The AI Hatchery
Build AI agents that hold up in production, and grow into AI DevRel, FDE and AI engineering roles.
© 2026 The AI Hatchery.
OSS4AI / THE AI HATCHERY · FREE CAREER FIELD GUIDE
Between the code and the customer
FDE, applied AI engineer, AI DevRel. Understand the work, choose a direction, and build proof that gets you into the conversation.
For engineers and PMs who already ship software. No signup required to read this guide. Hosted by The AI Hatchery on beehiiv.
Download the free PDF →01 / Get your bearings
Choose by the work you want to do, not the title. These roles overlap, and companies draw the boundaries differently.
One customer, deeply
Scope, build and run AI systems inside a customer’s infrastructure. Work through the messy data, integration and deployment details.
Proof: a working system, evals and a plan to operate it.
Several customers, often with sales
Turn technical discovery into pilots, prototypes and evaluation suites. Teach the customer team and defend the choices you made.
Proof: a build, demo and business case.
A developer community, one-to-many
Build sample apps, tutorials and talks. Help developers adopt a product, then bring their rough edges back to engineering.
Proof: code someone else can run and learn from.
Read the pay figures as a dated snapshot.
Research was checked October 24, 2026. Employer examples below are historical US posted ranges, not current vacancies or expected take-home pay. Location, level, equity and commission matter; these selected AI-company postings are not a market-wide benchmark. Verify the current posting before applying.
A / One customer, many capabilities
Palantir’s original distinction is useful: regular engineers focus on “one capability, many customers”; FDEs focus on “one customer, many capabilities.” [16]
Scope an ambiguous request, sequence the work and build on the customer’s infrastructure.
Handle integrations, security and procurement gates. Deliver a system the customer can operate. [17]
Build agents or retrieval workflows, plus evaluation and monitoring that expose failures.
Translate customer needs back to your own product and engineering teams.
Customer-facing experience under another title can count. Decagon names solutions engineering, consulting, implementation engineering and technical product/PM backgrounds. Anthropic explicitly says side projects and prototypes count. [21] [17]
Practice coding, open-ended scoping and explaining something you built. Palantir’s hiring guidance recommends delivering a functioning idea first, then expanding it. Company loops vary. [24] [26]
Make a small agent or RAG prototype useful in a realistic workflow. Show its data boundaries, failure cases, eval results and operating plan.
US research snapshot · Oct 24, 2026
Palantir · New York $135,000–$200,000Forward Deployed Software Engineer [3]
Anthropic · NYC / SF / Seattle $280,000–$320,000Staff Forward Deployed Engineer [17]
Travel? Customer assignment length? Deployment ownership? On-call responsibilities? What counts as success?
B / Build, demo, defend
Titles include applied AI engineer, solutions engineer or architect, and agent engineer. The common thread: helping customer teams get from discovery to a working pilot.
Build customized pilots, prototypes and evaluation suites. [34]
Lead hands-on workshops and code reviews with customer engineers. [34]
Partner with sales across accounts and explain technical choices in business terms. [6]
Send useful customer feedback to product and research teams. [6]
Anthropic names product engineers, software engineers, FDEs and technical PMs among relevant backgrounds. OpenAI lists prototypes built on its API as a plus. A project that works, and a clear explanation of its value, do more than a list of tools. [34] [6]
Databricks’ researched posting explicitly included design and architecture, live coding, a “Build, Demo, Pitch!” presentation, and reference checks. Practice a demo you can defend under questions. [7]
Bring one prototype with a short demo: the customer problem, eval results, tradeoffs, security boundaries and what it would take to deploy.
US research snapshot · Oct 24, 2026
Anthropic · SF / NYC $200,000–$320,000Applied AI Engineer, Enterprise [34]
Databricks · CA / SF $219,100–$301,300Senior Solutions Architect [7]
Databricks says commission-based ranges are on-target earnings (OTE), without clarifying this posting’s classification. Confirm base vs OTE with the recruiter.
Pre-sales or post-sales? Quota or commission? How many accounts? Pilot-to-production handoff? Rules on AI tools in each interview?
C / Code that teaches
Developer advocate, DevRel engineer and developer experience engineer are common titles. The audience is a community, not only one customer account.
Turn products into runnable demos, templates and open-source projects. [11]
Teach through tutorials, video, talks and technical documentation.
Support communities and office hours, and notice where developers get stuck. [8]
Hit rough edges early and get them fixed or documented before more people hit them. [11]
Build public proof: a sample repo with a clear README, a tutorial, a short recorded talk, and documentation contributions. Vercel’s researched posting required a link to something that taught developers something. [55] [11]
Community and ambassador programs can give you practice, not a job guarantee. Explore Claude Community Ambassadors and LangChain community programs; check their current terms and availability. [57] [58]
Practice teaching a technical topic to a mixed audience in 15–20 minutes, then taking questions. Some companies add coding or a take-home. PostHog’s published process uses a paid workday with a repo and tutorial. [50] [53]
Make a small repo someone else can run. Add a README, one useful tutorial, troubleshooting notes, and a recorded walkthrough.
US research snapshot · Oct 24, 2026
Pinecone · San Francisco $140,000–$240,000Developer Advocate · plus equity [8]
Vercel · San Francisco $168,000–$252,000DevRel Engineer, Agentic Infrastructure · base [11]
Who is the audience? How much code vs content? Travel and events? Who acts on developer feedback? How is adoption measured?
