The AI roles we staff — in plain English
Eighteen specialists make up a modern AI crew. You almost never need all of them on day one. This page explains what each role actually does, when you need it, and how the pieces fit together — written for leaders, not engineers.
The crew
18 AI roles, explained
These are the exact role titles used inside every project blueprint and custom build we staff. Rates and hours per role are visible in the configurator.
Project Manager
Runs the launch — sprints, status, risks, and deadlines.
Think of them as: The traffic controller who keeps everyone moving in the same direction on the same calendar.
What they're responsible for
- Plans sprints, runs standups, tracks dependencies
- Owns status reporting to your stakeholders
- Manages scope changes and risk mitigation
- Coordinates handoffs between specialists
When you need this role
Every project. The bigger the crew, the more this role pays for itself.
AI Product Owner / Program Lead
Decides what to build, for whom, and what success looks like.
Think of them as: The director of the project — turns a business problem into a clear AI use case with measurable ROI.
What they're responsible for
- Owns the use case, ROI model, and acceptance criteria
- Prioritizes the roadmap and decides what ships first
- Aligns engineers, stakeholders, and end users
- Drives delivery cadence sprint over sprint
When you need this role
Day one of every initiative. Without it, you get impressive demos that don't move the business.
AI Solution Architect
Designs the end-to-end blueprint — models, data, tools, and how it all fits your stack.
Think of them as: The architect drawing the building before the contractors start. Saves you from expensive rework.
What they're responsible for
- Designs the target architecture and data flow
- Picks the model + tool pattern that fits the problem
- Ensures enterprise fit (security, scale, integrations)
- Reviews engineering decisions against the long-term plan
When you need this role
Any project touching multiple systems, or anything that has to live inside an enterprise stack.
LLM / Agent Engineer
Builds the AI brain — prompts, agents, tool calls, and structured outputs.
Think of them as: The specialist who turns 'we want AI to do X' into a working, reliable AI workflow.
What they're responsible for
- Designs and ships prompts and agent workflows
- Wires the model to tools, APIs, and internal systems
- Enforces structured outputs the rest of the app can trust
- Tunes for accuracy, latency, and token cost
When you need this role
Nearly every modern AI build. If the AI has to think or act, this role does it.
Integration Engineer
Connects the AI to the systems your business already runs on.
Think of them as: The plumber wiring the AI into your CRM, HRIS, ticketing, ERP, contact center, and auth.
What they're responsible for
- Builds API integrations to CRM/ATS/ERP/HRIS/ITSM/CCaaS
- Handles authentication, permissions, and event flows
- Maintains stable contracts between systems as they evolve
- Troubleshoots when an upstream system breaks the AI
When you need this role
Any time the AI needs to read from or write to a system of record — which is almost always.
Data / Knowledge Engineer (RAG)
Teaches a generic AI model your company's specific information.
Think of them as: The librarian who connects your documents, wikis, and policies so the AI answers from your business, not the open web.
What they're responsible for
- Builds corpora, metadata, and embedding pipelines
- Owns the RAG retrieval layer and data quality
- Enforces permissions so users only see what they should
- Keeps the knowledge fresh as content changes
When you need this role
Whenever the AI must answer using your own data — support, sales enablement, internal help desks, compliance.
UX / Conversation Designer
Shapes how users actually talk to (and trust) the AI.
Think of them as: Part writer, part designer — figures out the flow, tone, and fallback when the AI doesn't know.
What they're responsible for
- Designs user journeys and conversation flows
- Sets tone of voice and brand-safe language
- Designs fallbacks, error states, and human handoff
- Tests usability with real users before launch
When you need this role
Anywhere humans interact with the AI directly — chat, voice, copilots, agent assist.
QA / Evals Analyst
Makes sure the AI actually works — and keeps working as it changes.
Think of them as: Traditional QA, but for models where the same input can give different outputs. Builds the scorecards the AI is graded on.
What they're responsible for
- Builds evaluation sets the model is scored against
- Tests for hallucinations, bias, and unsafe responses
- Validates accuracy, latency, and cost before each release
- Catches regressions when prompts or models change
When you need this role
Before launch and continuously after. Skipping this is the #1 reason AI projects fail in production.
Security / Privacy Architect
Protects the model, the data behind it, and the users in front of it.
Think of them as: Your security team's AI specialist — knows the new attack patterns AI introduces.
What they're responsible for
- Designs identity, DLP, secrets, and permissions
- Tests for prompt injection and data leakage
- Owns AI-specific red-teaming and audit trail
- Pairs with your existing security team for sign-off
When you need this role
Any production AI handling customer data or company-confidential information.
DevOps / MLOps Engineer
Keeps the AI infrastructure running smoothly in production.
Think of them as: The operations engineer for AI — handles deploys, monitoring, scaling, and the 2am pages.
