The role, decoded

AI Solutions Architect: Designing & Scoping Client AI Systems

What is an AI solutions architect?

An AI solutions architect designs AI systems for someone else's problem. Where an in-house AI architect owns the design for one organisation's own product, a solutions architect works across clients or business units: they sit between the customer and the engineering team, turn a fuzzy business need into a concrete AI architecture, and carry the technical credibility that closes the deal and de-risks the delivery. The title is most common at the hyperscalers and AI labs, and at the systems integrators and consultancies that build on top of them.

The job is genuinely two-sided. One side is architecture — the same agentic-system design, model selection, retrieval, trust boundary, and cost-at-scale work any AI architect does. The other side is the engagement: discovery workshops, requirements, a proposed reference architecture with a defensible rationale, effort and cost estimates, and enough delivery oversight to make sure what ships matches what was promised. A solutions architect who can't do the first half has no credibility; one who can't do the second half is just an architect with a different business card.

AI solutions architect vs AI architect, AI engineer & forward deployed engineer

The whole family overlaps; the useful split is who you serve and how much you build:

• vs AI architect. Same technical design skills. The solutions architect adds client-facing scoping, pre-sales, and stakeholder communication; the in-house architect points those same skills at one product and its rationale. Solutions architect is the outward-facing variant.
• vs AI engineer. The engineer builds and operates the system hands-on. The solutions architect designs it and scopes it, and typically writes less production code — though the best ones still build to stay credible.
• vs forward deployed engineer. Closest cousin. Both are client-facing. The FDE embeds inside the customer and actually builds the solution in their environment; the solutions architect stays more on design, scoping, and advisory across several accounts. FDE is hands-in-the-code; solutions architect is hands-on-the-whiteboard.

For the full side-by-side across the entire family, see the AI roles, decoded.

What an AI solutions architect actually does

A typical engagement runs through a repeatable arc, and the architect owns the technical spine of all of it:

• Discovery and scoping. Workshops to extract the real requirement behind the ask, the constraints (data residency, compliance, budget), and the definition of success.
• Reference architecture. The agentic-system design — the loop and its bounds, model selection and economics, where retrieval sits, the tool and trust boundary, and the deployment platform — plus the written rationale a client's own engineers will scrutinise.
• Platform and cost justification. Which platform (Anthropic, AWS Bedrock, Cloudflare) and why, and what the blended token cost is at the client's projected volume — the number that makes or breaks the business case.
• Proposal and estimation. Translating the architecture into effort, timeline, and cost that a buyer can approve.
• Delivery oversight. Enough involvement through the build to keep the shipped system faithful to the design, and to handle the inevitable "the model does X in production" surprises.

As-built: a cross-platform reference architecture, in the open

The solutions architect's core deliverable is a reference architecture with a defensible rationale — and this platform is one, built in public. The design decisions are the same ones you'd defend in a client workshop: the learning coach is a bounded agentic loop capped in src/lib/coach.ts at six model turns and eight tool calls so a runaway can't burn the budget; model selection is a routing decision in src/lib/llm.ts's modelFor(), cheap models for easy steps and strong ones for hard steps, with the per-token economics tracked in src/lib/cost.ts; and the platform choice spans Cloudflare, Anthropic, and AWS with a written reason for each.

It also carries the honesty a good solutions architect owes a client: the tradeoffs and the things that went wrong are documented, not hidden — including the time the AI Gateway's guardrails, in BLOCK mode, rejected the site's own prompt-injection lesson and had to be moved to FLAG mode. That "here's what we'd do differently" candour is exactly what earns technical trust in a client engagement. The full write-up is in the architecture notes.

AI solutions architect skills

The technical half is the standard AI-architecture skill set; the difference is the client-facing layer bolted on top:

• Agentic system design — the loop, tool use, MCP, retrieval, evals, the trust boundary.
• Model and cost economics — model selection and the blended-cost modelling that anchors a business case.
• Platform depth — at least one of Anthropic, AWS Bedrock, or Cloudflare to production standard, plus the cross-platform view that lets you justify a choice to a sceptical client.
• Safety and governance — the trust boundary, the OWASP-LLM risks, privacy and (in the EU) the EU AI Act — the concerns enterprise buyers ask about first.
• Scoping and estimation — turning a vague ask into requirements, architecture, effort, and cost.
• Communication and pre-sales — running a workshop, writing a proposal, and defending a design to both executives and engineers.

You do not need machine learning or model training for this — solutions architecture is about designing systems on top of existing models. Here's why ML isn't a prerequisite.

