This post is for technology and operations leaders who have to choose how to get an AI or software initiative into production, and are weighing four options that all promise "engineers on your problem": a forward deployed engineering (FDE) partner, a management consultancy, a staff augmentation firm, or a vendor's own FDEs.
They look similar in a proposal and behave very differently in month three. The differences come down to who writes the code, who is accountable when it breaks, and how the engagement ends. If you need the definition of FDE first, read what forward deployed engineering is.
What are the four models?
An FDE partner embeds senior engineers who build and hand off a system you own. A consultancy advises and designs. Staff augmentation supplies engineers who work under your direction. A vendor FDE is an engineer employed by a platform company to help you succeed with its product.
- A forward deployed engineering partner is an independent firm that places senior engineers in your team, writes production code in your environment and is accountable for a defined outcome. Kaprin works this way.
- A management consultancy handles strategy, operating model, vendor selection, governance and change management. Implementation may be included, subcontracted or left to you.
- With staff augmentation, contractors or a managed team are added to your headcount. You set priorities and review the work.
- Vendor-provided FDEs come from a platform, model or cloud provider. CIO.com reports Microsoft and AWS putting billions of dollars into thousands of FDEs.
The boundaries blur. Consultancies sell FDE pods (Deloitte is one), and some staffing firms have renamed contractors as FDEs. Judge a provider by what it does, using the table below, and ignore the label.
How do the four models compare?
They differ most on who writes code, who is accountable, who owns the result and how the work ends. This table generalizes typical patterns. Individual providers vary, so test any specific proposal against each row. Read down a column to see one model's profile, or across a row to compare all four on a single question.
| Dimension | Forward deployed engineering partner | Management consultancy | Staff augmentation | Vendor-provided FDE |
|---|---|---|---|---|
| Who writes code | Senior engineers from the partner, as the main output | Often not the consultants; implementation may be subcontracted or left to you | Contractors, following your instructions | The vendor's engineers, usually on the vendor's product or platform |
| Where the work happens | Inside your team, repositories and cloud account | Mixed: workshops, documents, sometimes on site | Inside your team | Your environment and the vendor's platform |
| What you are accountable for / who is accountable | Partner is accountable for a working system and an agreed metric; you provide access and an owner | Consultancy is accountable for advice and deliverables; you are accountable for the production result | You are accountable for the result; the provider is accountable for supplying people | Vendor is accountable for platform adoption; you are accountable for the wider system |
| Time to production | Weeks. Kaprin typically deploys within 30-45 days; Deloitte describes about 3 months for its FDE pods | Often months, because analysis and design come first | Depends on your management and backlog; no inherent time-box | Depends on the vendor's priorities and your fit with the product roadmap |
| Who owns the result | You, if the contract says so: code, database and IP in your own cloud account | You own the deliverables; any built system depends on who built it and where it runs | You, usually, as it is built in your repositories | Often runs on the vendor platform, which shapes what you can move later |
| How it ends | Planned handoff to your team, with documentation and runbooks | Final report, or a follow-on phase | Contract ends and knowledge leaves with the contractors | Often continues for as long as the platform commitment does |
| Typical failure mode | Scope too broad or no owner on your side, so the pod builds the wrong thing quickly | Strategy that never reaches production | Busy engineers, no outcome; the system depends on people who have left | Solutions shaped around one vendor, with lock-in risk (the concern analysts raise in the CIO.com reporting) |
Accountability is the row that decides most outcomes. If nobody is accountable for the production result, the project can fail while every party meets its contract.
What is the typical failure mode of each model?
Each model fails in a predictable way: FDE partners when scope or ownership is vague, consultancies when strategy never ships, staff augmentation when nobody owns the outcome, and vendor FDEs when the best answer lies outside the vendor's product. Knowing the failure mode in advance lets you write a safeguard into the contract.
FDE isn't a cure for everything, and it helps to be honest about that. An embedded pod with no business owner and an unclear problem will build the wrong thing efficiently. This is part of what Gartner means when it predicts that over 40% of agentic AI projects will be canceled by the end of 2027, citing cost, unclear value or weak risk controls. No engagement model fixes an unclear problem.
It works the other way too. A consultancy can be exactly right when the real question is which processes to automate, who should own them and how to manage the change. A well-run FDE engagement still depends on those decisions having been made.
How can you tell which model a proposal really is?
Ask three questions of any proposal, whatever it calls itself: who writes the code, who is accountable if it fails in production, and what you still own when the contract ends. The answers place it in one of the four models quickly.
- Who writes the code? If the answer is "your team, with our guidance," you're looking at consulting or staff augmentation, whatever the title says.
- Who is accountable for the production result? If the answer is "you," the provider is selling capacity and the outcome stays with you.
- What do you own at the end? Code, database and IP in your own accounts points to an independent model. Dependence on a platform points to a vendor model.
A proposal that blurs these answers is usually blending models, and not always in your favor.
Which model should you pick, and when?
Choose by the main thing you lack. Lack a built, owned system: FDE partner. Lack direction: consultancy. Lack hands under your own management: staff augmentation. Lack depth on one vendor platform: that vendor's FDEs. Most buyers lack more than one of these, so start with whichever gap blocks production first.
Choose a forward deployed engineering partner when:
- You know the problem and need a production system in weeks, built in your cloud.
- A pilot or vibe-coded prototype is stuck and needs production hardening.
- You want to own the code and have your team inherit it. Our Slender Morris case is an example: purchase orders flow from PDF into the ERP and freight systems, with about 90% faster order entry.
Choose a management consultancy when the question is strategy, operating model, vendor selection or organizational change. It also fits when you need board-level governance or a risk framework before you need a system, or when many people have to agree before anyone can build.
Choose staff augmentation when:
- You have a strong engineering manager and a clear backlog, and all you lack is capacity.
- The work is long-running, you want to control it day to day, and you accept accountability for the result and for retaining knowledge.
Choose a vendor-provided FDE when the problem sits inside one vendor's product and needs deep platform knowledge. That includes early adoption of a new product where only the vendor's engineers know the internals. You should be comfortable with the platform dependency. For the trade-offs, see vendor FDEs vs independent FDEs.
Can you combine the models?
Yes. A common pattern is a consultancy for direction, an FDE partner for the build and handoff, and a vendor FDE for hard platform questions. The rule is to name one party accountable for the production result, so the others have clear scopes.
When you combine them, put the baseline metric in one contract, require code and infrastructure to live in your accounts, and agree the handoff plan before work starts. Our buyer checklist lists the questions to ask each provider.
If you're not sure which of the four fits, tell us what you are trying to ship. If the answer is a consultancy or a vendor, we'll say so.
