[{"data":1,"prerenderedAt":486},["ShallowReactive",2],{"posts":3},[4,259,371],{"id":5,"title":6,"body":7,"date":248,"description":249,"draft":250,"extension":251,"meta":252,"navigation":253,"path":254,"seo":255,"stem":256,"updated":257,"__hash__":258},"posts\u002Fblog\u002Fforward-deployed-engineer-job-description.md","Forward deployed engineer job description: what to write",{"type":8,"value":9,"toc":236},"minimark",[10,14,23,28,38,47,50,54,57,60,64,67,89,92,96,99,102,141,144,148,151,154,158,161,204,207,211,214,221,224,228],[11,12,13],"p",{},"You have decided you need a forward deployed engineer, or someone has told you that is the title to use. Now you have to write the job description. Most of the ones you can copy were written by AI vendors hiring people to install their own product at customer sites, so they describe a different job from the one you have. Here is one written from the buyer's side: the responsibilities, the skills, what to leave out, and whether the role should be a hire at all.",[11,15,16,17,22],{},"If you want the role explained first, the longer version is on the ",[18,19,21],"a",{"href":20},"\u002Fforward-deployed-engineer","what a forward deployed engineer is"," page.",[24,25,27],"h2",{"id":26},"why-most-fde-job-descriptions-do-not-fit-your-company","Why most FDE job descriptions do not fit your company",[11,29,30,31,37],{},"The title comes from Palantir. Gergely Orosz's ",[18,32,36],{"href":33,"rel":34},"https:\u002F\u002Fnewsletter.pragmaticengineer.com\u002Fp\u002Fforward-deployed-engineers",[35],"nofollow","deep dive on forward deployed engineers"," in The Pragmatic Engineer explains that Palantir called these engineers \"Deltas,\" and that until around 2016 it had more of them than regular software engineers. Palantir's own description, quoted there, draws the line well: a product developer works on \"one capability, many customers,\" and a Delta on \"one customer, many capabilities.\"",[11,39,40,41,46],{},"The title has since spread to AI companies. The ",[18,42,45],{"href":43,"rel":44},"https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FForward_deployed_engineer",[35],"Wikipedia entry for the role"," notes that Anthropic, OpenAI, Amazon Web Services, Microsoft, and Google Cloud all announced forward deployed divisions in 2026. Those engineers work for the vendor. Their job is to make the vendor's product succeed inside a customer, so their postings mention sales, travel to customer sites, and fixes that go back into the vendor's platform. One Salesforce senior FDE posting that Orosz quotes lists ten expectations, including driving adoption of its own agent product.",[11,48,49],{},"Your situation is the reverse. You are the customer. You need one experienced person who will come into your systems and ship an AI feature your team keeps. So delete the parts about sales, quotas, and someone else's product roadmap, and write the rest around your own workflow.",[24,51,53],{"id":52},"start-the-job-description-with-the-workflow","Start the job description with the workflow",[11,55,56],{},"Before any list of duties, write one sentence that names the work. Something like: \"Our support team retypes refund reasons from email into the billing tool, and the errors show up in month-end reports.\" Name the process, the person who does it today, the system it lives in, and what goes wrong when it fails.",[11,58,59],{},"That sentence does more filtering than any list of frameworks. It attracts people who want to ship into a real operation and puts off people who want to pick a model and build a demo. If you cannot write it yet, you are not ready to hire a builder.",[24,61,63],{"id":62},"forward-deployed-engineer-responsibilities","Forward deployed engineer responsibilities",[11,65,66],{},"Write responsibilities as changes in your systems, not as activities. A version that fits most small and mid-size companies:",[68,69,70,74,77,80,83,86],"ul",{},[71,72,73],"li",{},"Sit with the people who run the workflow and write down how it actually works, including the exceptions.",[71,75,76],{},"Build one AI agent or LLM feature into the product or operation you already run, in your repository, behind your existing auth.",[71,78,79],{},"Set clear limits: what the system may do on its own, what it may only draft, and what a person still checks.",[71,81,82],{},"Build a small set of test cases from your real records, so \"it works\" means it works on your data.",[71,84,85],{},"Ship in small releases to the environment your team already uses, and review each one against a real case with the person who owns the process.",[71,87,88],{},"Leave the code, the accounts, and short handover notes in a state your team can run and change without the engineer.",[11,90,91],{},"The last line matters most, and it is the one vendor postings leave out. A vendor's FDE can leave behind a dependency on the vendor. Yours should leave behind a system your team owns.",[24,93,95],{"id":94},"forward-deployed-engineer-skills-to-ask-for","Forward