AI Architect Job Description Templates for Small and Growing Teams
6 templates covering the general architect, generative AI, ML platform, customer-facing solutions, the first AI hire, and a fractional contract scope. Download as DOCX.
Almost every AI architect job description online was written by a company that already has an ML platform team, a data engineering group, and a governance committee. It assumes the architect walks into an organization where somebody else runs the pipelines. If you are a small business hiring one person to decide how AI fits your product, copying that posting sets an expectation you cannot meet, and the candidate finds out in week two.
There is also no official version of this job to copy from. The Bureau of Labor Statistics does not publish an occupation called AI architect at all, which means there is no standard duty list, no wage figure, and no employment count for the title. Every posting you read is somebody’s interpretation, and most of them are a list of frameworks with no statement of what the person actually decides.
At FirstHR we write hiring templates for companies without a dedicated HR department. The six below cover the general internal architect, a generative AI specialist, an ML platform architect, a customer-facing solutions architect, a first AI hire at a small company, and a fractional contract scope, each with the classification note the generic versions leave out.
TL;DR
An AI architect owns the target-state design of your AI systems: data flow, model strategy, serving, evaluation, cost, and governance. There is no BLS occupation for the title, so benchmark against nearest classifications, which run from $116,580 to $175,140 at the median. The role is almost always exempt. Six templates below, downloadable as DOCX.
What an AI Architect Actually Owns
An AI architect owns decisions, not deliverables: which AI capability gets built, what it must cost per request, what latency it must hold, how it is evaluated, and what happens when the model is wrong. The output is a reference architecture and a set of documented tradeoffs that other engineers build inside.
That definition sounds abstract until you write the posting, at which point it becomes very concrete. The role differs enormously depending on which problem you are hiring against, and the four below produce four different candidate pools. Picking the wrong one is the most common reason an AI search stalls at the offer stage.
You have models but no system
Design problem
Somebody built a prototype that works on a laptop and nobody can say what happens at a thousand requests a minute. The architect is the person who turns that into a design with latency, cost, and failure behavior written down before the money is spent.
You have several teams building the same thing twice
Platform problem
Every team ships its own pipeline, its own deployment path, and its own monitoring. The architect defines the shared foundation and the paved path, which is a platform role wearing an architect title.
Your deals die in technical evaluation
Customer problem
Prospects want to know how the AI works, where their data goes, and what happens when it is wrong. That is a customer-facing solutions architect, judged on technical win rate rather than on internal system design.
You have nothing yet and one person to hire
First-hire problem
A small company hiring one person needs design judgment and working hands in the same body. Write the posting for a builder who thinks in architecture, not for a specialist who expects a platform team underneath them.
Write the Problem, Not the Technology List
The weakest AI postings are a list of frameworks with no statement of what the person decides. Senior architects read that as a company that has not worked out what it wants, and they have other options. Replace the list with three sentences: what exists today, what the architect will own, and the constraint that makes the problem interesting. A ranked set of two or three depth areas filters better than fifteen technology names, because a candidate can tell whether they qualify.
AI Architect vs the Adjacent Titles
The clearest way to choose a title is by what the role owns rather than by seniority. AI architect sits between several established titles, and each neighbor already has a candidate pool that reads postings differently, so a mistitled search reaches the wrong people.
Title
What it owns
Hire it when
AI architect
Target-state design across AI systems
Multiple teams or projects need one coherent design
AI engineer
Building and shipping an AI feature
You know what to build and need it built
ML engineer
Model training, serving, and pipelines
The work is modeling and production ML rather than design
Data architect
Data models, storage, and governance
The bottleneck is the data layer, not the AI layer
Solution architect
End-to-end solution for a customer
The design work is external and tied to deals
Cloud architect
Cloud foundation, scale, security, cost
Infrastructure is the constraint rather than the models
If the bottleneck is really the data layer, the data architect templates describe the role you need, and if the work is production modeling rather than system design, the machine learning engineer templates fit better. Where the design work is customer-facing and tied to deals, compare against the solution architect templates before you commit to a title.
What Belongs in the Posting
An AI architect job description does four jobs at once: it explains the situation honestly, it filters out people who cannot do it, it protects you legally, and it closes a candidate who has other offers. Most AI postings do only the third and fourth badly and the first not at all.
The parts a senior engineer reads first
What already exists: stack, data, cloud, team size
What the role owns, stated as a decision right
Whether it is greenfield or a rescue
Who the architect reports to and argues with
The parts that filter applicants
Production ownership, not prototypes
Depth areas ranked, not a list of every buzzword
Hands-on percentage, stated honestly
Whether the role has direct reports
The parts that protect you
FLSA classification stated on the posting
Essential functions written plainly
Data handling and confidentiality expectations
Equal opportunity statement
The parts that win the hire
A published salary range and equity terms
The constraint that makes the problem interesting
Decision speed and who signs off
A named person and a real deadline to apply
The single most common omission is the hands-on percentage. An architect who expects to be purely advisory and lands in a company where they deploy their own code leaves within a year, and the reverse mismatch is just as expensive. State it as a number. Our guide to writing a job description covers the general structure, and the whole hiring template library follows the same skeleton.
