The jobs behind forward deployed

In its September 2026 description of its engineering practice, DoiT says its team began as Customer Reliability Engineering and became Forward Deployed Engineering. DoiT says the underlying job stayed the same. It describes the new name as communicating the function more clearly to customers and potential recruits.

In his January 27, 2026 essay for a16z, Tom Hollands argues that titles can raise a function's standing and signal an organization's capabilities. The essay does not measure how many jobs changed.

DoiT's explainer also gives a consultant's endpoint as delivery of a report. Climb's AI-lead advertisement describes the company as a data and AI consultancy and gives the lead ownership of client engagements from discovery through executive handoff. The duties include directing architects and engineers, resolving delivery problems and shaping follow-on engagements.

An established team can acquire the prefix without the employer adding a function or hiring another person. A genuine expansion of delivery work and a change in its name could both increase the recorded prevalence of the title. Separating them requires a record of the duties before and after the rename.

Palantir's November 2020 role interview already described engineers configuring the company's platforms around a customer's needs, maintaining production systems and carrying useful field solutions back into product development. Customer-specific work and a route back into the shared product were part of that employer's account before the generative-AI boom.

In a March 2021 interview with the Society for Computers and Law, qualified solicitor Will Lilley described working as a legal engineer at Simmons Wavelength. His projects included improving due diligence, producing a legal-data reporting tool and automating parts of private-equity fund creation. He described understanding lawyers' procedures and working with data scientists to improve them.

Pave's November 2025 report measures the share of customer companies employing someone with an engineering title of this kind. Company title presence rose from 0.34% in January 2023 to 1.24% in September 2025, about 3.6 times the earlier rounded company share. The observations count companies with a matching title, rather than employees or hires.

The prose describes about 9,000 integrated customers; the chart footer cites 10,083 historical participants. Monthly denominators and a stable-panel rule are unpublished. These engineering-title observations cannot establish the prevalence of lawyer or creative titles.

LinkedIn's August 2026 US report gives forward-deployed engineering 7.3% of AI-builder postings in its partial 2026 data. It uses standardized occupations rather than literal title matching, and that engineering category does not cover every function using the prefix.

The two measures cannot be joined into one growth series, and neither supplies a baseline for the October 14 sample.

The main figures cover 24 original postings across 17 employer groups, checked on October 14, 2026, when each had forward deployed or its hyphenated form in the title. The full download preserves all 28 postings across 20 groups from that October 14 snapshot, including one deployment-title variant and three adjacent comparison roles; when rechecked on October 27, all 24 main postings remained listed, though some titles and terms had changed and one adjacent comparison posting had closed. Duplicate locations and shared-brand requisitions are reconciled.

The collection is purposive, assembled to inspect different functions rather than estimate their market shares. It covers English-language originals without a country restriction. A role appearing here establishes what an employer advertised; it does not establish how many people were hired, whether every duty occurs in practice or how common the role is elsewhere.

Ten of the 24 postings require coding, three prefer it and three explicitly say it is unnecessary; eight leave it unstated. An unstated requirement stays unstated, including where a title sounds technical enough to invite an inference.

Thirteen of the 24 advertisements contain engineer or engineers in the title, including Profound's management vacancy, a separate advertisement from its marketing role. Nine require coding, two prefer it and two explicitly waive it. This is a word-match check, not a classification of all thirteen jobs as software engineering.

In this collection, Baseten's engineering posting explicitly asks for hands-on software development on customer AI applications. The engineer owns evaluation, production deployment and monitoring; the posting names quality, latency and cost as outcomes without setting numerical targets. Python is preferred for the production code. The duties include turning vague objectives into specifications and tested services, working alongside the customer's engineers through sales, implementation and expansion.

Materialize's engineering role requires SQL and integration code inside customer systems. The posting names connectors and continuously updated SQL views, with database-change feeds from Postgres or MySQL and streams from Kafka or Kinesis. Its migrations involve writing to the old and new systems before a controlled switch. The duties also include debugging streaming queries, production hardening, performance tuning, monitoring and handoff. The engineer is expected to leave designs and runbooks that outlast the engagement.

Higharc's product-manager posting specifies a concrete data problem: turn a homebuilder's scattered plan catalog into structured information. The manager is expected to write SQL, use notebooks and script against APIs, while setting delivery criteria for coverage and data quality. Its advertised output includes a usable plan library, not just a recommendation to build one. Its acceptance criteria include catalog coverage, data completeness and error thresholds, with the manager expected to defend them under schedule pressure.

On October 14, Causal Labs used the same product-manager title for leadership of customer deployments involving physical systems. It asks for enough technical understanding to lead the team and explicitly says the person need not write the code. The role leads deployment teams across the provider and the customer. Both roles own customer delivery, but a coding exercise would address an advertised requirement in one and introduce a new prerequisite in the other.

Higharc offers remote work within the continental US with travel of up to 30%; Causal Labs requires travel and extended periods at customer sites.

By October 27, Higharc had retitled its posting Sr. Product Manager - Studio Data Platform, and Causal Labs had retitled its posting Member of Operations Staff - Deployment Product.

OpenAI's legal-domain engineering role asks for production software development across backend and frontend systems, including writing and reviewing code. Legal-sector experience is helpful rather than required, and the advertisement does not specify a JD.