deployed engineer skills to ask for",[11,97,98],{},"Software engineering comes first. Orosz notes that almost every FDE posting asks for a solid engineering background and real-world experience shipping. Applied AI is still software. The model is one dependency. The rest is permissions, data, deploys, and edge cases.",[11,100,101],{},"The skills worth listing:",[68,103,104,111,117,123,129,135],{},[71,105,106,110],{},[107,108,109],"strong",{},"Production experience."," Has shipped and maintained systems other people depend on, not only prototypes.",[71,112,113,116],{},[107,114,115],{},"Full stack range."," Comfortable across an API, a database, a user interface, and a deploy pipeline, because the AI feature touches all of them.",[71,118,119,122],{},[107,120,121],{},"Hands-on LLM work."," Prompts, retrieval over your own records, tool calls, and evaluating output against real cases.",[71,124,125,128],{},[107,126,127],{},"Working with operators."," Can sit with a dispatcher or a finance clerk, ask good questions, and explain a tradeoff without jargon.",[71,130,131,134],{},[107,132,133],{},"Judgment about scope."," Will say what should not be automated, and why.",[71,136,137,140],{},[107,138,139],{},"Clear writing."," Pull requests, notes, and a handover your team can read after the engagement.",[11,142,143],{},"What to leave off: a long list of model names and frameworks, a research degree requirement, and \"10x\" language. Those describe a different job and attract the wrong candidates.",[24,145,147],{"id":146},"forward-deployed-engineer-vs-solutions-engineer","Forward deployed engineer vs solutions engineer",[11,149,150],{},"These two roles get mixed up in job postings, and the difference changes who you should hire. Orosz describes how OpenAI separates them. Its solutions architects are closer to an advisory role. They rarely write code on customer infrastructure, and they usually build proofs of concept with anonymized or offline data. Its forward deployed engineers write code directly on customer infrastructure, using the customer's own tools.",[11,152,153],{},"If what you need is a proof of concept to decide whether an idea is worth pursuing, a solutions engineer, or a short advisory engagement, is the smaller and cheaper job. If your pilot already worked in a demo and never shipped, you need the forward deployed version: someone who will put it into production on your real data.",[24,155,157],{"id":156},"a-sample-job-description-you-can-adapt","A sample job description you can adapt",[11,159,160],{},"Replace the bracketed parts, and leave out anything that is not true for your company.",[162,163,164,169,189,196,199],"blockquote",{},[11,165,166],{},[107,167,168],{},"Forward Deployed Engineer (Applied AI)",[11,170,171,172,176,177,180,181,184,185,188],{},"We want ",[173,174,175],"span",{},"the workflow, in one sentence",". Today ",[173,178,179],{},"the team or person"," does this by hand in ",[173,182,183],{},"the system",", and when it goes wrong, ",[173,186,187],{},"the cost",".",[11,190,191,192,195],{},"You will work inside our team to put an AI feature into that workflow. You will sit with ",[173,193,194],{},"the operator"," to learn how the work really happens, build the feature in our codebase and environment, set limits on what it may do on its own, and test it against real cases before each release. When you are done, our team should be able to run and change the system without you.",[11,197,198],{},"You have shipped production software that other people depend on, across the API, the database, and the interface. You have built features on top of language models with real data and real permissions. You can explain a tradeoff to someone who does not write code, and you will tell us when a step should stay manual.",[11,200,201],{},[173,202,203],{},"Employment type, location, and reporting line.",[11,205,206],{},"The last line is where you decide whether this is a full-time seat or a bounded engagement, which is the next question.",[24,208,210],{"id":209},"when-a-full-time-hire-is-the-wrong-choice","When a full-time hire is the wrong choice",[11,212,213],{},"A full-time forward deployed engineer makes sense when you can see a year or more of AI work, a sequence of workflows that will keep a senior engineer busy, and you can wait out a senior search.",[11,215,216,217,22],{},"It is the wrong choice when you can only name one to three workflows and the need drops after that. A salary continues through the quiet months. It is also the wrong choice when the work cannot wait for a search, which is often the real problem. You need a senior AI engineer now, and you cannot justify, or wait for, a permanent seat. In that case the same job description works as the brief for a fractional engagement, a few months or a fixed sprint, ending in a handover. The engagement shapes (a monthly retainer, a fixed-scope sprint, or advisory hours) are on the ",[18,218,220],{"href":219},"\u002Fpricing","pricing",[11,222,223],{},"And