How I Built an AI Second Brain That Actually Gets Work Done

A few months ago, I was about to send an email to one of our partners when my AI caught a bad link. I hadn't asked it to do that, but it noticed the mistake, found the right one, updated the draft and flagged the change for my review.

Small save. But it was the first real sign that what I'd been building was working: an AI system that understands my work well enough to know when something needs attention, without being told to look.

Most people are still using AI like a search engine. You ask a question, get an answer, move on. But I wanted a system that continuously evaluates what's happening across Forward Future, connects the dots between applications and starts useful work on its own.

I call it my second brain.

It's built around Grok Bot, which serves as both the interface and the orchestrator. It remembers relationships and decisions. It preps me for meetings, tracks commitments, watches business performance and coordinates a team of specialized agents. Some routines run on a schedule, and others fire when new information arrives. I can kick off work just by talking to it.

If you don’t consider yourself to be an advanced AI user, here's the part I most want you to hear: This all started with a single email routine. If you can explain how you triage your inbox to a new employee, you can build version one.

Here's how it works, how I built it and what it took to make it reliable.

I didn't build the whole thing at once. When I started experimenting with Grok Bot in August 2026, I gave it exactly one job: reviewing my email every morning. Version one connected Gmail, HubSpot and Google Drive. It read incoming messages, pulled the relevant relationship and business context and helped me decide what deserved my attention.

Say a sponsorship email references a company we negotiated with six months ago. The old way: dig through old threads, check the CRM, hunt down the latest proposal. I wanted the agent to do that digging for me.

Then I added sources, one at a time. Fathom came in through its API so the system could pull new meeting transcripts. Asana, Calendar and Slack followed, which let it connect conversations to the work those conversations created. Eventually I brought in our audience and revenue platforms: Beehiiv, YouTube and Stripe.

Today, the system checks for important emails and new meeting transcripts approximately every 30 minutes during the workday. Other workflows respond to changes in connected systems, such as new meeting transcripts, HubSpot opportunities changing stages or overdue Asana tasks, using a combination of scheduled checks and event triggers.

You don't need all those integrations to start. Here's how to build a useful first version.

If I were starting over, I'd begin with the same workflow that started my second brain: an assistant that reviews my inbox and prepares a prioritized briefing before I start work.

Here's how to build it using Grok Bot.

Step 1: Connect your email

Open Grok Bot, select Connect apps and connect Gmail. Start with read-only access. You can add HubSpot or other sources later.

Step 2: Give it instructions

Open your Bot and paste this prompt, adjusting the priorities to fit your work:

Step 3: Test and refine

Ask Grok Bot to run the briefing immediately. Check whether it caught the important messages and excluded unnecessary noise.

My early briefings summarized too much. I improved them with feedback like:

Run it again and compare. A few rounds of refinement can make a big difference.

Step 4: Automate it

Once you're satisfied, tell Grok Bot:

Confirm the schedule and use Test to verify the routine works.

Step 5: Add context

After a few days, connect HubSpot or another relevant source. Now the assistant can recognize that a seemingly routine email relates to an active negotiation or outstanding commitment.

That's when it becomes genuinely useful: instead of simply reporting what's new, it starts identifying what matters.

The goal: A two-minute daily briefing with your most important priorities, supporting context and recommended next steps.

Chat history is fine for remembering that you prefer bullet points to long sentences. It's less reliable for precise business commitments, changing commercial terms and information that needs to stay accurate over time.

So I built a more structured memory system, using Convex as the primary database. It organizes Forward Future's information into records for people, companies, projects, conversations, agreements, decisions and commitments, and tracks the relationships between them.

Those relationships can get messy fast. One company can be a sponsor, an interview guest, a technology partner and the maker of a product we're considering covering. Sometimes it's all of those at once. The system has to recognize those connections while keeping commercial relationships firmly separate from editorial decisions.

Every important record includes relevant history, associated documents, outstanding responsibilities and links to its original sources. The system updates those records as new information arrives, distinguishing confirmed decisions from proposals, assumptions and outdated information. That way, a tentative sponsorship date doesn't carry the same authority as a signed agreement.