Blog · Technology Decisions
Forward Deployed Engineer or Consultancy? Five Ways to Get AI Built
How to choose between hiring a forward deployed engineer, using your vendor's, buying agency time, working with a consultancy that builds or using your IT team, and what each leaves you owning.
- The Foregrounds
- 15 min read

Salesforce said in November 2025 that it had committed to building a team of 1,000 forward deployed engineers (FDEs). The same post, citing an analysis by Indeed and The Financial Times, said job postings for the role rose by more than 800% between January and September of 2025. The OpenAI Deployment Company, launched in May 2026, said in a release on founding partner Advent's site that it would start with "approximately 150 experienced Forward Deployed Engineers and Deployment Specialists." Forward deployed engineering means that an engineer works alongside the customer's team and builds a solution around how that team works.
Those numbers describe what vendors have chosen to do for their own products and for the customers they select. If your team sells its expertise, as a consultancy, an agency, or an accounting or advisory practice does, we think a full-time FDE is rarely the answer, because a hire needs enough work beyond the first system. The decision on how to resource your AI and agentic work should be based on who learns the work and who owns the system, so below we compare five ways to get AI built, the four tests that decide among them and the questions to ask any provider.
Why the vendor numbers do not settle the question for your team
A vendor's FDE exists to get that vendor's product into use, and access can be selective, so a vendor's hiring tells you little about what your team should do. Anthropic's job posting describes an FDE "who embeds directly with our most strategic customers" and will "contribute insights back to our Product and Engineering teams." Salesforce's blog said in November 2025 that customers "must be nominated by their account executive" for a program "offered to select customers for free," with paid services expected later.
At least one vendor is also moving toward smaller customers: Anthropic's May 2026 services company "will work with mid-sized companies across sectors," although that page does not use the term FDE. So access may widen, but a vendor's engineer is still there to get that vendor's product working.
Forward deployed engineer vs consultant: five options and what each leaves you owning
For most teams like yours, we think the likeliest fit is a consultancy that builds, or the vendor's FDE when all the work sits in one product, and hiring comes later if at all. We sell the fourth option, so weigh our view against the tests below.

- Hire an FDE, or an engineer who works that way. A hire can keep the knowledge on your team and give you someone to handle later changes, but you are left with a job to fill and manage, one person holding the knowledge, and the task of finding that person enough work. The US Bureau of Labor Statistics puts the median wage for software developers at $135,980 in May 2025, and wages were 70.0% of employer compensation costs across private industry in June 2026. Dividing $135,980 by 0.70 gives about $194,000 a year before recruiting and tools. That is our rough illustration using a broad compensation ratio, not a measured cost of hiring an FDE. We think this is right when the work is steady but rarely right as the first step.
- Use the vendor's FDEs. They bring product knowledge and a direct line to the people who build it, but you first have to find out whether you qualify, and unless the proposal says otherwise, everything outside that product stays with you. In its November 2025 blog, Salesforce described pods, the small groups some of its FDEs work in, that focus on one client "for about three months," or as long as deployment takes, on "one or two use cases." This is the best option when you qualify and the work fits.
- Buy FDE time from an agency. You buy engineering time without making a hire, but you are left directing that time, and the knowledge can go when the hours stop unless someone on your team learns alongside the engineer. If you buy by the month or use a part-time engineer, that responsibility still sits with you. We would choose this option when you already know what to build and have an owner.
- Work with a consultancy that builds. One team decides what to build and builds it, with several skills in one engagement, but you still have to make your people available during the build and agree what the handoff gives you, which depends on the proposal. A proposal for advice need not include a build, so check who writes the code. It fits when the work crosses tools and you have no engineer.
- Have the in-house IT team do it. They know your systems, access rules and security, and can keep responsibility for the system inside your business, but you are left with all of it, including capacity taken from their other work and any AI skills they lack. Stanford's April 2026 The Enterprise AI Playbook, which covers 51 successful cases at 41 organizations and relies mainly on self-reported data, says first attempts failed "when led by technical teams without business ownership." The section on in-house work below says when it fits.
What the research says about outside builders, and where it stops
The one comparison we found favors outside help, although it does not show that hiring a consultancy doubles your odds. MIT NANDA's July 2025 report, linked as a hosted copy and labeled preliminary, says "external partnerships with learning-capable, customized tools reached deployment ~67% of the time, compared to ~33% for internally built tools." Those percentages come from interviews at 52 organizations and describe self-reported outcomes, but the report does not give exact initiative counts behind them. Its partnership category includes buying external tools and co-developing with vendors, so it does not separately measure consultancies or FDEs. The authors say the gap "may reflect organizational capabilities rather than implementation approach alone" and "does not necessarily prove causation." We read this as support for going outside as a reasonable default, but you still need to apply the four tests below.
Four things decide which option fits: learning, data, support and ownership
Who learns the work? Whoever learns your exceptions holds much of the value, so someone on your team needs to learn them alongside the builder. Stanford's logistics case had "Two junior IT staff embedded from day one," and the report says "the company could operate the system independently." That company is listed at "$1B+," well above a small team, but the point carries, because the test for your business is whether one of your people works alongside the builder from the start.
Who builds on your data? Your team holds client material under promises it has made to those clients, and one buyer told MIT's researchers, "I can't risk client data mixing with someone else's model, even if the vendor says it's fine." So ask whoever builds it whose account and environment the system runs in, and who can see the client documents.
Who can your team call after launch? The work keeps changing after launch, so someone has to change the system with it. Another MIT buyer said, "Our process evolves every quarter. If the AI can't adapt, we're back to spreadsheets." An employee can handle ongoing changes, a contract may end, and a vendor's pod is planned around a deployment, although Salesforce's partner announcement does say "FDEs stay engaged after launch." It does not say for how long, so ask.
Who owns the system? Ownership means one named person who answers for the system, plus rights and access to what was built. In Stanford's mid-market translation-services case, one change on the second attempt was that "the CEO took ownership rather than delegating to the CTO." The rights matter because, as a chief information officer told MIT's researchers, "Once we've invested time in training a system to understand our workflows, the switching costs become prohibitive." So check what you can carry forward when an engagement ends.
When a vendor's forward deployed engineer is enough, and when it is not
A vendor's engineer can be enough when the data and the steps all sit in one product on its own platform, the vendor will give you access and you are content to keep using it. Salesforce described a similarly bounded scope in November 2025, "one or two use cases." We think it falls short when the work crosses other tools, shared drives and past client documents, unless the proposal names who builds those connections, because knowing the product does not make the engineer responsible for the rest.