6 AI Architect Job Description Templates to Download
Download all six as one file or copy them individually. Each follows the same structure: company context, position summary, key responsibilities, required and preferred qualifications, a classification note, an equal opportunity statement, and how to apply. The bracketed fields are the only parts you need to change.
Download All 6 AI Architect Job Description Templates
General, generative AI, ML platform, customer-facing solutions, first AI hire, and fractional contract. All in one download.
AI Architect (General)
Target-state design
The baseline internal role: reference architecture, build-versus-buy decisions, integration patterns, evaluation standards, and governance.
Generative AI / LLM Architect
Retrieval, evals, guardrails
For companies past the demo stage on language models, with retrieval design, model routing, evaluation harness, and a real cost model.
ML Platform Architect
The paved path
For teams where every data scientist builds their own pipeline. Registry, serving, drift monitoring, reproducibility, and infrastructure cost.
AI Solutions Architect
Customer-facing
For technical evaluations and pilots, with security questionnaires, scoping, handoff to delivery, and a shared revenue target.
First AI Architect
Half design, half build
For a small company hiring one person. Picks the stack, ships the first system, sets minimum viable governance, and writes it down.
Fractional / Contract
Scoped by deliverable
For architectural judgment without a full-time hire, written as a scope of work with the misclassification warning stated plainly.
Template 1: AI Architect (General)
The baseline internal role, with reference architecture, build-versus-buy decisions, integration patterns, evaluation standards, and model governance written in.
AI Architect Job Description (General)
AI ARCHITECT JOB DESCRIPTION
Company: __ ([City, State] / Remote)
Reports to: [CTO / VP Engineering / Head of Data]
Employment type: Full-time, W-2
FLSA status: Exempt (see classification note)
Compensation: $_ to $_ per year, plus [bonus / equity]
ABOUT [COMPANY NAME]
[Company Name] is a [industry] company in [City, State] with [team size]
employees and [product or customer detail]. We already run [current systems:
data warehouse, application stack, cloud provider], and we are hiring an AI
Architect to design how AI capability fits into that estate rather than beside
it.
POSITION SUMMARY
The AI Architect owns the target-state design for our AI systems: how data
reaches models, how models are served and monitored, how AI features integrate
with our existing applications, and which build-or-buy decisions we make. The
role is a technical authority, not a people-management track.
KEY RESPONSIBILITIES
•Own the reference architecture for AI and machine learning systems end to end:
data ingestion, feature and context stores, training or fine-tuning, serving,
evaluation, and monitoring
•Translate business objectives into a sequenced technical roadmap with cost,
latency, and accuracy targets stated up front
•Make and document build-versus-buy decisions across models, vector stores,
orchestration, and observability, including total cost of ownership
•Define integration patterns between AI services and our existing [application
stack / data warehouse / identity provider]
•Set evaluation standards: offline benchmarks, human review, regression checks,
and the acceptance bar a feature must clear before release
•Establish model and data governance: lineage, access control, retention,
approved data sources, and the review path for a new use case
•Review designs from engineering and data teams and mentor engineers on
architectural reasoning
•Partner with [security / legal / compliance] on risk review for each release
REQUIRED QUALIFICATIONS
•[Number] years building and running production software, with [number] years on
machine learning or AI systems specifically
•Demonstrated ownership of an AI or ML system in production, not a prototype:
tell us what it served, at what scale, and what broke
•Depth in at least two of: data engineering, ML platform and MLOps, applied
modeling, retrieval and search, cloud infrastructure
•Fluency in Python and at least one cloud provider's AI and data services
•Written communication strong enough that a design document settles an argument
•Bachelor's degree in computer science or equivalent practical experience
PREFERRED QUALIFICATIONS
•Experience in [our industry] or with [regulated data type]
•Experience with model risk documentation or an AI governance framework
•Experience choosing between hosted models and self-hosted inference on cost
CLASSIFICATION NOTE (read before posting)
An AI Architect at this level is normally exempt from overtime, most often under
the computer employee exemption or the learned professional exemption. The
computer employee exemption requires payment on a salary or fee basis of at
least $684 per week, or an hourly rate of at least $27.63, plus duties centered
on systems analysis, design, development, or documentation. Classification turns
on actual duties, not on the title. This is general information, not legal
advice.
EEO STATEMENT
[Company Name] is an equal opportunity employer and provides reasonable
accommodations for the essential functions of this role.
COMPENSATION AND HOW TO APPLY
Compensation: $_ to $_ per year, [equity], [benefits summary]
To apply, email __ with your resume and one architecture
document you wrote that you are willing to discuss.