if you cannot write the workflow sentence at the top of this post, do not hire anyone to build yet. Start with a short advisory block that forces the choice, then write the job description.",[24,225,227],{"id":226},"next-step","Next step",[11,229,230,231,235],{},"Write the workflow sentence and the responsibilities list. If you want a second opinion on whether the result is a full-time seat, a sprint, or not a build at all, request an intro call on the ",[18,232,234],{"href":233},"\u002Fcontact","contact"," page and bring the draft.",{"title":237,"searchDepth":238,"depth":238,"links":239},"",2,[240,241,242,243,244,245,246,247],{"id":26,"depth":238,"text":27},{"id":52,"depth":238,"text":53},{"id":62,"depth":238,"text":63},{"id":94,"depth":238,"text":95},{"id":146,"depth":238,"text":147},{"id":156,"depth":238,"text":157},{"id":209,"depth":238,"text":210},{"id":226,"depth":238,"text":227},"2026-11-04","A forward deployed engineer job description written from the buyer's side: the responsibilities, skills, and limits to list, plus when not to hire.",false,"md",{},true,"\u002Fblog\u002Fforward-deployed-engineer-job-description",{"title":6,"description":249},"blog\u002Fforward-deployed-engineer-job-description","2026-11-05","VhT84ZuC7pbEtzlvRXbz3NoBd5r3BcPyksCikOtD3KA",{"id":260,"title":261,"body":262,"date":364,"description":365,"draft":253,"extension":251,"meta":366,"navigation":253,"path":367,"seo":368,"stem":369,"updated":364,"__hash__":370},"posts\u002Fblog\u002Fforward-deployed-engineer-vs-ai-consultant.md","Forward deployed engineer vs AI consultant",{"type":8,"value":263,"toc":356},[264,267,273,277,280,283,286,290,293,296,299,303,306,309,317,321,324,327,331,334,337,343,347,350],[11,265,266],{},"A founder asks for \"someone who knows AI.\" Two kinds of people answer. One will spend a few weeks interviewing the team and will hand over a recommendation. The other will spend the next month in the repository, and the recommendation will be a working change. Both can be the right hire. They are not the same hire.",[11,268,269,270,22],{},"This note is a draft for review. The longer version of the role is on the ",[18,271,272],{"href":20},"forward deployed engineer",[24,274,276],{"id":275},"what-each-person-is-paid-to-produce","What each person is paid to produce",[11,278,279],{},"An AI consultant is paid for judgment at a distance. The good ones are specific. They will tell you which workflow is worth software, which vendor is a fit, and which idea should wait. The artifact is a document, a workshop, or a decision in a meeting. When they leave, the work of building is still ahead of you.",[11,281,282],{},"A forward deployed engineer is paid for a change in the system you already run. The artifact is a pull request, a release, and a person on your team who has used the result on a real case. The judgment is still there. It shows up as scope, as a refusal to automate a step that needs a person, and as a short note next to the code. It does not show up as the thing you were invoiced for.",[11,284,285],{},"If you are comparing proposals, ask each person what you will hold on the last day. A plan and a build can both be honest answers. A plan described as a build is the mismatch to avoid.",[24,287,289],{"id":288},"time-and-proximity","Time and proximity",[11,291,292],{},"Consultants batch their time. A day on site, a week of writing, a readout. That rhythm fits a decision that needs interviews and then a quiet synthesis. It fits poorly when the truth of the workflow is in the exceptions, and the exceptions show up on a Wednesday when a customer does something the interview did not mention.",[11,294,295],{},"A forward deployed engineer is in the tools your team already uses, for the length of the engagement. Questions get answered against the record, not against a recollection from last Tuesday's workshop. When the model fails, the person who hit the failure is in the same conversation as the person who will change the prompt or the code.",[11,297,298],{},"Proximity has a cost. You have to make someone available. An embedded engineer without access to the operator who does the job will rebuild the consultant's problem: a clean solution to a process that does not exist.",[24,300,302],{"id":301},"who-owns-the-result","Who owns the result",[11,304,305],{},"After a consulting engagement, you own the document. The reasoning is yours to accept or ignore. The implementation risk sits with whoever you hire next, and that person may not have been in the room.",[11,307,308],{},"After a forward deployed engagement, you own the code, the accounts, and the operational notes. The engineer was a guest in your repository. There should be no private environment, no license that expires when the contract does, and no step that only the guest can run.",[11,310,311,312,316],{},"Ask, in writing, where the code will live and who can deploy it on the day after the contract ends. The answer you want