Top forward deployed engineering firms in the US / Cost and method
Questions to ask an FDE firm before you sign
This checklist is for the first meeting with any forward deployed engineering firm. It has nine parts: the firm and its owners, the engineers, their time with your people, access to your systems, testing, platforms and models, price, handover, and partner claims. Each question comes with what a clear answer contains and a fact from this guide that shows why the question matters. Print it, send it to every firm before the meeting, and compare the written answers side by side.
Published October 21, 2026. Editorial.
Key takeaways
- Send every firm the same questions before the meeting and ask for written answers, so that you can compare them.
- Ask who owns the firm, whether the engineers are employees or contractors, and where they will work on each day of the week.
- Ask what has to pass before each release, what part of the system would run on the firm's own software, and how the work moves to your own staff at the end.
- Check partner claims yourself: Anthropic's directory is public, and AWS's partner FDE programme had named no firms by 20 October 2026.
Print this page and take it to the first meeting with any forward deployed engineering firm. Send the questions to every firm on your shortlist before the meeting, in the same order, and ask for written answers. That includes Reveneau, Fluxon and every other firm in the list of US FDE firms. Written answers placed side by side are easier to compare than notes from a call.
Each question below comes with what a clear answer contains and a fact from this guide that shows why it matters. For the general case of any embedded engineering vendor, the FDE guide also has questions to ask a forward deployed engineering vendor.
How to use the checklist
- Send the questions a week before the meeting.
- In the meeting, ask for evidence where it exists: a contract clause, a test report, a partner letter, a name.
- Mark each question the firm could not answer, or answered only in general terms.
- Compare the firms after the last meeting, then pay the two clearest for the same short task on your own data.
Part 1: The firm and its owners
| Question | What a clear answer contains | Why ask |
|---|---|---|
| Who owns your firm? | Owners and main investors, by name | TechCrunch reports that Anthropic co-owns Ode with Anthropic [11] |
| Does any owner sell AI models, cloud services or software you might recommend? | A yes or no, with names | A model maker's stake can shape which models a firm suggests |
| Does any owner also own, or plan to buy, our company or a competitor? | A yes or no | InvestmentNews reports that General Catalyst owns Percepta and co-led the deal to buy its client Janus Henderson [9] |
| If your firm is sold, can we end the contract? | A clause, with notice in days | Ode bought Fractional AI, and AlixPartners bought Artium, both in 2026 |
Part 2: The engineers
| Question | What a clear answer contains | Why ask |
|---|---|---|
| Who exactly will work on our project? | Names, roles and years in the role | The people in the sales meeting may not be the people who do the work |
| Are they your employees or contractors? | A count of each | TechCrunch reported in 2024 that Tribe AI builds with a network of over 500 contractors [5] |
| Where do they live, and in which time zone? | Cities and working hours | 10Pearls's about page lists offices in Costa Rica, Colombia, the UK, Pakistan and Peru [15], and its FDE page says it uses those offices to overlap with your working hours [14]; Caylent says its staff work remotely [16] |
| Are they hired already? | A yes, or a hiring date | Tredence's FDE page gives a goal of 200 FDEs over the next 12 months [10] |
| How do you replace someone who leaves? | A notice period and a process | A change of people slows the work |
Part 3: Their time with your people
| Question | What a clear answer contains | Why ask |
|---|---|---|
| How many days a week will the engineers be at our site? | A number, written into the contract | Forward deployed work depends on time with users |
| How will their time split between talking with users and building? | A share for each | First Round Review reports that at Serval, FDEs are expected to spend "around 20%" of their time with customers [4] |
| Who pays for travel and hotels? | A line in the quote | Site visits add cost |
Part 4: Access to your systems
| Question | What a clear answer contains | Why ask |
|---|---|---|
| Which systems and data will the engineers need? | A list, by system | Access is the first thing your security team approves |
| Who on our side approves each access, and who removes it at the end? | Named roles | Access should end when the work ends |
| Will any AI tool your engineers use send our code or data outside our company? | Each tool, and where data goes | Your data rules apply to the firm's tools too |
The FDE guide's page on security and access for embedded engineers explains how to set this up.
Part 5: How the work is tested
| Question | What a clear answer contains | Why ask |
|---|---|---|
| What has to pass before a change reaches our users? | A named test process, and who runs it | Code can be written fast, and tests show whether it does what you asked |
| Who writes the tests, and from what? | The source, such as your specification | Tests written from your requirements check what you asked for |
| Can we see and keep the test results? | Yes, in a stated format | You need them to run and change the system later |
| How do you test AI parts that can give a different answer each time? | Repeated runs, and how they are scored | AI output can change from run to run |
One firm's answer to the first question, as an example of the detail to ask for: Reveneau says every change has to pass a large eval suite before release. The suite is a set of automated tests written from the specification. Reveneau says Jev grades the suite and makes it run ten times faster than its previous grader did. Reveneau also says AI writes 100 percent of its code, and that it takes responsibility for the whole project through production, meaning daily use by real users, and after release. Ask every firm for an answer with the same level of detail.