Template 2: Generative AI / LLM Architect
For companies past the demo stage on language models: retrieval design, model routing, an evaluation harness, guardrails, and a cost model that survives real volume. If the role is building rather than designing, the AI engineer templates are the closer fit.
Generative AI / LLM Architect Job Description
GENERATIVE AI / LLM ARCHITECT JOB DESCRIPTION
Company: __ ([City, State] / Remote)
Reports to: [CTO / VP Engineering / Head of Product]
Employment type: Full-time, W-2
FLSA status: Exempt (see classification note)
Compensation: $_ to $_ per year, plus [bonus / equity]
ABOUT THIS ROLE
[Company Name] is building [product feature] on top of large language models.
We have moved past the demo stage and need someone to design the system that
makes it reliable, affordable, and defensible: retrieval, evaluation, guardrails,
and the cost model underneath all three.
POSITION SUMMARY
The Generative AI Architect designs our LLM application architecture end to end:
retrieval and context assembly, prompt and tool orchestration, model selection
and routing, evaluation, guardrails, and the observability that tells us when
quality drifts.
KEY RESPONSIBILITIES
•Design retrieval architecture: chunking, embeddings, indexing, ranking, and
freshness, with measurable retrieval quality targets
•Define model strategy: which tasks use which model, when to route to a smaller
model, and when fine-tuning beats prompting on cost and quality
•Design agent and tool-calling patterns with explicit failure and fallback paths
•Build the evaluation harness: golden sets, automated scoring, human review
workflow, and regression gates in the release pipeline
•Design guardrails for prompt injection, data leakage, unsafe output, and
hallucination, and document what each control does and does not catch
•Own the cost model: tokens per request, caching strategy, context budget, and
the unit economics of each feature at projected volume
•Set data handling rules for customer content sent to third-party model
providers, including retention and training opt-out terms
•Partner with product on what "good enough to ship" means in measurable terms
REQUIRED QUALIFICATIONS
•[Number] years in software or ML engineering, including hands-on work shipping
an LLM-backed feature to real users
•Practical command of retrieval-augmented generation beyond a tutorial: you can
explain why your retrieval failed and what you changed
•Experience designing evaluation for systems whose output is not deterministic
•Working knowledge of inference cost, latency tradeoffs, and context limits
•Python, plus experience with at least one orchestration or serving framework
•Clear written reasoning about tradeoffs under uncertainty
PREFERRED QUALIFICATIONS
•Experience with self-hosted inference and the cost case for it
•Experience with [our industry] data sensitivity or regulatory review
•Experience running a red-team exercise against an LLM feature
CLASSIFICATION NOTE
This role is normally exempt under the computer employee or learned professional
exemption. Classification depends on actual duties. If the work is primarily
operating a tool rather than designing systems, review the classification before
posting. This is general information, not legal advice.
EEO STATEMENT
[Company Name] is an equal opportunity employer and provides reasonable
accommodations for the essential functions of this role.
COMPENSATION AND HOW TO APPLY
Compensation: $_ to $_ per year, [equity], [benefits summary]
To apply, email __ with your resume and a short note on an
LLM system you designed, including what it cost to run.
Still Using Spreadsheets for Onboarding?
Automate documents, training assignments, task management, and track onboarding progress in real time.
For teams where every data scientist builds their own pipeline. Registry, serving, drift monitoring, reproducibility, CI/CD gates, and infrastructure cost ownership.
Reports to: [VP Engineering / Head of Platform / CTO]
Employment type: Full-time, W-2
FLSA status: Exempt (see classification note)
Compensation: $_ to $_ per year, plus [bonus / equity]
ABOUT THIS ROLE
[Company Name] has [number] data scientists and ML engineers who each build
their own pipeline, their own deployment path, and their own monitoring. That
does not scale. We are hiring a Platform Architect to design the shared
foundation they all build on.
POSITION SUMMARY
The ML Platform Architect designs the internal platform for training, deploying,
serving, and monitoring models: the pipelines, the registry, the serving layer,
the feature store, and the paved path that makes the correct way the easy way.