is your repository and your team. ",[18,313,315],{"href":314},"\u002Fservices","Services"," are scoped that way.",[24,318,320],{"id":319},"when-the-consultant-is-the-right-buy","When the consultant is the right buy",[11,322,323],{},"Buy the consultant when you do not yet know which problem is worth software. A few weeks of interviews can save a quarter of building the wrong agent. Buy the consultant when the decision is a vendor choice and your own team will implement it. Buy the consultant when you want a challenge to a plan you already love, from someone who will not then be paid to build that plan.",[11,325,326],{},"In those cases the distance is a feature. The person is not attached to a particular architecture because they will not be the one maintaining it.",[24,328,330],{"id":329},"when-the-forward-deployed-engineer-is-the-right-buy","When the forward deployed engineer is the right buy",[11,332,333],{},"Buy the engineer when you can name the workflow. You know the system it lives in. You know who feels the pain. You need the change in production, and you do not have a senior person free to do it. A fractional engagement covers a few months of that work without opening a full-time seat.",[11,335,336],{},"Buy the engineer when a previous consulting round already produced a clear recommendation and the recommendation has been sitting in a doc. The missing piece is implementation in your stack, with your permissions and your data.",[11,338,339,340,342],{},"The ",[18,341,220],{"href":219}," page separates a monthly retainer, a fixed sprint, and advisory hours. Advisory hours are the slice of this practice that behaves like consulting. The retainer and the sprint are the build.",[24,344,346],{"id":345},"a-practical-way-to-choose","A practical way to choose",[11,348,349],{},"Write one sentence: \"In ninety days, we will have ___.\" If the blank is a decision, talk to a consultant, or book advisory hours. If the blank is a feature your staff or your customers will have used, talk to a forward deployed engineer. If the blank is both, do the decision first and keep it short, then start the build with a scope that names the workflow.",[11,351,352,353,355],{},"If you want to test the fit, the ",[18,354,234],{"href":233}," page is a 30-minute intro call. Bring the sentence with the blank filled in.",{"title":237,"searchDepth":238,"depth":238,"links":357},[358,359,360,361,362,363],{"id":275,"depth":238,"text":276},{"id":288,"depth":238,"text":289},{"id":301,"depth":238,"text":302},{"id":319,"depth":238,"text":320},{"id":329,"depth":238,"text":330},{"id":345,"depth":238,"text":346},"2026-11-03","A forward deployed engineer ships in your stack. An AI consultant leaves you a plan. Here is how to tell which one you are buying.",{"robots":250,"sitemap":250},"\u002Fblog\u002Fforward-deployed-engineer-vs-ai-consultant",{"title":261,"description":365},"blog\u002Fforward-deployed-engineer-vs-ai-consultant","cYs5cWZXfQ1_xJYPMv_GUziw6aEhclmAssLwOjM4TPs",{"id":372,"title":373,"body":374,"date":364,"description":480,"draft":253,"extension":251,"meta":481,"navigation":253,"path":482,"seo":483,"stem":484,"updated":364,"__hash__":485},"posts\u002Fblog\u002Fhow-to-hire-an-applied-ai-engineer.md","How to hire an applied AI engineer",{"type":8,"value":375,"toc":472},[376,379,385,389,392,395,398,402,405,408,411,414,418,421,424,427,430,434,437,440,446,450,453,456,460,463],[11,377,378],{},"Hiring an applied AI engineer goes wrong in a predictable way. The job post lists models and frameworks. The interviews test puzzle skill. The person who joins can demo a chatbot and cannot put it inside the product you already ship. This note is a draft for review, and a shorter path through the same decision.",[11,380,381,382,384],{},"An applied AI engineer is someone who can take a real workflow, put a model in it with limits, and leave the system in a state your team can run. A ",[18,383,272],{"href":20}," is that person working inside your company for a bounded period. The hiring test is the same either way.",[24,386,388],{"id":387},"start-from-a-workflow","Start from a workflow",[11,390,391],{},"Write the job from the work, not from the technology. Name the process, the person who does it today, the system it lives in, and what goes wrong when it fails. \"Build AI features\" will attract people who want to pick a model. \"Stop the support team retyping refund reasons into the billing tool\" will attract people who want to ship.",[11,393,394],{},"If you cannot name the workflow, you are not ready to hire a builder. You may be ready for a short advisory engagement that forces the choice. Hiring a builder before the choice is how prototypes pile up.",[11,396,397],{},"The workflow also tells you seniority. A narrow internal tool with a willing operator is a smaller job than a customer-facing feature in a product with strict permissions. Match the person to that weight. A senior full stack engineer who has shipped production systems is the usual