This post is for CTOs and technology buyers who are being offered forward deployed engineering (FDE: engineers who write production code inside your team and environment) by an AI lab, a hyperscaler or a platform vendor, and who want to compare it with an independent firm. We're an independent firm, so we have a stake in the answer. We'll try to be fair about where a vendor is the better choice.
The useful questions are whose interests are in the room and what you can still change in two years.
What is the difference between a vendor FDE and an independent FDE?
A vendor FDE is employed by a platform company and helps you succeed with that company's product. An independent FDE works for a firm with no product to sell and is accountable to you for the system you asked for. Both write code, but their incentives and portability differ.
Both can be excellent. The differences come from the structure of the job. A vendor engineer who tells you a competing tool fits better is acting against their employer's interest. An independent engineer who recommends a vendor product they have no stake in is easier to believe. That's incentive design, and it says nothing about the people involved.
| Question | Vendor FDE | Independent FDE |
|---|---|---|
| Who employs them | The platform, model or cloud vendor | An independent engineering firm |
| Main goal | Your successful adoption of the vendor product | A working system you own, built to your requirements |
| Strongest at | Deep platform knowledge, access to the product team, early features | Multi-vendor integration, portability, honest tool selection |
| Typical limit | Reluctance to recommend outside the platform | Less access to vendor roadmaps and internals |
| How it ends | Often tied to the platform commitment | Planned handoff to your team |
Why do vendors invest so heavily in FDEs?
Vendors fund FDEs because getting customers from pilot to production drives product usage and renewals. Forrester counts about $9B committed to FDE-style units by four providers between 4 May and 15 July 2026. Our own reading, separate from Forrester's, is that these units exist first to sell and expand the vendor's own products.
The figures, from Forrester: OpenAI $4B or more, Microsoft $2.5B, AWS $1B, and Anthropic's services venture about $1.5B (Forrester's own figure). Forrester also argues that FDEs are "training wheels," temporary by design, and that services firms should sell outcomes, build for handoff and let their footprint shrink. We unpack the numbers in the $9 billion FDE bet.
That commercial logic isn't sinister. It's how platform businesses work, and it explains why vendor engineers are often generous with time and expertise. It also tells you what they're optimizing for, so keep it in mind when you read a recommendation.
What does lock-in look like with a vendor FDE?
Lock-in shows up as code, data and workflows that depend on one platform's proprietary services, so moving later means rewriting. Analysts quoted by CIO.com warn that hyperscaler FDEs raise this risk, even if each design decision looked sensible at the time.
CIO.com reports Microsoft and AWS putting billions into thousands of FDEs, and quotes analysts at Info-Tech and Carmi Levy cautioning about vendor lock-in. The reporting doesn't say any vendor acts in bad faith. The warning is about structure.
In practice, lock-in often arrives through small choices:
- Workflow logic written in a vendor-specific orchestration layer instead of plain code.
- Prompts, evaluations and routing tied to one model provider's interfaces, with no abstraction in between.
- Data stored in a proprietary format or in an account you don't control.
- Knowledge that stays in the vendor engineer's head, with no documentation or runbooks.
None of that is a reason to avoid a vendor. It is a reason to ask where the system runs, who owns the code and what a migration would take. Our own builds go the other way: deterministic software in your own cloud, with schema-validated model outputs behind a thin interface, so swapping a model or vendor is an engineering task instead of a rewrite.
How does Anthropic's applied AI engineer model fit in?
Anthropic's model pairs vendor engineers with outside firms. In March 2026 it said it was growing its partner-facing team fivefold to provide dedicated applied AI engineers on live partner deals. That makes the vendor engineer a supporting expert, with a partner firm owning delivery.
The announcement is on Anthropic's Claude Partner Network page (12 March 2026), which also says membership is free and open to any firm. For a buyer the interesting part is the structure. The vendor supplies product depth, and the partner firm is positioned to own delivery.
That points to the mixed approach many buyers will land on. It also raises two questions: who is accountable for the production result, and who owns the code. Settle both in writing, whoever supplies the engineers.
When is a vendor's own engineering team the right call?
Use a vendor's engineers when the problem sits inside their product and needs knowledge that only they have. The clearest cases are a new product with thin documentation and deep platform issues that outside engineers cannot see or fix. Outside those cases, ask what the vendor gains from each recommendation.
- If a feature is only weeks old, the vendor's engineers are the experts, and they may be able to change the product for you.
- Quotas, performance limits, undocumented behavior and security reviews often need the vendor's internals.
- If you've chosen one vendor deliberately, the dependency is a decision and won't surprise you later.
- Some vendors will fund engineering time for strategic customers. Take it, with your eyes open about the commitment attached.
When is an independent FDE team the right call?
Use an independent team when your system spans several vendors, when you want to avoid lock-in, or when you need a clear exit. The engineers owe you advice with no product to sell and a system that works without them.
Real workflows cross ERP, CRM, payments, freight and several AI providers, and no single vendor's engineers own that whole picture. If you want models and clouds to be swappable parts, an outside team has no reason to resist it. Regulators, boards or auditors may also want you to show how you could leave a supplier, which means you own the code, database and IP, in your own cloud account or private VPC. And if you want someone to say "a standard SaaS product is enough here" or "don't build this," that advice comes more easily from a team with nothing to sell.
Our FansChance work is an example of a system that needed to work across hardware, payments and ticketing under heavy load. The result ran in the client's own environment, with the client owning the code. For how our engagements end, see handoff by design.
What 6 questions should you ask either kind of FDE?
Ask the same six questions of a vendor or an independent team: who is accountable, who owns the code, where it runs, what a migration would take, how it ends, and how success is measured. The answers tell you more than the brand.
- Who is accountable for the production result, and what happens if it misses the agreed metric?
- Who owns the code, database and IP, and where is it stored from day one?
- Where will it run: in my cloud account, or on your platform?
- What would a migration to another vendor involve, and which parts are portable by design?
- How does the engagement end, and who on my team will hold the knowledge?
- What's the baseline metric, and who measures it before and after?
A vendor that answers these well is a strong partner, and an independent firm that can't answer them isn't one. Our fuller buyer checklist is in how to scope, buy and evaluate FDE services, and the wider comparison of models is in FDE vs consulting vs staff augmentation.
If you're weighing a vendor offer against an independent team, send us the proposal and we'll tell you what we'd ask before you sign.
