Box Architect will let a team describe a document process in plain language, then design the workflow, build the forms and agents it needs and suggest fixes once it runs. The catch: Box says Architect begins rolling out in beta to Enterprise Advanced customers in Q4, its top plan, and there is no public price yet.
Box announced it on 5 November at its BoxWorks conference, alongside agent security controls that teams can use today.
What Architect builds
You start with a goal, such as onboarding a client with complete paperwork. Box’s Architect product page says the tool proposes the steps, routing rules and policies. It then builds Box Forms for intake, metadata templates (structured fields attached to files), custom AI agents, Box Automate workflows and a Box App to track progress, with no code.
Before anything goes live, teams can test the workflow on real files and edge cases, preview agent outputs and routing, and add human review at decision points. Box says every step keeps the permissions, classifications and retention policies already set in Box.
Each step can use a different model. Box names Anthropic, OpenAI, Google, NVIDIA and IBM among the providers. After launch, Box says Architect will flag bottlenecks and recurring missing information and recommend changes your team accepts or skips.
Box’s BoxWorks announcement post says Architect will redesign a process “from the ground up” rather than copy the one your team runs today. The draft may not match how your people actually work, so the test stage matters.
What it costs and who gets it
Architect is not generally available, and Box’s press release warns that unreleased features “may not be delivered on time or at all.” Computerworld reported that pricing for Architect has not been disclosed. It added that Enterprise Advanced includes 20,000 AI Units, Box’s measure of AI usage.
Box’s pricing page shows no Enterprise Advanced price in its main plan table. A second comparison on the page lists Enterprise Advanced “starting at $70 per user/month”, billed annually, with a three-user minimum. We could not confirm Architect ships in that package, and extra AI units are sold on top.
The security controls you can use now
Three controls are generally available to Shield and Enterprise Advanced customers. They are classification-based policies that keep agents away from sensitive files, MCP Guardrails, and alerts when agent activity passes limits you set. MCP, the Model Context Protocol, is the open standard that lets outside AI assistants act inside apps like Box.
Box’s support note on MCP Guardrails lists the write actions an admin can restrict. They include uploads, copying and moving files and folders, adding collaborators and creating shared links. Guardrails are off by default and only on Enterprise Advanced. Reading files is governed separately by Shield access policies.
Assistants now reach company apps with less friction, as with Claude’s automatic connections for Team and Enterprise users, so these limits matter.
Where it can break
Box demonstrated Architect on stage at BoxWorks. Ron Miller wrote in his FastForward newsletter that workflows like Box’s examples take “a fair bit of human tweaking and back and forth” to reach a steady state. He added that model changes can break what worked. Box has not published error rates, and we have not tested Architect.
Jason Andersen of Moor Insights & Strategy said in the Computerworld report that he would still recommend human oversight while building agents. An agent takes actions instead of only answering, so a mistake lands in your files. Compare AWS letting agents run Step Functions workflows but not rewrite them. Architect is meant to build and revise workflows, with a person approving changes.
What to do this week
On Enterprise Advanced, turn on MCP Guardrails now, starting with shared links and external collaborators. Then write down one paper-heavy process with its steps and exceptions. That is your test case when the beta arrives. On a lower plan, ask Box what Architect will cost before you plan around it.
Sources
Get the weekly email
The week's most useful AI news, a tutorial and an interview.
Anna Jakovleva
Anna Jakovleva is the senior editor for AI tools at AI Observer, leading coverage of agents and automation, design and product, and the site's tutorials. She worked in operations and product teams before joining, so she judges tools by setup time, cost and where they break. Her writing is hands-on and practical.
Spotted a mistake? Tell us on the corrections page.