Three hypothetical teams and the option that fits each
The same tests point to different options as the work changes.
- A wealth-advisory practice whose client records, tasks and billing all sit in one vendor's platform. It wants drafts of annual-review summaries from those records, and everything is in one product, so the fit is the vendor's FDE, or a partner the vendor certifies, if the practice can get access. It still names an owner.
- A research and communications agency whose client briefs draw on a project tool, a shared drive, past reports and a billing system. Nobody there writes code, so the team needs a provider whose proposal covers the whole process. The fit is a consultancy that builds, or agency FDE time if a named person can direct it, and the research agency names one owner to learn alongside the builder.
- An accounting practice with a systems engineer who has free time, and a year-end document checklist that barely changes. The rules are stable and the data sits in systems the engineer runs, so the fit is in-house, with a partner named as the business owner.
What to ask any provider of forward deployed engineering services
The label does not tell you how a provider works, because it can describe an employee, a vendor team or outside engineering time. So ask anyone who says "forward deployed" the same five questions, and get the answers in the proposal.
- Who sits with our people? Ask for names or roles, and how much of their time.
- Who writes the code? It should be the person in those conversations, or you should know where the hand-off happens.
- What do we own at handoff? Ask about code, accounts, configuration, test cases, written instructions and anything that stays licensed.
- How is our team trained? Ask on which work, and who the owner is by the end.
- Who do we call a month after launch, and on what terms?
The Foregrounds is one option that works this way. The engineers who sit in the interviews with your people write the prototype, with no hand-off between the strategy and the code. We build on your own data, in your environment, and we train your team as part of every build. At handoff, you have a named owner, review points and a runbook, meaning written instructions for running the system. Ask us the fifth question in a proposal, as you would anyone else.
When your in-house team is the right answer
In-house can be the best answer on this list when your team has an engineer with time and the work is stable, so nobody needs an outside builder to keep up with change. It needs a business owner beside the engineer, and it should leave room for a false start, since Stanford reports that "61% of successful projects included at least one prior failure," in a sample of successful cases only. If you have that engineer and that owner, a short outside review of the plan may be all the outside help you need.
What we recommend you do next
- Decide who will own the system before you decide who builds it. Name one person on your team this week and give them time during the build, because if you cannot name anyone, every option gets weaker. In that case it may be better to first decide whether to build anything at all.
- Trace one piece of work through every tool it touches before you call a vendor. If it stays in one product, ask that vendor about access first, and if it crosses tools, ask consultancies that build and agencies for proposals.
- Hire only when you can list a year of work for the person. Write down what the engineer would handle over the next twelve months, and compare that list with about $194,000 a year, our rough US illustration using BLS wage data and private-industry compensation data, rather than a quote for your hire. If the list runs out after the first system, pay an outside team for the build and make sure your people end up running it.
Questions buyers ask
What is the difference between a forward deployed engineer and a consultant?
The main difference is who writes the code, because a consultant's traditional job ends in advice and a plan, while an FDE builds the system inside the customer's work. The line is not fixed, since some consultancies build and some FDEs spend much of their time working out what to build. So ask whether the person advising you will also write the code.
Which companies offer forward deployed engineering services for small and midsized businesses?
It depends on the kind of provider, and we do not rank them here. AI vendors and their certified partners suit work in one product, with access that varies, while specialist agencies sell embedded engineers, consultancies that build may not use the term, and recruiters place an engineer you then employ. To tell whether one serves a business your size, ask for its smallest current type of client, its minimum engagement and whether the engineer's time is dedicated to you.
Can we start with a consultancy and hire an engineer later?
Yes, if you plan for it at the start. The risk is that the system gets built in a way nobody you later hire can pick up, so write into the proposal that your people have access from the first day and that the code, accounts and instructions come to you. If a hire is likely, involve that person or your IT contact during the build.
What is at stake is who can run and change the system once the outside help has gone. See how we build AI systems with your team.
Keep reading
More from the practice
AI Agent Governance for a Small Team: Six Decisions to Write Down
How a small team can govern AI agents with six written decisions, where an AI usage policy stops, and when frameworks such as NIST AI RMF and ISO/IEC 42001 are worth the time.
Read moreForward Deployed Engineer: What Is Forward Deployed Engineering?
What is forward deployed engineering, what does a forward deployed engineer do, and what should your team look for when you get AI built?
Read moreHow to Measure AI Automation ROI After Launch, Beyond Hours Saved
How to judge an AI automation after launch by the business number it moved, not by the hours its dashboard says it saved.
Read moreStart withthe businessproblem.
Thirty minutes. Tell us what's slow or manual today, and we'll tell you honestly whether we can fix it.
30 minutes