Free guide for employers
Know when you need a Forward Deployed Engineer, Agentic AI Engineer, or another AI specialist. Get practical interview questions, a candidate scorecard, and a hiring brief your team can use.
Fourteen pages. Sent to your inbox and downloadable straight away.
Ready to hire? Talk to Focus GTS
Inside the guide
Fourteen pages built to be used in a hiring meeting, not filed away. Two of the pages are fillable, so your team can complete them on screen and share them.
01
Six AI engineering roles side by side: Forward Deployed Engineer, Agentic AI Engineer, ML Engineer, Data Scientist, Data Engineer and MLOps lead. Each one with its primary outcome and the delivery evidence to ask a candidate for.
02
The failure patterns that surface after the offer: buying a title instead of a result, mistaking a demo for delivery, screening on tool names, and expecting a recruiting partner to invent a role the business has not defined.
03
Core questions with what to listen for and when to probe further, so every candidate for the same role answers the same questions against the same criteria. Plus a bounded work sample you can run in ninety minutes.
04
A weighted hundred point view across production delivery, evaluation judgment, engineering depth, problem framing, security and communication. A discussion aid for the debrief, not an automated decision.
05
A fillable page covering the business outcome, what must be working by day ninety, must have capabilities, systems and access, engagement model and budget range. Complete it digitally and send it to your hiring manager.
The evaluation gap
A generalist recruiter can run a good search. The risk is in the process: when screening leans on titles and tool names without the technical depth to test what a candidate actually delivered, the evaluation happens on your team's calendar instead. These are checks you can apply to any recruiting partner, including us.
| Screening shortcut | What it can miss | What to ask for instead |
|---|---|---|
| Has AI on the resume | Whether the person built a prototype once or has owned a system that real users depend on. | A named system, their specific scope within it, and what happened after it went live. |
| Knows the frameworks | Whether they can evaluate output quality, handle failure paths and set permission boundaries. | How they measured whether the system was good, and what they changed when it was not. |
| Impressive demo | The distance between a working demo and production: monitoring, cost, latency, rollback and on-call ownership. | A walkthrough from prototype to production, including what broke and who fixed it. |
| Senior title | Whether the role was ever defined well enough for anyone to succeed in it. | A written brief: the outcome, the day ninety deliverable, and who the collaborators are. |
Why Focus GTS
Focus GTS combines recruiting experience with hands-on AI building and operating experience.
Hundreds of engineers placed
Placement experience gives us context for role scope, candidate backgrounds, and the realities of assembling a delivery team.
We build and run AI ourselves
We build and operate AI internally and through Navigator. Integration, evaluation, adoption and ongoing ownership are practical concerns for us, not talking points.
Active in the AI ecosystem
Focus GTS is an Anthropic partner and a member of the Agentics Foundation. Those relationships complement our recruiting and delivery experience.
What we staff
Across foundation models, classical ML, data infrastructure, and the new agentic and forward-deployed categories that didn't exist five years ago.
Senior ML engineers who productionize models — training pipelines, model serving, monitoring, drift detection, and the engineering rigor it takes to move past notebook prototypes.
Senior data scientists with statistics depth, experimentation design experience, and the business-facing communication chops to ship insights that actually change product decisions.
As dedicated data science recruiters, we vet for both technical depth and business fluency. Every data scientist we present has been screened on statistics, experimental design, and the ability to translate model outputs into decisions that move enterprise teams forward — not just keyword-matched on a résumé.
Specialists building production agentic systems — LangChain, LlamaIndex, the Anthropic Claude SDK, OpenAI Agents SDK, custom orchestration. Tool-use design, RAG architectures, evaluation harnesses.
Engineering generalists embedded with enterprise clients to ship AI systems against production constraints. The role pioneered at Palantir, now standard at OpenAI, Anthropic, and the leading AI companies.
Engineers who build with foundation models — prompt engineering, fine-tuning, evals, guardrails, latency optimization, and the cost engineering required for production LLM deployments at scale.