05 / A direction, not a diagnosis
Pick the work you would enjoy doing repeatedly. This is a static self-check, not an automated or validated career assessment. Count A, B and C choices on paper or in your own notes; there is no answer submission form.
A. One customer, deeply, for months.
B. Several customers, each to a working pilot.
C. Many developers through reusable examples.
A. Shipping a fix into a customer’s system.
B. Walking a team through a demo you built.
C. Publishing a tutorial or giving a talk.
A. Inside a customer’s systems and workflows.
B. With an account team and customer engineers.
C. At a meetup, in a community or on camera.
A. Owning a scoped customer deployment.
B. Working next to sales on technical discovery.
C. Helping developers adopt a product at scale.
A. Unblock the customer and get the deployment live.
B. Explain the cross-customer pattern to product.
C. Fix or document it before more developers hit it.
A. A system integrated with a messy existing codebase.
B. A build, demo and pitch for a customer panel.
C. A repo, talk or video that teaches something.
Count your answers: mostly A → FDE; mostly B → applied AI; mostly C → AI DevRel. A mix is normal. Read the responsibilities in individual postings.
All three roles need technical skill, communication and an understanding of customer value. FDEs can work with sales; DevRel can support commercial goals. These are tendencies, not hard boundaries.
06 / Turn interest into evidence
Pick one direction. Finish one useful thing. Rehearse the part of the interview that makes that role different.
Week 1
Choose one role. Mark the responsibilities, experience, stack, location, travel and deployment ownership. Check whether pay means base or OTE.
Output: a short list of repeated requirements and one skill to demonstrate.
Weeks 2–3
FDE: an agent or retrieval workflow with integrations, failure cases and an operating plan.
Applied AI: a pilot with an eval and a demo that explains the customer value.
DevRel: a runnable repo and tutorial that help another developer solve a real problem.
Output: one link you can put on an application.
Week 4
FDE: scope a vague request out loud, including data, interfaces, edge cases and tradeoffs.
Applied AI: build, demo and pitch. Defend the evals, security and path to deployment.
DevRel: teach one topic to a mixed audience in 15–20 minutes. [50]
Output: a recording, feedback and one improved version.
Can someone else run or understand your project from the README?
Can you explain when it fails, not only when it works?
Can you connect your existing experience to the role’s actual responsibilities?
Have you checked the current posting and asked about the interview’s AI-tool rules?
Use synthetic or public data in portfolio projects. Do not publish customer data, employer code, credentials or confidential work.
07 / Follow the evidence
Adapted from The AI Hatchery’s careers research, checked October 24, 2026. Reference numbers match the original research. Job pages are historical examples and may change, redirect or close.
Salary examples are posted US ranges from selected employers, not offers or a compensation survey. Interview formats differ by company, role and date. Preparation suggestions and the 30-day plan are guidance, not hiring guarantees. No paid-product launch promises or live-vacancy claims are included.
[3] Palantir job posting: Forward Deployed Software Engineer, New Yorkchecked 2026-10-24. Historical research reference, not an open-job guarantee.
[16] Palantir blog: Dev versus Delta: demystifying engineering roles at Palantir2019 · checked 2026-10-24. Publisher may require browser verification. Historical research reference, not an open-job guarantee.
[17] Anthropic job posting: Staff Forward Deployed Engineerchecked 2026-10-24. Historical research reference, not an open-job guarantee.
[21] Decagon job posting: Agent Deployment Engineerchecked 2026-10-24. Historical research reference, not an open-job guarantee.
[24] Palantir careers: The phone interviewchecked 2026-10-24. Historical research reference, not an open-job guarantee.
[26] Palantir careers: Navigating open-ended questionschecked 2026-10-24. Historical research reference, not an open-job guarantee.
[34] Anthropic job posting: Applied AI Engineer, Enterprisechecked 2026-10-24. Historical research reference, not an open-job guarantee.
[6] OpenAI job posting: Applied AI Engineer, Startupschecked 2026-10-24. Historical research reference, not an open-job guarantee.
[7] Databricks job posting: Sr. Solutions Architect, Digital Native Business (Strategic Accounts)checked 2026-10-24. Historical research reference, not an open-job guarantee.
[36] Sierra job posting: Software Engineer, Agentchecked 2026-10-24. Historical research reference, not an open-job guarantee.
[11] Vercel job posting: DevRel Engineer, Agentic Infrastructureposted 2026-07-22, updated 2026-10-20 · checked 2026-10-24. Historical research reference, not an open-job guarantee.
[8] Pinecone job posting: Developer Advocateposted 2026-10-20 · checked 2026-10-24. Linked page now titled Staff Developer Advocate; the pay example retains the researched historical title. Historical research reference, not an open-job guarantee.
[50] DevRelJobs: Acing the DevRel interviewdated 2026-02-15 · checked 2026-10-24. Historical research reference, not an open-job guarantee.
[53] PostHog handbook: DevRel hiring processchecked 2026-10-24. Historical research reference, not an open-job guarantee.
[55] DevRelJobs: Building a DevRel portfoliodated 2026-02-10 · checked 2026-10-24. Historical research reference, not an open-job guarantee.
[57] Claude Community Ambassadors (Anthropic)checked 2026-10-24. Historical research reference, not an open-job guarantee.
[58] LangChain community programschecked 2026-10-24. Historical research reference, not an open-job guarantee.
Build AI agents that hold up in production, and grow into AI DevRel, FDE and AI engineering roles.