Eudia's government posting adds work inside a government customer's environment, a legal qualification and an active clearance or strong eligibility to obtain one. Existing clearance is preferred. Rejecting an otherwise eligible applicant for not already holding it would impose a stricter bar than the advertisement states.

Luma's US creative role asks for a senior practitioner with a required portfolio and experience in production pipelines. The work includes making content in Dream Machine, developing customer production methods and taking partners' needs back to the company. Its practical bar is visible in the finished creative work and the ability to help a brand or agency produce it.

Tribe AI's staff product designer has another output: interaction designs and clickable prototypes for customer AI systems. The responsibilities include conversation flows, escalation paths and error handling, supported by design experience and a portfolio. Customer-specific delivery can involve designing what a system should do before another person implements it.

Profound's marketing role explicitly waives coding and a computer-science degree. It seeks marketing experience and judgment to configure useful agents, evaluate outputs and drive customer adoption.

GC AI's solutions-attorney advertisement separates a sales-focused track from a forward-deployed track within the same role. The latter includes implementation, tailored prompting and continuing customer support. This adjacent comparison stays outside the main title-matched count.

On October 14, Hippocratic AI's customer-success executive posting explicitly separated the job from deployment: implementation managers owned configuration, data onboarding and getting the system live, while the executive took the customer relationship after go-live for renewal, adoption and expansion, with travel, customer-site work, required health-system account experience and a preferred clinical license. By October 27, the employer had rewritten that listing to give the executive post-sale ownership from discovery through rollout and adoption, removing the separate implementation-manager handoff. This adjacent comparison role is outside the main title-matched count.

Harvey's buyout finance role asks for investment-side experience and financial modeling, with Python or SQL as an advantage. The work includes pilots using live deal materials, followed by screening, underwriting, diligence and investment-committee workflows in a firm's own formats. It continues into portfolio monitoring after a deal closes. Financial models and committee documents are outputs of the advertised investment work.

A successful demonstration can still leave a customer without a working process, and that process can still depend on the person who configured it.

Five of the 24 postings establish US annual base ranges in US dollars: Luma, two roles at OpenAI, Materialize and Titan AI. The comparison preserves each role and location. A generic salary field is insufficient when its base component is not established, and a dollar sign does not establish both currency and period.

The OpenAI San Francisco role and legal-domain New York role establish the unit and base identity across separate parts of their listings. Structured salary data supplies USD and an annual period, while the compensation description identifies base pay. The inclusion decision needs the complete original, rather than a salary-summary label alone.

Eudia's Palo Alto posting lists $190K to $220K base pay without an explicit annual USD unit and remains excluded. The October 14 snapshot also excludes Profound's and Harvey's on-target earnings; Harvey still identifies its ranges as on-target earnings, while by October 27 Profound's listings stated base pay before variable compensation.

The included base ranges mix roles, levels and locations. They describe advertised offers, not actual employee pay. Estimating a premium for the title would require a comparison that controls for those differences.

The following hiring exercise is unvalidated and hypothetical. It separates discovering a workflow rule from applying it to an availability decision, without claiming to predict employee performance. An employer should choose a deliverable that matches its vacancy and give every applicant the same stated conditions.

Suppose a fictional equipment-rental company requests an availability dashboard. Its records show when an item was returned, but staff do not consider it available until an inspection is approved. The initial brief omits that rule. Every candidate receives the same return and inspection records and is told they can clarify the workflow with a staff member working from common documented answers. Candidates are not expected to guess the rule.

Score discovery of the inspection rule separately from whether the query or acceptance criteria follow it.

For a software role, supply a small runnable API with return and inspection tables. Its availability query currently relies on return timestamps alone. Ask for a corrected query and an endpoint test. The common documented answers should specify that approval applies to the current return, not the item's entire history. Include an item approved on an earlier rental and returned again with its new inspection pending. Joining approvals by item alone would make it available incorrectly; the expected endpoint result must keep it unavailable. The reviewer can follow that case from the shared records through the query to the test result.

For a product-manager vacancy that does not require coding, ask for the expected availability result and acceptance criteria a developer could implement. The candidate can propose owners for the inspection and the software change.

A legal or creative exercise would need its own domain criteria and qualified reviewer, rather than importing the rental example as a universal screen.

Eighteen of the 24 postings explicitly give the customer-facing person a duty to return field needs to internal product direction. a16z's June 2025 services-led-growth argument describes customer implementation as a way to learn what an AI application needs and strengthen the product. The advertisements establish that employers want this work. They do not establish that feedback was accepted, other customers benefited or margins improved.

An employer can trace field feedback into the shared product's change history. Start with a recorded customer problem and identify who accepted the work. For a software fix, retain an example that reproduces the failure alongside the pull request and regression test, then check whether a later deployment uses the fix without a customer-specific patch. A reusable template counts as a different result from a change in the underlying product, and both differ from leaving a suggestion in an account note.

A person spending the week supporting one customer may have little time to document a general problem or help another team resolve it. The product owner needs a way to prioritize that contribution, and the deployment role needs time allocated to making it useful. A feedback bullet in the advertisement does not create either arrangement.

The product owner also needs to record when a customer request will remain a custom solution. A deployment worker should not remain responsible for delivering a product improvement the receiving team has declined.