KEY RESPONSIBILITIES
•Design the end-to-end ML lifecycle platform: data access, feature computation,
training orchestration, model registry, deployment, and rollback
•Define the serving architecture for batch, real-time, and streaming inference
with stated latency and availability targets
•Design monitoring for data drift, model drift, and prediction quality, with
alert thresholds that an on-call engineer can act on
•Set reproducibility standards: versioned data, versioned code, versioned model
artifacts, and a documented lineage for anything in production
•Design the CI/CD path for models, including automated evaluation gates and a
documented rollback procedure
•Own infrastructure cost: training spend, GPU utilization, inference cost per
prediction, and the tradeoffs between them
•Define access controls and audit logging for training data and model artifacts
•Work with data science to make the platform something they choose rather than
something they route around
REQUIRED QUALIFICATIONS
•[Number] years in platform, infrastructure, or ML engineering, with [number]
years designing ML infrastructure specifically
•Production experience with container orchestration and infrastructure as code
•Experience with an ML lifecycle or orchestration toolchain and a clear opinion
on what you would use again
•Strong understanding of distributed systems, storage, and cost at scale
•Python plus at least one systems language
•Track record of internal tools that engineers actually adopted
PREFERRED QUALIFICATIONS
•Experience with GPU capacity planning and scheduling
•Experience migrating a team off bespoke pipelines onto a shared platform
•Experience with feature store design and its failure modes
CLASSIFICATION NOTE
Normally exempt under the computer employee exemption, which covers systems
analysis, design, and development duties and requires payment on a salary or fee
basis of at least $684 per week, or an hourly rate of at least $27.63. Confirm
against actual duties. This is general information, not legal advice.
EEO STATEMENT
[Company Name] is an equal opportunity employer and provides reasonable
accommodations for the essential functions of this role.
COMPENSATION AND HOW TO APPLY
Compensation: $_ to $_ per year, [equity], [benefits summary]
To apply, email __ with your resume and a description of a
platform you designed and who used it.
Template 4: AI Solutions Architect (Customer-Facing)
For technical evaluations and pilots, with security questionnaires, scoping, a written handoff to delivery, and a shared revenue target stated up front.
AI Solutions Architect Job Description (Customer-Facing)
AI SOLUTIONS ARCHITECT JOB DESCRIPTION (CUSTOMER-FACING)
Company: __ ([City, State] / Remote)
Reports to: [VP Sales Engineering / Head of Customer Engineering / CTO]
Employment type: Full-time, W-2
FLSA status: Exempt (see classification note)
Compensation: $_ base plus $_ variable, [equity]
ABOUT THIS ROLE
[Company Name] sells [AI product or service] to [customer type]. Deals stall in
the technical evaluation, not the commercial one. We are hiring an AI Solutions
Architect to sit in those conversations, design the customer's implementation,
and stay accountable through go-live.
POSITION SUMMARY
The AI Solutions Architect designs customer-specific AI implementations, leads
technical evaluations and pilots, and translates between what the customer needs
and what our platform does. The role is externally facing and carries a shared
revenue target.
KEY RESPONSIBILITIES
•Lead technical discovery with prospects and map their use case to a concrete
architecture, including data flow and integration points
•Design and run proofs of concept with defined success criteria agreed in
writing before the pilot starts
•Answer security, privacy, and model-governance questionnaires, and represent
our controls accurately to customer security teams
•Scope implementations: effort, sequence, dependencies, and what the customer
must provide
•Hand off to delivery or customer success with a written architecture and open
risks named
•Feed recurring customer blockers back to product and engineering as ranked
requests, not anecdotes
•Support [number] active opportunities with [sales team size] sellers
•Travel [percentage] for onsite workshops and executive briefings
REQUIRED QUALIFICATIONS
•[Number] years in solutions architecture, sales engineering, or consulting,
with hands-on AI or machine learning delivery
•Ability to design a working architecture on a call and defend it to an engineer
•Experience with enterprise integration: identity, data residency, APIs, and
private networking
•Comfort with security and compliance review cycles
•Presentation skills in front of both engineers and executives
•Willingness to carry a number alongside the sales team
PREFERRED QUALIFICATIONS
•Experience selling into [our target industry]
•Cloud provider architecture certification
•Prior ownership of a technical win rate metric
CLASSIFICATION NOTE
Customer-facing architects are normally exempt, but the analysis is less
automatic than for an internal architect: review whether the primary duty is
systems design (computer employee exemption), advanced knowledge work (learned
professional), or sales. Commission structure does not by itself decide the
question. Confirm before posting. This is general information, not legal advice.
EEO STATEMENT
[Company Name] is an equal opportunity employer and provides reasonable
accommodations for the essential functions of this role.
COMPENSATION AND HOW TO APPLY
Compensation: $_ base, $_ variable at target, [equity], [benefits]
To apply, email __ with your resume and one deal you won on
technical merit.
Template 5: First AI Architect (Small Business or Startup)
For a small company hiring one person: picks the stack, ships the first system, sets minimum viable governance, and writes enough down that the next two hires can follow. Pair it with the AI product manager templates if the second hire is on the product side.
First AI Architect Job Description (Small Business / Startup)
FIRST AI ARCHITECT JOB DESCRIPTION (SMALL BUSINESS / STARTUP)
Company: __ ([City, State] / Remote)
Reports to: [Founder / CTO]
Employment type: Full-time, W-2
FLSA status: Exempt (see classification note)
Compensation: $_ to $_ per year, plus [equity: __%]
ABOUT THIS ROLE
[Company Name] is a [team size] person company. We do not have an ML platform, a
data team, or an AI governance committee, and we are not pretending otherwise.