Part 6: Platforms and models
| Question | What a clear answer contains | Why ask |
|---|---|---|
| Will any part of our system run on your own software? | Each product, with its licence | Distyl AI says it deploys its own products [12]; Turing pairs its FDEs with its own control plane [13] |
| What will that software cost each year, and after you leave? | A price per year | The licence continues after the engineers leave |
| Which AI models would you use, and would you test another maker's model? | Model names, and a yes or no | A firm tied to one maker may favour its models |
| Who owns the code, prompts and settings you write for us? | Our company, in a clause | You need them to change or move the system |
Part 7: Price
| Question | What a clear answer contains | Why ask |
|---|---|---|
| Do you publish a rate card, meaning a list of your prices? | A yes or no | None of the fifteen firms in this guide publishes a price |
| How do you charge: by time, by team, by fixed fee, by outcome or by output? | One method, explained | Tenex says it charges by story points it defines [7]; Caylent prices on outcomes or a fixed fee where results can be measured [8] |
| If you charge by output or outcome, who measures it? | A written measure and a named person | The fee depends on the measure |
| What will running the system cost each month after you leave? | Model, hosting and licence costs, estimated | These costs stay with your company |
What forward deployed engineering costs from a US firm explains each pricing method.
Part 8: Handover and exit
| Question | What a clear answer contains | Why ask |
|---|---|---|
| When does the engagement end, and how do we know it is done? | A date and a written test of done | An end date gives both sides a point to check the work against |
| What will our own staff be able to do on the day after you leave? | A list of tasks they can do alone | Amazon says its own FDE model is designed so customers are self-sufficient when a deployment ends [3] |
| Will you train our people? | A plan, with sessions and dates | HatchWorks AI says its FDEs train the client's organization [6] |
| What does the handover cost? | A price and dates | A handover uses the engineers' time, which you pay for |
The FDE guide's page on handover and exit lists what a handover should contain.
Part 9: Partner claims
| Question | What a clear answer contains | Why ask |
|---|---|---|
| Are you in Anthropic's partner directory, and at which tier? | The name as listed, and the tier | You can check it yourself; on 20 October 2026 it held 321 entries [1] |
| If you say you are a partner but are not listed, what proof do you have? | A letter or a link from the platform | HatchWorks AI says it is a direct partner, and the directory did not list it on 20 October 2026 [1] [6] |
| Are you in AWS's partner FDE programme? | Written proof from AWS | AWS's 30 June 2026 post names no partner firms [2] |
| What does your certification mean, and who issued it? | The issuer and what it tested | HatchWorks AI and Artium both describe their FDEs as certified |
AWS says its partner programme will use engineers who "pass an AWS-defined technical bar before engaging with any customer" [2], meaning a technical test that AWS sets. If a firm says its engineers passed that test, ask for the record.
After the meetings
Lay the written answers side by side, one column per firm. Look first at the questions a firm could not answer, and at answers that stayed general after you asked for detail. Then give the two firms with the clearest answers the same short paid task on your own data, and compare the code, the tests, the written explanation and the questions each firm asked you.
To put these questions to Reveneau, contact Reveneau, and read about its forward deployed engineers. For firms that build AI software without placing engineers inside your team, see the guide to AI development companies in Silicon Valley.
Common questions
What should I ask an FDE firm before I sign a contract?
Ask about nine areas: who owns the firm; who the engineers are and whether they are employees; how many days they will be at your site; which systems they need; what has to pass before each release; whether your system will run on the firm's own software; how it charges; the handover; and any partner claims, with proof.
How do I find out whether an FDE firm's engineers are its employees?
Ask the firm for a count of employees and contractors on your project, by name and role, and ask for it in writing. TechCrunch reported in 2024 that Tribe AI builds client projects with a network of over 500 contractors. Contractors can do good work, but your security team needs to know who will have access to your systems.
How many days a week should a forward deployed engineer spend at my site?
This guide found no public standard. First Round Review reported that at Serval, a software company, FDEs are expected to spend around 20 percent of their time with customers. For your project, agree on a number of days at your site, write it into the contract, and ask who pays for travel and hotels when engineers come from another city.
What should I ask an FDE firm about testing before release?
Ask what has to pass before a change reaches your users, who writes those tests and from what source, whether you can keep the results, and how the firm tests AI parts that can give a different answer each time. Reveneau's answer, as one example, is an eval suite, a set of automated tests written from the specification, that every change must pass.
How do I check an FDE firm's Anthropic partner claim?