Platform engineers who run the infrastructure ML lives on — feature stores, training infra, model registries, deployment pipelines, observability for both classical ML and LLM workloads.
Specialized researchers and engineers for vision (multimodal models, OCR, visual search, object detection) and natural language (entity extraction, classification, document AI).
Senior data engineers (Snowflake, Databricks, dbt, Airflow) and AI architects who design the cross-system architecture for enterprise AI programs spanning multiple platforms.
How we hire
Most staffing firms run keyword searches for "TensorFlow" and "LLM" and forward résumés. We don't.
Minimum 5 years for ML engineers, 7+ for AI architects. No junior submissions for senior architecture work.
100% US-based candidates for senior AI roles. No offshore submissions for production architecture.
Screened by people who've actually shipped production AI — not generic IT recruiters running keyword searches.
Verified production experience on the specific stack the role requires. Real prior-manager calls, not LinkedIn endorsements.
Why we hire differently
The most underrated thing about hiring AI talent in 2026: the engineers worth hiring are the ones who use AI every day — not just the ones who can build it. Most staffing firms can't tell the difference. We can, because we use it ourselves.
We've placed Forward Deployed Engineers onto AI teams where, when we briefed the candidate, we knew more about the modern AI stack than they did. They could train a transformer. They couldn't tell you why your product team should be running a Claude agent for a workflow they were trying to build. There is a real, growing gap between "I shipped an ML model" and "I am an active operator of modern AI in my own work." Most résumés don't disambiguate it. Most recruiters can't either.
When you've built MCP servers and shipped production agents, you can spot an AI engineer who hasn't. We screen accordingly.
Failure modes we see
Pattern recognition from the AI hires we've watched succeed and fail across enterprise teams.
Companies try to ship LLM features on top of data that isn't ready. The model is fine. The data is missing rows, half-formatted, or living in three systems with no canonical merge. We watch teams hire an AI engineer to fix what's actually a data infrastructure problem — and then wonder why six months later the project hasn't shipped. Get the data layer right before you hire the model layer; sometimes the right hire is a senior data engineer first, AI engineer second.
The biggest risk in 2026 is not doing anything. Companies are still debating whether to "do AI" while their competitors are shipping their fourth agentic workflow. AI compounds — every quarter you delay, the gap between you and the teams that started 18 months ago widens. We watch enterprise programs lose six months to procurement loops that should have been a two-week conversation. The teams that win in this market aren't the smartest; they're the ones who started.
The AI engineers worth hiring are operating on a different cadence than enterprise hiring processes. Eight-week interview loops, quarterly headcount approvals, three-round panel screens — the candidates worth hiring are getting two competing offers in the time it takes you to schedule round two. Companies that win in the AI talent market move in days, not weeks. Slow processes are a quiet selection bias against the best engineers — they self-select out before you know they were available.
Cross-platform expertise
A staffing firm that only knows one part of the AI stack can't help when your Marketing Cloud needs to talk to Databricks, or your Salesforce Einstein team needs to integrate with Anthropic Claude.
Focus GTS staffs senior talent across the platforms that actually run enterprise AI: Adobe AI products (Firefly, GenStudio, Brand Concierge, LLM Optimizer), Salesforce Einstein and Agentforce, Databricks (ML, feature stores, AI Functions), Snowflake Cortex, and the foundation-model providers (OpenAI, Anthropic Claude, Google Gemini). Most production AI programs span three or more of these — we can staff specialists across the stack without piecing together separate staffing relationships.
Ready to hire
Already past the guide and working on a live requisition? Bring us the outcome you need and we will help you define the talent. Contract, contract to hire, full time and executive search across AI, ML and data roles.
If you have filled in the one page role brief from the playbook, bring that. It is the fastest way for us to get to a shortlist that matches the work rather than the job title.
Tell us a little about the role. We'll route you to the right specialist on our AI team, usually within one business day.