We have [describe the actual data and systems] and a real problem worth solving
with AI. You would be the first person here whose job is to decide how.
POSITION SUMMARY
The first AI Architect sets our AI direction and then builds it. Expect roughly
half design and half hands-on implementation. You will pick the stack, ship the
first system, and write down enough for the next two hires to follow.
KEY RESPONSIBILITIES
•Choose the initial AI stack with a written rationale, biased toward managed
services over anything we would have to operate ourselves
•Build the first production AI capability end to end and own it in production
•Establish the data foundation the AI work needs: sources, quality, access, and
a retention rule we can defend to a customer
•Define what success means before building: the metric, the baseline, and the
threshold at which we stop
•Set the minimum viable governance: approved data sources, human review where
the output touches a customer, and an incident path when it goes wrong
•Write short architecture notes so decisions survive your vacation
•Advise the founder honestly on where AI is not the answer
•Interview and onboard the next [number] technical hires
REQUIRED QUALIFICATIONS
•[Number] years shipping production software, including at least one AI or ML
system you personally put in front of users
•Comfort operating without a platform team: you deploy, you monitor, you page
•Judgment about scope: you can name three things you deliberately did not build
•Python and a cloud provider you know well enough to estimate the bill
•Direct communication with non-technical stakeholders about risk and cost
PREFERRED QUALIFICATIONS
•Prior first or founding technical hire experience
•Experience in [our industry]
•Experience keeping an AI feature under a fixed monthly budget
CLASSIFICATION NOTE
Normally exempt under the computer employee exemption when the primary duty is
systems analysis, design, and development, at a salary or fee basis of at least
$684 per week or an hourly rate of at least $27.63. Equity does not count toward
the salary basis. If the role drifts into general IT support, revisit the
classification. This is general information, not legal advice.
EEO STATEMENT
[Company Name] is an equal opportunity employer and provides reasonable
accommodations for the essential functions of this role.
COMPENSATION AND HOW TO APPLY
Compensation: $_ to $_ per year, __% equity over [vesting], and
[benefits summary]
To apply, email __ with your resume and the AI project you
are proudest of, including what it cost and what you would do differently.
Template 6: Fractional / Contract AI Architect
For architectural judgment without a full-time hire, written as a scope of work with deliverables rather than duties, and with the misclassification warning stated plainly rather than buried.
Fractional / Contract AI Architect Job Description (1099)
FRACTIONAL / CONTRACT AI ARCHITECT SCOPE AND JOB DESCRIPTION (1099)
Company: __ ([City, State] / Remote)
Engagement owner: [Founder / CTO]
Engagement type: Independent contractor, [number] hours per [week / month]
Term: [Number] months, renewable
Rate: $_ per [hour / month], or $_ fixed fee per deliverable
ABOUT THIS ENGAGEMENT
[Company Name] needs architectural judgment about AI without a full-time hire.
This engagement covers [scope: evaluate our options, design the first system,
review a vendor decision, unblock the team]. It is scoped by deliverable, not by
supervision, because the working relationship is that of an independent
contractor.
SCOPE OF WORK
The contractor will deliver:
•A current-state assessment of our data, systems, and AI readiness
•A written target architecture with two or three sequenced options, each with
cost, risk, and time estimates
•A build-versus-buy recommendation with total cost of ownership at our volume
•An evaluation and monitoring plan for whatever we build
•[Optional] Hands-on implementation of [specific component]
•A written handoff a permanent hire can pick up
WORKING ARRANGEMENT
•The contractor controls how and when the work is performed and uses their own
equipment and tools
•We define the deliverable and the deadline; we do not set working hours or
supervise method
•The contractor may work for other clients during the term
•Invoices are submitted [frequency]; payment terms are net [number] days
•IP assignment, confidentiality, and data handling terms are in the separate
services agreement
REQUIRED QUALIFICATIONS
•[Number] years designing and shipping production AI or ML systems
•References from at least two comparable engagements
•Ability to produce written deliverables a third party can act on
•Own business entity, insurance, and equipment
CLASSIFICATION WARNING (read before posting)
Do not use this template to convert a full-time role into a contract role. If you
set the schedule, direct the method, supply the equipment, and the work is
integral to your ongoing business, that is an employee under federal and state
tests regardless of what the agreement says, and several states apply a stricter
test than the federal one. Misclassification exposes you to back taxes, unpaid
overtime, penalties, and benefit claims. Use this template only for a genuinely
independent, deliverable-scoped engagement, and have counsel review the services
agreement. This is general information, not legal advice.
EEO AND NONDISCRIMINATION
[Company Name] does not discriminate in the selection of contractors on any
protected basis.
HOW TO SUBMIT
Rate: $_ per [hour / month] or $_ per deliverable
To propose, email __ with your rate, availability, two
references, and a sample deliverable from a prior engagement.