Search Anthropic's Claude Partner Network directory for the firm's name, and try other spellings. On 20 October 2026 the directory held 321 entries, most with a tier of Global Premier, Preferred or Select. If a firm says it is a partner and is not listed, as with HatchWorks AI on that date, ask for written proof from Anthropic.
How do I check a claim to be in AWS's partner FDE programme?
Ask for written proof from AWS. AWS announced on 30 June 2026 that it is starting a partner FDE programme with a limited set of partners, whose engineers must pass a technical test that AWS sets, but its post names no partner firms. No public list was found, so a firm's own statement is the only evidence until AWS confirms it.
What should I ask an FDE firm about who owns it?
Ask who owns the firm and its main investors, whether any owner sells models, cloud services or software the firm might recommend, whether an owner also owns or plans to buy your company or a competitor, and whether you can end the contract if the firm is sold. InvestmentNews reported one case where an owner sits on both sides.
What should the handover section of an FDE contract say?
It should give an end date and a written test for when the work is done, list what your own staff will be able to do alone the day after the firm leaves, set out training sessions with dates, and price the handover. Amazon says its own FDE model is designed so customers can run their systems by themselves when a deployment ends.
What should I ask about pricing by story points or by outcome?
Ask who sizes the work or measures the outcome, how you can challenge a count or a result, what one unit costs, and what happens when the scope changes. Tenex says it charges by story points that it defines. Caylent says it prices on outcomes or a fixed fee where results can be measured. Get the measure in writing.
What should I do after meeting several FDE firms?
Lay the written answers side by side, one column per firm, and mark the questions each could not answer or answered in general terms. Then pay the two firms with the clearest answers for the same short task on your own data, and compare the code, the tests, the written explanation and the questions each firm asked you.
References
- [1] Anthropic, Claude Partner Network directory (read 20 October 2026): 321 entries; HatchWorks AI not listed.
- [2] AWS Partner Network blog, Introducing Forward Deployed Engineering for Partners (30 June 2026): a limited set of partners; an AWS-defined technical bar; no partner firms named. Amazon's own statement.
- [3] Amazon, aboutamazon.com, AWS forward deployed AI engineers (30 June 2026): designed so customers are self-sufficient when a deployment ends. Amazon's own statement.
- [4] First Round Review, So you want to hire a forward deployed engineer (24 February 2026): Serval's FDEs spend around 20% of their time with customers.
- [5] TechCrunch, on Tribe AI's seed round (17 July 2024): a network of over 500 contractors.
- [6] HatchWorks AI, Anthropic page (read 20 October 2026): Anthropic-certified Forward Deployed Engineers; training for the client's organization; states it is a direct partner.
- [7] Tenex, AI engineering (read 20 October 2026): charges based on what it produces, measured by story points it defines.
- [8] Caylent, press release launching its Anthropic practice (28 April 2026): outcome-based pricing and fixed-fee terms where measurement is feasible.
- [9] InvestmentNews, Janus Henderson taps Anthropic for all-in bet on AI (11 June 2026): work with Percepta; take-private led by Trian Fund Management and General Catalyst.
- [10] Tredence, Forward deployed engineering (read 20 October 2026): committed to building a pool of 200 FDEs over the next 12 months.
- [11] TechCrunch, on Ode with Anthropic (15 July 2026): joint venture of Anthropic, Blackstone, Hellman & Friedman, Goldman Sachs and others.
- [12] Distyl AI, homepage (read 20 October 2026): deploying purpose-built products customized to the business.
- [13] Turing, Enterprise AI (read 20 October 2026): forward-deployed engineers and an AI control plane.
- [14] 10Pearls, Forward deployed engineers (read 20 October 2026): engagement shapes; global offices used to match a client's hours.
- [15] 10Pearls, About (read 20 October 2026): offices in Costa Rica, Colombia, the UK, Pakistan and Peru.
- [16] Caylent, Careers (read 20 October 2026): a remote-first organization whose workforce has been 100% remote since 2015.
More in Cost and method
What forward deployed engineering costs from a US firm
None of the fifteen forward deployed engineering firms in this guide published a price on 20 October 2026. Two published how they charge: Tenex by story points of delivered work, and Caylent on outcomes or for a fixed fee where results can be measured. This page gathers the public figures that do exist, mostly pay data for engineers, explains the common ways firms charge in plain words, and lists what a quote should show so that you can compare firms on the same terms.
How the FDE firms in this guide were chosen
Every firm in this guide passed three tests on 20 October 2026. Its own pages offer engineers who work inside a client's team. Its own pages or an independent source name a US office, or it is a US company that works remotely. And it sells the engineers as its main offer, so the in-house FDE teams of software makers such as Palantir and AWS are left out. This page lists each check, how the sources are labelled, the candidates left out with the reason, and the facts that could not be confirmed.