backlogs are long because nothing on them reaches production. A pod does not add hands to the backlog; it changes what the team ships each month.
A backlog that grows
- Twelve AI items on the board; none has been in production
- The one engineer who understands the models is also on call for everything else
- Each initiative starts a new notebook and a new sample export
- Review waits for a person who is in another meeting
- The quarter ends with a demonstration and a longer backlog
A pod in the repository
- Two workflows named for the quarter, each with a metric and an owner
- Pull requests from the pod in your review queue every week
- One evaluation harness, reused by every workflow that follows
- Your engineers pair on each workflow and can change it without the pod
- The quarter ends with what shipped, written down, and a renewal decision
FDE / 02How a pod works
Four things a pod holds to.
A pod is the one engagement that runs more than one workflow. These rules keep it from becoming staff augmentation.
The quarter has goals, written before it starts
Two or three workflows, each with a metric and an owner, signed before week one. The pod is judged on those, not on velocity.
The lead is accountable
One named engineer leads the pod, reports to the owner on your side weekly in writing, and is the person you hold to the quarter's goals.
Your engineers pair on every workflow
Nothing is built that only the pod understands. Each workflow has an engineer on your side who has reviewed it, released it and rolled it back.
Renewal is a decision, not a default
In the last week of the quarter both sides read what shipped and decide whether to run another. There is no notice period to work through and no automatic rollover.
FDE / 03The quarter
M1 – M3
Three months, drawn as a plan.
One quarter, in months. The diamonds are the cut-overs and the review.
The goals for a quarter are fixed at its start. What is discovered during it goes on the list for the next.
FDE / 04What exists at the end
6 ITEMS
The end of a quarter.
Six things, ticked by your side, before renewal is discussed.
- The workflows in production
Each one named at the start of the quarter, running on real traffic under your account.
- The repository
Every commit in your repository, reviewed by your reviewers, intellectual property assigned before week one.
- The evaluation harness
One harness, in your pipeline, with a golden set for each workflow and a regression run on every release.
- The runbooks
One per workflow, executed by your engineers, not the pod's.
- The quarter report
What shipped against the goals, what did not and why, and what the pod would run next.
- The renewal decision
Another quarter, a handover or a stop, decided in writing in the last week.
FDE / 05Specified
10 FIELDS
The pod, specified.
The terms as they stand before the first call. The goals move with the quarter; the shape does not.
- Term
- One quarter, thirteen weeks, renewed by decision
- Team
- Two to three senior engineers: a named lead, a platform engineer and, where the work needs one, a data engineer
- Start
- A scoping call, then the quarter's goals agreed with the owner of the outcome
- Goals
- Two or three workflows per quarter, each with a metric and an owner, signed before week one
- Cadence
- Your standups daily; a written report from the lead weekly; a demonstration on real records at each cut-over
- Where
- In your repository, your pipeline and your review queue; on site for the first week and where the work needs it
- Access needed
- Repository, the systems of record, real records, and a reviewer who can approve a pull request
- Your engineers
- At least one paired on each workflow, with time to review and release it
- Fee
- Fixed per quarter, agreed before week one
- Ending
- When a quarter is not renewed, its last two weeks are the Handover & Enablement
FDE / 06Proof
3 ROWS
Case studies
Illustrative engagements written end to end, chapter by chapter.
- 03RetailSupplier catalogue onboarding and enrichmentSubmission to live listing 16 → 5 days; no legal field ever written by a model
- 06ManufacturingMaintenance: alarms and notes to planned workAlarm to approved work order 19h → 3.1h; safety text never paraphrased
- 11LogisticsShipment exception handling at a 3PLTime to customer update 4.1h → 0.9h; claims filed in window 71% → 96%
FDE / 07Questions
6 QUESTIONS
Asked about pods.
How is a pod different from staff augmentation?
A pod is judged on workflows in production, not hours. Its goals are written before the quarter starts, its lead reports on them weekly, and renewal depends on what shipped. Staff augmentation gives you hands; a pod gives you an outcome per quarter.
Who is in the pod?
A named lead who has taken systems into production, a platform engineer, and a data engineer where the workflows need one. You meet each of them before you sign, and the same people stay for the quarter.
Can the goals change mid-quarter?
The workflows named at the start do not change. What is learned during the quarter is written down for the next one. If a goal turns out to be impossible, the lead says so in the weekly report and the owner decides what replaces it.
How many workflows does a pod ship in a quarter?
Two or three, depending on the systems each has to connect to. That is agreed at scoping, in writing, before the quarter begins.
How does this compare with the Production Sprint?
The Production Sprint takes one workflow to production in six weeks with one engineer. A pod runs several, in sequence, with a team, and builds the harness and the habits your engineers keep. Many organisations run a sprint first and a pod after.
What happens when we stop?
The last two weeks of the final quarter are the Handover & Enablement: runbooks, decision records and a dry run by your engineers. The pod holds no standing access after it leaves.
FDE / ENDStart
One quarter,and a team that ships each month.
Book a scoping call; you leave it with a view on what a first quarter would contain.