■ MODERATE RISK ■ Technology
Not wholesale, but the documentation-heavy half of the role thins out considerably. What survives is the part that happens in rooms with people who disagree about what the system is actually for.
“Connecting business problems to technical solutions still requires a human who can talk to both sides.”
Our AI replacement risk score — how we score jobs
Solutions architects sit between a customer's messy requirements and a technical design that can be built and paid for. The week involves discovery workshops, drawing reference architectures, sizing and costing a cloud deployment, writing statements of work and design documents, running proofs of concept, answering security questionnaires, and translating the same idea three different ways for an engineering lead, a procurement officer, and a CIO who wants to know why it costs that much.
A large slice of that is document production, and document production is where models excel. AI now drafts architecture documents, generates infrastructure diagrams and Terraform from a described design, produces cost comparisons across cloud providers, fills in security questionnaires from a knowledge base, and turns meeting notes into a solution outline. Cloud vendors ship well-architected review tooling and AI advisors that recommend patterns directly. Standard integrations — a fairly typical data platform, a fairly typical migration — increasingly come as templated blueprints rather than bespoke thinking. Junior architects who mainly assembled slides and diagrams are the exposed group.
The resistant part is political and diagnostic. Customers rarely know what they need; they present a stated requirement that conflicts with their actual constraints, their existing vendor contracts, their team's real capabilities, and internal factions who want different outcomes. Extracting the true problem takes reading a room, asking uncomfortable questions, and building enough trust that someone admits the legacy system nobody documented is load-bearing. Then the architect has to make tradeoffs — cost against latency, speed against maintainability — and defend them when a stakeholder pushes back. Trust also gets sold: enterprise buyers want a person who will still be reachable when the migration goes sideways. Our 45 reflects a role where the artifacts automate but the relationships don't.
Automatability: our editorial assessment of current and near-term AI capability
Documentation and design-artifact work is being automated now, and by roughly 2030 producing a competent reference architecture will be a prompt rather than a week. Expect fewer junior architect roles and more expectation that a senior architect covers more accounts with AI leverage. The consultative and trust-based portion of the job holds up well into the 2040s, particularly in enterprise and regulated sales where accountability is part of the purchase.
It replaces a good portion of what they produce, not what they do. Reference architectures, cost models, and design documents are increasingly generated in minutes. The job survives because customers arrive with the wrong problem statement, competing internal agendas, and undocumented systems — and untangling that requires someone in the room asking questions the customer didn't expect.
Yes, if you're moving toward the customer-facing end rather than the diagram-producing end. The architects doing well are the ones who own relationships, understand a specific industry deeply, and can arbitrate between stakeholders. Those who defined their value as knowing cloud service catalogues are competing directly with tooling that knows the catalogue better.
Anything that ends in a document. Proposals, design docs, architecture diagrams, infrastructure code from a described design, cost comparisons, and security questionnaire responses are all being generated from prompts and knowledge bases. The practical effect is that the artifact stops being the deliverable and becomes a by-product of the conversation that produced it.
By being the person who knows what the customer actually needs, which is usually different from what they asked for. Invest in discovery skill, industry-specific fluency, and credibility. Then use AI aggressively for the artifacts so you can cover more accounts and spend more time on the parts of the engagement where judgment and trust decide the outcome.