How to become an AI solutions architect

For a senior engineer or existing solutions/cloud architect, the transition is short — your scoping and stakeholder muscles already exist; you're adding the AI-native design layer:

• 1. Fundamentals. The shift to probabilistic systems, the agentic loop, prompt-as-spec.
• 2. Build a real agent. You can't scope what you can't build — stand up a tool-calling agent end to end.
• 3. Make it production-grade. Evals, cost-modelling, and the trust boundary — the parts a client will grill you on.
• 4. Design across platforms. Produce a cross-platform reference architecture and justify every choice, because "why not just OpenAI / why Bedrock" is the first client question.
• 5. Package it. A runnable system plus a one-page rationale and a rough cost model — the artifact that proves you can architect and sell the design.

Cloud and enterprise solutions architects have the smallest gap — your discovery, NFR, and estimation habits transfer directly. See the AI engineer roadmap for the technical sequence.

AI solutions architect certifications

There's no single "AI solutions architect" exam, but a few certs map to the skill set and carry weight with enterprise buyers:

• AWS Certified Generative AI Developer – Professional (AIP-C01) — agentic architecture, foundation-model selection, RAG, and governance on Bedrock; the closest architect-grade AWS GenAI exam.
• AWS Certified AI Practitioner (AIF-C01) — foundational; a useful on-ramp, not an architect credential.
• Anthropic Claude (CCAR-F) — Claude-specific building and prompting depth.

Treat certs as checkpoints that prove the profession, not the destination. For the full breakdown, see the AI certification exam guide and our verdicts on which certs are worth it.

AI solutions architect salary & job outlook

AI solutions architect sits among the higher-paid AI roles, and often above the in-house AI architect because the role carries commercial responsibility and, at vendors, sometimes a variable/commission component. US aggregators in 2026 put "AI architect" roughly in the $185k–$190k median band with a spread into the $260k+ range, and solutions-architect listings frequently higher again once total comp is counted. Treat these as directional — aggregator methodologies and titles vary widely.

On demand: growth is driven by every consultancy, systems integrator, and cloud provider needing people who can both design AI systems and sell the design to enterprise buyers moving from experiments to production. Fewer postings than "AI engineer" (the larger, more liquid market), but senior and well-paid. For the sourced picture see the AI skills split and where to find these jobs.

AI solutions architect courses & how to train

Most AI courses teach neither half of this job well — they start from ML theory you don't need and never touch scoping, cost-modelling, or defending a design. The faster route assumes your seniority and adds the AI-native architecture layer — agentic design, evals, cost and routing, the trust boundary, and deployment across Anthropic, AWS, and Cloudflare — and makes you produce a reference architecture with a written rationale, which is exactly the artifact a client engagement turns on.

That's how aiArch's curriculum is built: backward-designed from the job, every claim cited, and the platform you learn on is itself a production AI system built in public.

Frequently asked questions

What is an AI solutions architect?

An AI solutions architect designs production AI systems for clients and owns the customer-facing side: scoping the engagement, mapping requirements to an architecture, justifying platform and cost tradeoffs to stakeholders, and steering delivery. It's the AI architect role plus pre-sales and communication, most common at cloud providers, AI vendors, and their partners.

What's the difference between an AI solutions architect and an AI architect?

The technical design skills are identical. The solutions architect adds client-facing scoping, pre-sales, and stakeholder communication and usually works across accounts; the in-house AI architect points the same skills at one organisation's own product and its decision rationale.

Does an AI solutions architect need to code?

Yes, enough to stay credible. You need to be able to build and debug an agentic system — you can't scope or defend an architecture you couldn't build — but the role's centre of gravity is design, scoping, and communication rather than day-to-day implementation. You do not need machine learning.

How do you become an AI solutions architect?

If you're already a senior engineer or a solutions/cloud architect, add the AI-native layer: learn the fundamentals, build a real agent, make it production-grade with evals and cost and safety, design a cross-platform reference architecture, and package it with a rationale and a cost model. Your existing scoping and stakeholder skills transfer directly.

How much does an AI solutions architect make?

US aggregators in 2026 put architect-family medians roughly in the $185k–$190k range, spreading past $260k, with solutions-architect roles often higher once variable comp is counted. Figures are directional — methodologies and titles vary — but the role consistently sits among the better-paid AI positions because it's senior and carries commercial responsibility.

Is an AI solutions architect the same as a forward deployed engineer?

They're close cousins and both client-facing, but not the same. A forward deployed engineer embeds inside the customer and builds the solution hands-on in their environment; an AI solutions architect stays more on design, scoping, and advisory, usually across several accounts.