Running an NFL Weather App With a Team of AI Agents

Testing the tools on work he knows cold showed Brendan where they're sharp and where they miss. He carries that read into areas he can't judge himself, like code and weather forecasts. And where he can't judge a claim, he checks the evidence instead of trusting the source: he reviews every PR, and the site keeps a public report card. That's what lets a partnerships lead run an engineering bot.

Why Grok Bot

Brendan has built a couple dozen apps with tools like Replit and Lovable. For Game Day Weather he moved to Grok Bot, xAI's agent product, because it's chat-based ("it kind of looks like iMessage") and lets him run a team instead of a single assistant. Each bot gets its own virtual machine and email address. "It's like my own little Ocean's Eleven," he says. "You're hiring a group of bots to manage this for you."

The crew

• Sideline, the weather researcher. Pulls local forecasts from several sources, blends them, and grades every game card FLAG, WATCH, CLEAR or DOME.

• Cursor Ops, the engineer. Works inside Cursor, runs a preflight check on each chip, pushes updates to the live site, and opens small PRs for UI changes.

• Chief of Staff. Coordinates between research and engineering, then proofs and sends.

• Comms. Handles the Buttondown newsletter and the @getGDW posts on X.

Brendan's job is the one he describes as "idea guy." He gives one directive, watches the team work it out in a group chat, and the bots "raise their hand" when a pull request is ready for him to review and merge on GitHub. Merge PRs and pick ideas. That's the whole human role.

The bots weigh in on product decisions too. When Brendan asked whether to add branded MCP UI cards to the app's AI-facing results, Sideline recommended a small test on one tool while keeping plain data as the main answer, since many agents only read text. Cursor Ops went off to check which chat apps actually render it before answering.

The weekly loop

The team runs on the NFL clock, with no out-of-cycle weather asks:

1. Research. Pull kickoff-window weather for the Thursday, Saturday international, Sunday and Monday slates.

2. Board. Grade every game and write the notes.

3. Live. Preflight, ship to the site.

4. Post. Newsletter and social, then on to the next clock.

It also grades itself. The site keeps a public report card, and Week 5 came in at 9 hits, 1 partial and 0 misses.

The takeaway

"This has helped me learn how to build and develop an agentic-led team," Brendan says. A few weeks of shipping taught him how the team improves, and how he can work better inside that structure.

You don't need to write the code to run the team. You need a clear directive, a loop, and a check on what you can't verify yourself.

See it live at gamedayweather.app. The Boston Builder Series highlights people across Boston building with AI, from @MGMT Boston in partnership with @Trove.