Exempt Status and Classification
An AI architect is almost always exempt from overtime, but you still have to name which exemption applies and confirm the duties independently of the salary. Salary level alone never establishes an exemption, and that misunderstanding is the single most common wage error at small technology companies.
The usual fit is the computer employee exemption, which requires a salary or fee basis of at least $684 per week or an hourly rate of at least $27.63, plus a primary duty of systems analysis, design, development, documentation, testing, or modification. For research-leaning or modeling-heavy roles the learned professional exemption is the better theory, carrying the same weekly salary threshold.
The computer employee exemption has two pay routes
Most AI architect roles are exempt, and the exemption that usually fits is the computer employee exemption under section 13(a)(17) and 29 CFR 541.400. It is unusual because it offers two pay routes: a salary or fee basis of at least $684 per week, or an hourly rate of at least $27.63. Almost no other exemption allows an hourly-paid exempt employee. The duties test still has to be met independently: the primary duty must be systems analysis, design, development, documentation, testing, or modification of computer systems or programs. A senior architect clears that comfortably. Someone whose day is operating a vendor tool, running reports, or troubleshooting laptops does not, whatever the title says. Classification follows the actual work. This is general information, not legal advice.
The learned professional exemption is the fallback
Where the computer employee exemption is awkward (a research-leaning architect, a role heavy on statistical modeling rather than systems design), the learned professional exemption under 29 CFR 541.301 is usually the better fit. It requires work requiring advanced knowledge in a field of science or learning, customarily acquired by a prolonged course of specialized intellectual instruction, plus the standard $684 weekly salary threshold. The practical point for a small employer is that you should pick one theory and be able to state it, rather than assuming that a six-figure salary makes a role exempt by itself. Salary alone never establishes an exemption. Write the classification onto the job description while the duties are fresh in your mind, because reconstructing the reasoning two years later during a wage claim is much harder. This is general information, not legal advice.
Fractional does not mean contractor by default
Fractional AI architects are a real and sensible answer for a company that needs judgment more than headcount, and they are also where small companies most often get misclassification wrong. The federal test looks at economic reality: control over the work, opportunity for profit or loss, investment, permanence, skill, and how integral the work is to your business. Several states apply a stricter standard that presumes employment unless the worker is free from control, performs work outside your usual course of business, and is independently established in that trade. An architect designing the core of your product is arguably inside your usual course of business, which is the clause that catches people. Scope the engagement by deliverable, do not set hours, do not supply the laptop, and have counsel review the services agreement before the first invoice. This is general information, not legal advice.
Governance duties belong in the posting, not in a later memo
An AI architect makes decisions with compliance consequences: which data trains or grounds a model, whether customer content leaves your tenant, what human review sits between a model and a customer, and what you can prove about any of it later. Writing those duties into the job description does two things. It tells senior candidates you are serious, which is a genuine differentiator against postings that read like a list of frameworks. It also gives you a documented owner when a customer security review, an insurer, or a regulator asks who is accountable for model risk. Name the specific artifacts you expect: an inventory of AI use cases, documented data sources, an evaluation record per release, and an incident path. Vague language about responsible AI is not a duty and cannot be assessed in an interview.
Anyone who falls outside both tests needs hours tracked and overtime paid past forty in a week. If you are unsure which side a role lands on, our breakdown of exempt versus non-exempt classification works through the tests in order. Equity is worth saying out loud here: it does not count toward the salary basis, so a package that is mostly equity with a low cash salary can fail the test even when total compensation looks generous.
Governance Duties Worth Naming in the Posting
Write the governance duties into the job description rather than leaving them to a later memo, because they are the duties most likely to be asked about by a customer, an insurer, or a regulator. An AI architect decides which data trains or grounds a model, whether customer content leaves your tenant, and what human review sits between a model and a customer.
The vocabulary is already standardized enough to borrow. The NIST AI Risk Management Framework, published by the National Institute of Standards and Technology, organizes AI risk work into four functions: govern, map, measure, and manage. Using those words in a posting signals to a senior candidate that you have thought about the problem, and it gives you a structure for the interview.
Vague Responsible AI Language Is Not a Duty
A line about a commitment to responsible AI cannot be assessed in an interview and cannot be enforced after the hire. Name the artifacts instead: an inventory of AI use cases with an owner for each, a documented list of approved data sources and what may not be sent to a third-party model provider, an evaluation record attached to each release, a human review step wherever output reaches a customer, and a written incident path when a model produces something harmful. Those are things a candidate can describe having built, and things you can point to later when a customer security review asks who is accountable.
The same logic applies to how you use AI in your own hiring process. If you screen candidates with algorithmic tools, that is a decision with legal exposure of its own, and our overview of AI in HR covers where the risk sits.
What to Pay an AI Architect
There is no BLS wage figure for AI architect, because the Bureau of Labor Statistics does not publish an occupation with that title. Benchmark against the nearest classifications instead, then adjust upward for production AI experience, which is genuinely scarce and priced accordingly in major metros.
