The AECdesign.ai blog
Plain-English guides to AI review workflows, source control, tool setup, security, and where AEC firms fit. Deeper than the FAQ; always with the engineer of record in control.
The document tax is where AI review has to prove itself
Recent construction AI coverage points to RFIs, submittals, daily reports, and document control as the practical proving ground for AI review workflows.
Document control is the first AI review control
Recent construction AI and document-control guidance points to the same conclusion: AI review is only useful when source status, versions, and reviewer authority are controlled.
Construction AI should support the reviewer, not replace the reviewer
Recent construction AI discussion is converging on a practical boundary: use AI for document-heavy support work, but keep issued decisions with qualified reviewers.
Jobsite AI still needs a defined review lane
Practical construction AI is moving closer to jobsites, submittals, and project records. That makes source control and EOR review boundaries more important, not less.
Claude policy updates are a reminder to define the review boundary
Anthropic's June privacy-policy update and MCP security guidance point to the same rule for AEC firms: define access, local files, and review authority before production use.
AI review belongs in the project record, not outside it
Recent construction AI signals point to a practical rule: use AI where RFIs, submittals, and project records already live, with the engineer of record still in control.
Construction's AI money is flowing into submittal and RFI review
A $13.8M seed round and a wave of new tools point AI first at submittal and RFI review — what it does well, and the engineer-of-record line it can't cross.
The state of AI in AEC — and where MCP fits
Where AEC actually stands on AI in mid-2026 — early and cautious — and how a standard-based, human-reviewed tool fits a careful industry.
What legal AI's MCP moment signals for AEC
Major legal vendors wired into Claude via MCP, framed around accountability rather than automation. Why that's a direct preview for AEC.
Agentic AI and tool use, explained
Tool use, the autonomy spectrum, and why a focused, human-adjudicated review tool is a deliberate choice for licensed work.
The open MCP ecosystem and avoiding lock-in
Thousands of servers and tens of millions of downloads — and why an open standard protects AEC buyers from vendor lock-in.
MCP security: prompt injection and tool poisoning
The lethal trifecta, tool poisoning, and what to check before connecting any AI tool to a firm's project data.
Server-side tools vs. client sampling
MCP primitives and where the work runs — and why server-side logic protects a firm's method and data.
How remote MCP servers authenticate
A remote server needs a lock. Since March 2025 it's been OAuth — how it works and why it matters for firm data.
Remote vs. local MCP servers
Local vs. remote MCP servers: transports, maintenance, and why AECdesign.ai runs remote.
Why MCP became the standard for AI tools
How MCP went from a quiet release to the default AI-tooling standard — and why that matters for AEC.