What they're responsible for
- Builds deploy pipelines and environments
- Monitors uptime, latency, error rates, and spend
- Scales infrastructure when traffic spikes
- Owns rollback and incident response
When you need this role
Once AI is live with real users. Often starts inside the LLM Engineer's role, then splits out as you scale.
Business SME
The subject-matter expert who validates the AI reflects how the work actually gets done.
Think of them as: The senior person from operations, sales, HR, finance, or clinical who knows the edge cases the AI must respect.
What they're responsible for
- Validates workflows match real-world reality
- Catches policy nuance and escalation rules
- Flags edge cases the engineering team would miss
- Owns acceptance from the business side
When you need this role
Every project. Lent from your team or staffed by us when needed.
Change / Training Lead
Makes sure your people actually adopt the AI you just built.
Think of them as: The internal launch lead — the difference between a tool that sits unused and one that changes how work gets done.
What they're responsible for
- Builds the adoption plan and training program
- Owns internal communications and rollout cadence
- Enables managers to coach their teams on the new way
- Tracks adoption metrics post-launch
When you need this role
Whenever the AI changes how a team does its job. Skipping this kills ROI more often than any tech issue.
Legal / Compliance Advisor
Keeps the AI program compliant, ethical, and audit-ready.
Think of them as: The translator between regulations, internal policy, and the engineering team.
What they're responsible for
- Maps AI use cases to applicable laws and policies
- Decides what data can be used and what must be disclosed
- Documents decisions for auditors and regulators
- Owns the AI risk register and review process
When you need this role
Early in regulated industries — finance, healthcare, legal, HR. Everyone else before scaling past pilot.
Voice / Telephony Engineer
Builds AI that picks up the phone — speaks, listens, and routes calls.
Think of them as: The specialist who makes AI sound human on a real call and integrates with your contact center.
What they're responsible for
- Implements ASR (speech-to-text) and TTS (text-to-speech)
- Designs IVR flows and CCaaS routing
- Handles recording, transcription, and call controls
- Tunes for latency and barge-in so calls feel natural
When you need this role
Voicebots, agent assist, AI receptionists, outbound calling, or any AI inside a contact center.
Data Scientist / Optimization Scientist
Builds the predictive and optimization models behind smarter decisions.
Think of them as: The math brain — forecasting, scoring, ranking, A/B testing, causal analysis.
What they're responsible for
- Builds forecasting, scoring, and optimization models
- Runs experiments to prove what actually works
- Performs causal analysis to separate signal from noise
- Hands trained models to engineering for production
When you need this role
Anything involving predictions, pricing, ranking, demand, churn, or optimization — not just chat.
Computer Vision Engineer
Builds AI that sees — images, video, documents, and live camera feeds.
Think of them as: The specialist for anything visual: detecting damage in photos, reading forms, watching a production line.
What they're responsible for
- Designs image and video processing pipelines
- Handles labeling, training, and detection accuracy
- Optimizes for edge devices or cloud inference
- Integrates vision outputs into downstream workflows
When you need this role
Document understanding, quality inspection, identity/KYC, claims, retail analytics, safety monitoring.
BI / Analytics Engineer
Turns AI activity into business metrics leadership can actually use.
Think of them as: The analyst who builds the dashboards proving the AI is (or isn't) earning its keep.
What they're responsible for
- Designs the semantic layer and KPI definitions
- Builds executive dashboards on AI usage and outcomes
- Owns data lineage so numbers are trustworthy
- Connects AI activity to revenue, cost, and CX metrics
When you need this role
As soon as leadership asks 'is this working?' — usually within 60 days of launch.
Technical Writer / Knowledge Curator
Writes the source material the AI (and your team) rely on.
Think of them as: The editor who turns scattered tribal knowledge into clean, citable content the AI can use.
What they're responsible for
- Authors SOPs, FAQs, and policy content
- Curates source citations the AI quotes from
- Maintains playbooks as products and policy change
- Partners with the RAG engineer on what to index
When you need this role
Any RAG or knowledge-assistant project. Garbage in → garbage out — this role prevents that.
How to combine them
You don't need all eighteen. Start with the right four or five.
Most successful AI initiatives launch with a small crew matched to a clearly scoped problem, then add specialists as the work demands. Here are the starting shapes we see most often.
Quick internal pilot
Product Owner + LLM Engineer + part-time PM
Prove a use case in 4–8 weeks with the smallest crew possible.
Customer-facing knowledge assistant
Product Owner + LLM Engineer + Data/Knowledge (RAG) + UX Designer + QA
AI that answers using your documents and is safe to put in front of customers.
Workflow automation
Product Owner + Solution Architect + LLM/Agent Engineer + Integration Engineer + QA
AI that takes action across systems, not just answers questions.
Regulated industry rollout
Add Security/Privacy Architect + Legal/Compliance + Change Lead before launch
Finance, healthcare, legal, and HR need oversight built in, not bolted on.
Not sure which roles your project needs?
Pick a project blueprint and our configurator suggests the right mix of roles, hours, and bill rates across four delivery regions — in under a minute.