The Nearest Classifications, National Medians
According to the Bureau of Labor Statistics Occupational Employment and Wage Statistics survey (May 2025), computer and information research scientists had a national median annual wage of $140,300, with the 25th percentile at $103,570, the 75th at $188,700, and the 90th at $230,630. The computer systems engineers and architects group had a median of $116,580, ranging from $79,370 at the 25th percentile to $188,470 at the 90th (U.S. Bureau of Labor Statistics, OEWS national estimates).
Nearest BLS classification
National median (OEWS, May 2025)
Use it when
Computer and information research scientists
$140,300 per year
The role is research-leaning or invents rather than assembles
Software developers
$135,980 per year
The architect still ships production code most weeks
Computer network architects
$134,050 per year
A reasonable floor for any architect title in infrastructure
Data scientists
$120,230 per year
The work leans modeling and analysis over systems design
Computer systems engineers and architects
$116,580 per year
The closest match by title, reported inside a broad group
Computer and information systems managers
$175,140 per year
The architect carries direct reports and a budget
Two adjustments matter. The classifications above are national and include every industry, so a technology company in a high-cost metro should expect to sit well above the median, often between the 75th and 90th percentiles for the research scientist classification. And the Bureau of Labor Statistics projects much faster than average employment growth through 2034 for computer and information research scientists, software developers, and the computer systems engineers and architects group, which is the demand side of why these searches take longer than they used to. Publish a good-faith range where pay transparency laws apply, and treat the range as a filter rather than an opening position.
Companies Using FirstHR Onboard 3x Faster
Join hundreds of small businesses who transformed their new hire experience.
Small company AI hiring fails in three predictable places: the posting cannot compete for attention, the interview cannot distinguish design ownership from exposure, and the first week is improvised. Each has a fix that does not require headcount.
You are competing for a candidate who has three offers and none of them are from you
Senior AI talent is scarce and the large employers move first with more money, so a small company that tries to win on base salary loses slowly and expensively. Compete on the things a small company actually has: one architect who owns the whole design instead of a fifth of it, a decision that takes a week rather than a quarter, direct access to the founder, real equity, and a problem with a constraint that makes it interesting. Say all of that in the posting rather than saving it for the second interview. Then compress your process: a screen, a technical conversation about a system the candidate actually built, one design exercise on your real problem, and a decision. Every extra round is a chance for a faster employer to close.
You cannot tell whether a candidate designed the system or watched somebody design it
The AI title market is noisy, and a resume full of frameworks tells you nothing about judgment. Skip the trivia and the take-home. Ask the candidate to walk through one AI system they put into production: what it served, at what scale, what it cost to run, what broke, and what they would do differently. People who owned a design answer with specifics and volunteer the failures. People who did not, talk about the technology stack instead of the tradeoffs. Then run one live design exercise on a real constraint from your business, and grade on the questions they ask before drawing anything. Two conversations structured that way separate the two groups more reliably than five rounds of general interviews.
The first week arrives and there is no HR person to run it
A senior technical hire notices a sloppy first week more than anyone, because it is the first evidence of how the company operates. At a small company the paperwork lands on the founder: the offer and signature, confidentiality and IP assignment, equity documents, equipment, cloud and repository access, security training, and the policy acknowledgments. FirstHR was built for exactly this. The onboarding wizard runs the same sequence for every hire, e-signature handles the offer and the IP agreement, document management stores signed acknowledgments against the employee profile, and training modules cover security and data handling before the architect touches production data. Applicant tracking is coming soon to FirstHR. Note that FirstHR is an onboarding and HR platform, not a payroll provider.
Once the offer is signed, the work becomes a repeatable onboarding checklist, and for a technical hire specifically the sequence that matters is access, security training, and data handling rules before the first production commit. Getting that order right is also the first governance decision your new architect will observe you making.
Key Takeaways
An AI architect owns decisions rather than deliverables: what gets built, what it must cost per request, how it is evaluated, and what happens when the model is wrong.
The Bureau of Labor Statistics publishes no occupation called AI architect, so benchmark against the nearest classifications and adjust for scarce production AI experience.
National medians (BLS OEWS, May 2025) run from $116,580 for computer systems engineers and architects to $140,300 for computer and information research scientists and $175,140 when the role carries direct reports.
The role is almost always exempt, usually under the computer employee exemption at $684 per week or $27.63 per hour, but the duties test is independent and salary alone never establishes an exemption.
Name governance duties as concrete artifacts, an AI use case inventory, approved data sources, an evaluation record per release, and an incident path, rather than as a responsible AI slogan.
Write the hands-on percentage as a number, because an architect who expected advisory work and lands in a hands-on company leaves within a year.
