Pete Carter

Gemini 4 Carbon is now number 1! Is this really a surprise? We are talking about Google. Never count them out. Although, OpenAI and and Anthropic will be making their next moves soon. Also, with the talent Meta has, wouldn't be surprised to see them at the top within the next year... Hell of a time for AI!

Multiple AIs. Multiple Computers. One Workflow. I go to bed. The work keeps moving. My setup uses three separate Claude accounts, one Codex account, and Grokbot across PCs, Macs, and cloud environments to help build a startup. One AI is building a feature. Another is building something else. They're reviewing each other's work, catching bugs, sharing findings, and figuring out what comes next. By morning, Grokbot has emailed me a briefing: what happened overnight, what passed review, what's blocked, and what needs me. They don't share one giant conversation. They share a workflow. The goal isn't to have one AI finish before another starts. It's to have them making progress at the same time without stepping on each other's work. Here's how you can build something similar. 1. GitHub: Give everyone a shared project Create a GitHub repository and connect your AI coding assistants. Assign each one a task it can work on independently. For example, Claude might build the user interface while Codex builds the backend. Before they start, agree on how the pieces will communicate and which files each assistant can change. Give each assistant its own branch and working directory. Git worktrees can provide separate working copies on the same computer, but you still need clear file ownership. Both assistants build. Both can review each other's work. 2. Google Drive: Create a shared inbox Create a shared project folder containing: START_HERE: Project goals, rules, responsibilities, and approval boundaries. CURRENT_STATE: What's completed, underway, blocked, and next. HANDOFFS: Assignments, questions, review findings, and replies. HANDOFFS/done: Messages that have been handled. I also have a separate Shared Brain. It preserves important decisions, preferences, project knowledge, and lessons learned across accounts and conversations. Each AI can contribute what it learns, and the others can build on that knowledge. Over time, the information stacks, giving them more context about me, my projects, and how I work. The inbox keeps the work moving. The brain keeps the knowledge growing. Connect each assistant through supported Google Drive integrations or synchronized local files. Check that it can read and write from the session you're actually using. Try this first: have Claude write a note, Codex read it and reply, and Claude read the reply. Check the files yourself. 3. Give the AIs the same operating rules Point each assistant to the shared folder and tell it: "Before working, read the project instructions, current status, and relevant handoffs. Check the Shared Brain for context relevant to your assignment. Confirm your task and which files you're responsible for. Don't edit another AI's assigned files. Coordinate dependencies through the shared inbox. Document changes, tests, blockers, and next steps. When assigned a review, send your findings directly to the responsible builder." That gives everyone the same starting point without you having to repeat the instructions in every conversation. 4. Build in parallel. Review each other's work. Claude and Codex can work on separate features simultaneously. When a change is ready, the builder commits and pushes it so another assistant can review the pull request. The reviewer documents problems. The original builder fixes them, and the updated change gets checked again before merging. Other independent assignments can keep moving. In one actual review, an AI caught typed text disappearing, incorrect upload feedback, and microphone cleanup issues. The builder corrected the problems and added tests. That's the kind of review I want from this setup. 5. Make every handoff actionable A handoff should identify who it's for, what task was completed, what changed, and what needs to happen next. Include the branch or pull request and exact revision, the checks run and their results, and anything that's blocked. The receiving AI reads the note, acknowledges it, and picks up the next action. Google Drive alone doesn't wake an AI. Start manually. Have one assistant write the handoff, then tell the receiving assistant to read it and act on it. Once that works, introduce supported scheduled checks or file watchers that notify or trigger the appropriate assistant. Test the full handoff before relying on it overnight. Local assistants need their computers and required apps or processes running. Cloud assistants need their own configured access and scheduling. 6. Give the team a coordinator As the setup grows, have a designated AI manager track assignments, progress, and shared status. I also use Grokbot as my assistant while I'm at my regular job. It requests updates, helps relay questions and decisions, contributes to reviews, and emails me progress reports. That morning briefing tells me what happened across the different computers and accounts, and where I need to step in. The team works within its assigned authority. Security approvals, important product decisions, unresolved disagreements, repeated review failures, and releases requiring my authorization still come back to me. Where to start You don't need my entire setup. Start with two AIs, one GitHub repository, and one shared Google Drive inbox. Assign two small, non-overlapping tasks, with a separate branch and working directory for each assistant. Have both build. Exchange reviews. Fix the findings. Test the handoffs manually before automating them. Then add more machines, accounts, scheduling, and reporting. For me, the value is having the AIs work together without constantly starting over or needing me to relay every message. Build the process first. Then scale it. The goal isn't to remove yourself from every decision. It's to stop being the bottleneck for every decision.