Frequently asked
Real answers to questions hiring managers ask us most often.
It is free. Fourteen pages covering six AI engineering role profiles, five expensive hiring mistakes, structured interview questions with what to listen for, a bounded work sample, a weighted candidate scorecard, engagement model comparisons, and a one page role brief. The scorecard and the role brief are fillable, so your team can complete them on screen. Give us a work email and a company name and you can download it immediately; we also email you a copy.
Forward Deployed Engineer, Agentic AI Engineer, ML Engineer, Data Scientist, Data Engineer, and MLOps or AI Platform Engineer. Each has its own primary outcome and its own evidence to request, because the same AI label is attached to very different jobs. Those are also the roles we staff, along with LLM and foundation model engineers, computer vision and NLP engineers, and AI architects.
Use the hiring form in the Ready to hire section. Tell us the role and your company and we route you to the right specialist on our AI team, usually within one business day. If you have already filled in the one page role brief from the playbook, bring it: it is the fastest route to a shortlist built around the work instead of the job title.
Data Scientists focus on extracting insight from data — statistical analysis, experimentation design, and model prototyping. ML Engineers productionize machine learning systems — training pipelines, model serving, monitoring, and the engineering rigor that takes a notebook to live infrastructure. AI Engineers (a more recent category) typically build with foundation models — prompt engineering, RAG, agentic workflows, and LLM-powered application engineering. Many roles blend the categories; we screen for the actual production work the candidate has shipped, not the title.
Forward Deployed Engineers (FDEs) are engineering generalists embedded directly with enterprise clients to ship AI/ML systems against real production constraints. The role originated at Palantir and has expanded across the AI ecosystem (OpenAI, Anthropic, and others now staff FDEs). They combine ML engineering, software engineering, and customer-facing communication. Demand for senior FDEs is high; supply is thin — we maintain a network specifically for this role.
Yes. The agentic AI and LLM engineering category has emerged as its own discipline since 2023 — building with frameworks like LangChain, LlamaIndex, Anthropic's Claude SDK, and OpenAI's Agents SDK; designing tool-use patterns, retrieval-augmented generation (RAG) systems, evaluation harnesses, and production agent orchestration. We actively place agentic engineers and LLM application developers for enterprise teams shipping these systems into production.
Yes. We staff AI/ML roles across all engagement models. Contract is the fastest path; contract-to-hire works well for teams who want to validate fit before converting; full-time and executive search take longer because compensation alignment for senior AI talent is the bottleneck — the AI labor market has compressed significantly since 2023.
Yes. 100% of our actively-marketed AI candidates are US-based. We don't present offshore-only candidates for senior AI architecture or ML engineering work. Production AI systems require collaboration with internal product, security, and infrastructure teams in real time — offshore-only engagements consistently produce drift, missed context, and rebuilds.
Every AI candidate is technical-screened by people who have actually shipped production AI systems — not generic IT recruiters running keyword searches for "TensorFlow" or "LLM". We verify production experience on the specific stack the role needs (e.g., training infra, model serving, agentic frameworks), validate against actual project work, run reference checks with prior managers, and only present senior candidates with real production track records.
Yes. We staff specialists for the broader enterprise AI stack: Adobe AI products (Firefly, GenStudio, Brand Concierge, LLM Optimizer), Salesforce Einstein and Agentforce, Databricks ML and feature stores, Snowflake Cortex, and the standalone foundation-model providers (OpenAI, Anthropic, Google). Most enterprise AI programs span multiple platforms; we can staff specialists across the stack.
Senior AI/ML roles typically take longer to fill than other tech disciplines because the market is supply-constrained. Contract placements move fastest — often 2–4 weeks for senior ML engineers when the client process is decisive. Full-time AI architect placements run 6–12 weeks because compensation alignment and technical interview cycles are extensive. The Forward Deployed Engineer category specifically runs longer because the candidate pool is small.
Senior US-based candidates only. Vetted by people who've shipped production AI. Tell us what you need.