A senior technical hire judges you by the first week. FirstHR runs the same onboarding sequence every time, with e-signature for the offer and the IP assignment, document storage for signed acknowledgments, and training modules for security and data handling before anyone touches production. Applicant tracking is coming soon to FirstHR.
Frequently Asked Questions
What does an AI architect do?
An AI architect owns the target-state design of a company’s AI systems: how data reaches models, how models are trained or selected, how they are served and monitored, how AI features integrate with existing applications, and which capabilities get built versus bought. The role is a technical authority rather than a management track, and its output is decisions and documents more than code, though at a small company an architect is usually hands-on half the time. The practical distinction from an AI engineer is scope and time horizon: an engineer builds a system, an architect decides which system gets built, what it must cost per request, what latency it must hold, and what happens when the model is wrong. At a company with fewer than a hundred employees, the honest version of the role is design judgment plus working hands, and the job description should say so.
Is there a BLS occupation code for AI architect?
No. The Bureau of Labor Statistics does not publish a Standard Occupational Classification called AI architect, so there is no official wage or employment figure for the title itself. Benchmark against the nearest classifications instead. According to the Bureau of Labor Statistics Occupational Employment and Wage Statistics survey (May 2025), the national median annual wage was $140,300 for computer and information research scientists, $135,980 for software developers, $134,050 for computer network architects, $120,230 for data scientists, and $116,580 for the computer systems engineers and architects group. Computer and information systems managers, the right comparison when the architect carries budget and direct reports, had a median of $175,140. Pick the classification that matches the actual duties, then adjust for your metro and for the scarcity of production AI experience.
Is an AI architect exempt from overtime?
Yes, in nearly every case, but you still have to name which exemption applies. The computer employee exemption under 29 CFR 541.400 is the usual fit: it requires payment on a salary or fee basis of at least $684 per week, or an hourly rate of at least $27.63, plus a primary duty of systems analysis, design, development, documentation, testing, or modification. The learned professional exemption under 29 CFR 541.301 is the better theory for research-leaning or modeling-heavy roles, and it carries the same $684 weekly salary threshold. Two cautions for small employers. Salary level alone never establishes an exemption; the duties test is independent and controls. And a high total compensation package that is mostly equity does not satisfy the salary basis, because equity is not counted toward it. State law may set a higher threshold than the federal one, so check yours.
What is the difference between an AI architect and an AI engineer?
Scope, time horizon, and decision rights. An AI engineer builds and ships AI features: writing the pipeline, integrating the model, tuning retrieval, and keeping the service healthy. An AI architect decides what should be built and why, sets the reference architecture that engineers work inside, makes build-versus-buy calls, defines evaluation and cost targets, and carries the consequences of those decisions across multiple projects. In practice the split only becomes real at a certain size. Below roughly twenty engineers, one senior person does both, and hiring a pure architect who expects a platform team underneath them is a common and expensive mismatch. Above that, the split pays for itself because the cost of inconsistent designs starts to exceed the cost of the role. Write the posting for the version you actually need today.
When should a small company hire an AI architect?
Hire one when the cost of an inconsistent design has started to exceed the cost of the role, and not before. Three signals matter. First, more than one team is building AI capability and they are solving the same problems differently, which produces duplicated spend and incompatible data handling. Second, an AI feature has reached real customers and you cannot answer basic questions about cost per request, evaluation, or what happens on failure. Third, a customer security review or an insurer has asked who owns model risk and there is no name to give. If none of those apply, a fractional engagement scoped to a written architecture and a build-versus-buy recommendation is usually the better spend, because it produces a decision without a permanent salary. The fractional template on this page is written for that case, with the contractor classification warning included.
What should an AI architect job description include?
Eight things: what already exists (stack, data, cloud, team size), what the role owns stated as a decision right rather than a wish list, the honest hands-on percentage, ranked depth areas instead of a list of every framework, the governance duties you expect by name, the FLSA classification, a published salary range with equity terms, and a named person to apply to with a real deadline. The most common failure in AI postings is a list of technologies with no statement of what the person decides, which attracts volume and no fit. Senior candidates are choosing between offers, so the posting has to answer what they will own, what constraint makes the problem interesting, and how fast you move. The six templates here follow that structure so you only rewrite the company-specific parts.
How do I hire an AI architect without an HR department?
Run a short, structured process and move faster than a large employer can. Publish a specific job description with a real salary range, because senior candidates filter on it. Screen once, then spend the main conversation on a single AI system the candidate put into production: what it served, at what scale, what it cost, what broke, and what they would change. Follow with one live design exercise on a genuine constraint from your business and grade on the questions asked before anything is drawn. Check references with someone who saw them make a decision that turned out wrong. Then decide within a week. Once the offer is signed, run onboarding as a fixed checklist rather than a pile of email: signed offer, confidentiality and IP assignment, equity documents, equipment, access, security training, and policy acknowledgments. Applicant tracking is coming soon to FirstHR.