fit for the second one.",[24,399,401],{"id":400},"what-to-look-for","What to look for",[11,403,404],{},"Look for someone who has put software into a system other people depend on. Applied AI is still software. The model is one dependency. The rest is auth, data, deploys, and the boring edge cases.",[11,406,407],{},"Look for someone who talks about limits. What the system is allowed to do on its own. What it may only draft. What a person still checks. Candidates who answer every failure with a better prompt have not sat with a workflow that has money or trust on the line.",[11,409,410],{},"Look for someone who can explain a past project to a non-engineer in the room. If the explanation needs a tour of the framework, they have not had to hand a system to the team that will live with it. Ask them to describe the handover, not the architecture diagram.",[11,412,413],{},"Look for taste about scope. The useful answer to \"how would you start\" is one slice, in production, reviewed by the person who does the job. A program of five agents in the first month is a signal that the candidate has not had to maintain what they shipped.",[24,415,417],{"id":416},"questions-worth-asking","Questions worth asking",[11,419,420],{},"Ask what they would refuse to automate, and why. You are listening for a real reason, such as a decision that needs a person, or a data source nobody trusts. You are not listening for a slogan about humans in the loop.",[11,422,423],{},"Ask them to walk through a failure from a system they shipped. What broke, who noticed, what changed in the product. A candidate with no failure story has not been close to production, or is not willing to tell you.",[11,425,426],{},"Ask where the code would live, and who could deploy it after they leave. For a fractional hire this is the whole point. For a full-time hire it still matters, because people leave.",[11,428,429],{},"Ask what they need from you in the first two weeks. The right answer includes time with the operator and access to a real sample of the data, with the permissions your own staff would have. A candidate who only needs a prompt and an API key is planning a demo.",[24,431,433],{"id":432},"fractional-or-full-time","Fractional or full-time",[11,435,436],{},"Hire full-time when you can see a year of this work, and you can wait out a search. Hire fractional when the work is real and bounded: one to three workflows, a few months, a team that can take the system over at the end. The cost shapes differ. A salary continues in quiet months. A retainer or a sprint stops when the scope is done.",[11,438,439],{},"A fractional applied AI engineer is also a way to learn what the full-time role should say. The job post you write after one shipped workflow is better than the job post you write from a list of model names.",[11,441,442,443,445],{},"The engagement shapes I use, retainer, sprint, and advisory hours, are on the ",[18,444,220],{"href":219}," page. The bracketed prices there are filled in after a scope, not before.",[24,447,449],{"id":448},"signs-to-walk-away","Signs to walk away",[11,451,452],{},"Walk away from a proposal that starts with a platform and has no workflow. Walk away from a promise of accuracy with no mention of your own cases. Walk away from a build that lives in the vendor's tenant and cannot be exported. Walk away from a team that will not name who reviews a real case each week. The engineer cannot invent that person.",[11,454,455],{},"Walk away, too, from your own brief if it is still \"explore AI.\" Exploration without a process produces a chat window. You can buy that more cheaply than a senior engineer, and you should.",[24,457,459],{"id":458},"the-first-conversation","The first conversation",[11,461,462],{},"Keep the first conversation to half an hour. Bring the workflow in one sentence, the name of the system, and the name of the person who will review the work. Ask the candidate to say the outcome back to you in their own words, and to say what they would need in the first week.",[11,464,465,466,468,469,188],{},"You are done with the conversation when you know whether the next artifact is a decision, a fixed build, or a month of embedded work. If you want that conversation with me, it is on the ",[18,467,234],{"href":233}," page. The kinds of build I take are on ",[18,470,471],{"href":314},"services",{"title":237,"searchDepth":238,"depth":238,"links":473},[474,475,476,477,478,479],{"id":387,"depth":238,"text":388},{"id":400,"depth":238,"text":401},{"id":416,"depth":238,"text":417},{"id":432,"depth":238,"text":433},{"id":448,"depth":238,"text":449},{"id":458,"depth":238,"text":459},"How to hire an applied AI engineer: the workflow first, the questions that matter, and when fractional beats a full-time seat.",{"robots":250,"sitemap":250},"\u002Fblog\u002Fhow-to-hire-an-applied-ai-engineer",{"title":373,"description":480},"blog\u002Fhow-to-hire-an-applied-ai-engineer","a4JrqPc94khy6XM7IFtlMufHL64HVjfIaxsVopuEP5s",1793853305675]