What the studio built
Light Sail VR first built the coordinator on a Framework desktop with an AMD Ryzen AI Max processor and 128GB of unified memory. It later added an AMD Radeon PRO R9700 graphics card to improve performance. AMD's own page uses the fuller product name, Radeon AI PRO R9700. AMD quotes the studio saying the speed increases from the card were "phenomenal." No model, throughput figure or test conditions are given for that claim.
The AMD page says the agent is built to keep project timelines aligned. The studio set out to build an agentic production coordinator that could keep project timelines aligned and moving. The Tribune reports these functions:
- Tracking deadlines, schedules, documents and stakeholders, and sending updates to staff through Slack.
- Revising production timelines.
- Updating Google Sheets documents.
- Preparing emails from instructions sent through Slack.
AMD gives a concrete example of the handoff it replaces. Client feedback arriving through platforms like Frame.io used to be moved into Google Docs, with schedule adjustments made individually. Now founder and CEO Matthew Celia can message the agent from Slack and ask it to move due dates forward or backward.
Why local AI
The Tribune says Celia turned to local AI because of cost and privacy concerns with externally hosted models. It reports that the system runs locally on Ryzen AI processors and Radeon graphics. It says this helps protect confidential client information and reduces cloud AI processing expenses. AMD's account adds that the studio works with Hollywood studios and technology companies under strict secrecy and NDA requirements.
The Tribune quotes Celia saying he is not comfortable sending client information outside "these four walls."
These are the studio's stated reasons. The available reporting does not include a cost comparison between cloud and local operation. It also has no security assessment, so privacy here is a goal rather than a verified guarantee. It does not say whether every part of the workflow stays on local hardware. Slack, Google Sheets and email are third-party services, so some data presumably still passes through them. That is my inference, not something the sources state.
The productivity claim
Celia told the Tribune: "Having the AMD computer allows me to handle six, seven, eight, or nine jobs with the same amount of personal effort." Previously, he could personally oversee only one or two projects at a time.
Treat this as a self-reported figure. The sources do not define a "job" or give a measurement period. They also include no independent before-and-after evaluation. The case study comes from AMD, which sells the hardware, so it is marketing material.
What the sources don't say
Several details that IT administrators would want are missing:
- The software stack and the models used.
- The permissions the agent holds in Slack, Google Workspace or email.
- Whether emails and document changes need human approval before they go out.
- How errors are handled.
- How many R9700 cards are actually deployed. AMD's narrative describes adding one card, but its summary reportedly refers to cards in the plural.
These matter for anyone copying the setup. An agent that can edit schedules and prepare emails needs tight access controls, an audit trail and clear rules on when a person signs off. None of these are described.
What is only planned
The Tribune says the studio is exploring more agents. They would help review interview footage, analyze story structure, monitor social media performance and identify comments that need moderation. None of these are reported as deployed. The agent that exists today does administrative project coordination. It does not edit video or analyze footage.
Why PC and IT readers should care
Independent testing shows the R9700 is a credible local-inference card, which makes the studio's choice plausible. It has 32GB of VRAM, and Puget Systems' dual-card tests found 8B-class models running at production-usable speeds. Back2Gaming's review noted a practical advantage over 16GB cards. A 27B model with a long context fit entirely in the R9700's memory, while two 16GB cards had to push some layers onto the CPU.
Those tests are not Light Sail VR's workload, so they are context rather than proof. They do suggest why a 32GB card paired with a large unified-memory desktop is a reasonable platform for small local agents.
The takeaway is narrow. A small studio says a Slack-driven agent on workstation-class AMD hardware has taken over specific administrative tasks and kept its work off external model services. That is worth watching for small businesses with confidentiality obligations. It needs independent measurement and a clear account of safeguards before anyone treats it as a template.