I'm on a Mission to 10x Myself. Here's My AI Stack.

I was listening to a podcast when a VC said something that hit me hard.

"Man, those AI-native founders," he said. "They are just built differently."

That sentence stuck with me. I've been building BenchSci for ten years. We've scaled from 4 people to 400 and raised over $200M. I build AI for a living. And in that moment I understood that I was probably way more behind than I thought.

Going deep on AI wasn't only about staying relevant. If I want to be a leader, a founder and CEO of tomorrow and not of yesterday, I have to rethink everything: how the company works, and how I work.

That moment kicked off a company-wide transformation at BenchSci. It also kicked off a personal one. I'm sharing the personal one here.

Today I'm all in on AI. Not as a talking point. As the way I work every single day.

I'm on a mission to 10x myself. Not 10% better. Ten times the meetings I'm ready for, the relationships I keep warm, the follow-ups that actually happen. The company needs more of me than there is, and AI is the only way I know to close that gap.

But this isn't only about my own output. We build AI for scientists. If I want us to build the best products for our customers, I can't understand AI from a slide deck or an analyst report. I have to immerse myself in it. I have to feel where it breaks, where it surprises me, and where it changes how work actually gets done.

The best and most forward-looking companies in the world are run by leaders who do exactly this. They use AI themselves, every day, and it shapes what they build. What I learn from running my own stack shows up in how we think about Emet and how we work with our customers.

Here's where I am today:

• 9 parts of my job rebuilt around AI
• 5+ core AI tools
• 50+ skills I've built
• 21 scheduled agents running on their own
• 13 data sources feeding all of it

Below is the whole stack: the tools, how I use each one, and why I focused on the parts of my job I did.

Before I'm a CEO, I'm a salesperson

Every startup chasing enterprise contracts has the same truth: the CEO's first job is sales. We sell to biopharma. We have zero customers in Canada, where we're based. Every customer is in the US or Europe.

So here's what my job actually looks like. For about eight months of the year, I'm on the road half the time. I spend roughly two-thirds of my time in meetings. I get 300 to 400 emails a day. I manage close to 500 relationships with customers, investors and partners.

That's a recipe for one thing: dropping the ball. When you do this many things this fast, you miss stuff. A follow-up slips. A commitment I made in a hallway at a conference disappears. Someone waits three weeks for a reply.

So I designed my whole stack around one principle: assume I'll drop the ball, and build the system that catches it. Every question I asked was some version of: what am I holding in my head right now that an agent could hold instead?

That's why I focused on nine areas:

1. Being on the go
2. Getting ready for meetings
3. Following up after them,
4. Having a AI chief of staff
5. My inbox
6. Knowing how the company is doing
7. Managing relationships
8. Thought leadership
9. and making the whole system improve itself.

Each one is either where my time goes or where the balls get dropped. Usually both.

The software, and why it's so good

My stack is deliberately small: a handful of tools, wired together, with Cowork at the center. Here's each one and why I love it.

Cowork: the brain. This is where everything comes together. Cowork is working as an agent on my desktop, connected to the 13 data sources I live in: Gmail, Google Calendar, Slack, Granola, Google Drive, Salesforce, Jira, Confluence and more. What makes it great isn't just the quality of the answers. It's three things on top. Skills are reusable playbooks I've written, more than 50 now, for things I do over and over: prepping for a meeting, drafting a board update, writing in my voice. Scheduled tasks run those skills on their own, so 21 agents work for me whether or not I open my laptop. Memory means it knows my people, our accounts and how I like things done, so I never repeat myself.

Granola: every call, captured. Granola takes notes on my approved video calls without a bot joining the meeting. The notes are good, and every conversation becomes data my agents can read. Most of my prep and follow-up agents start with Granola.

Plaud: Granola for the real world. A small recorder with excellent microphones for approved in-person meetings, conferences and dinners. Those are some of my most important conversations, and before Plaud they lived only in my head. Now they flow into Cowork like everything else.

Wispr Flow: talk instead of type. Voice dictation that works in any app: email, Slack, WhatsApp, Claude, Gemini. I talk the way I think, and it turns my rambling into clean written text. It's the fastest win in my stack, 3 to 5 hours back every week.

Gmail and Google Sheets: where the work lands. I didn't replace my inbox or buy a fancy CRM. My agents write into the tools I already use. Drafts wait for me in Gmail. My relationship list lives in a Google Sheet that Cowork reads and works from. No new app to check is a feature, not a compromise.

The stack, one job at a time

On the go. For about eight months a year I'm traveling half the time, so my phone is my office. Plaud records the in-person meetings. Wispr Flow lets me answer anything by talking: in a cab, at the airport, between meetings. It's why my messages have gotten longer lately.

Before every meeting. About 40 to 45 minutes before each meeting, an agent sends me a briefing on who I'm meeting and what I need to know. It pulls from my email, Granola, Slack, everything. I walk into every room knowing what needs to happen.

After every meeting. This is the one I love. Every day, an agent reads my Granola notes and my email, figures out every meeting I had, and drafts the follow-up: a thank-you, the action items, whatever the meeting needs. It's sitting in my drafts. I review it and hit send.

A chief of staff. Every morning I get an email that tells me what matters most today, what's happening across the company, and which commitments I made that I haven't followed up on yet.

Email. An agent reads every email I get and writes a reply. It's waiting in my drafts. Then it gets better: every time I edit a draft, rewrite it, or don't send it, the system learns from that and writes the next one more like me. To be clear, I read every email that goes out under my name. But starting from a draft in my voice instead of a blank page saves me about 80% of the time.

Briefings. I've built a set of artifacts that track the company, our deals, Emet, and the market and competitors. An agent reads across all of them and sends me daily and weekly briefings so I always know where we stand.

500 relationships, still personal

Every new person I meet in the industry gets added to a Google Sheet. My EA helps me a lot here. On top of that sheet, I built a CRM inside Claude Cowork. It's at almost 500 people: customers, partners, investors, everyone.

For each person it knows their company, where they're based, when we last spoke, every email we've exchanged, and anything new published about them. It also has a page that tells me who I should talk to next and why.

Where it really earns its keep is travel. Before a trip I ask: who's in Boston? San Francisco? San Diego? Europe? When I publish a new post, I ask: which 30 people should get this? Then I email them straight from the system. That's how I do things at scale and still keep them personal.

The agents that catch what I drop

This is the principle at its purest. I assume I'm going to miss things, so I built agents whose only job is to find what I missed. One tells me everything I might have dropped and every commitment I made but haven't followed through on. Another tells me who I emailed that never replied.

I don't trust myself to remember everything. I shouldn't have to.

Writing in 10 minutes

I write and post a lot, because thought leadership is one of the ways BenchSci wins. The ideas are mine. The drafting isn't. I trained agents on everything I've ever written, and built others modeled on the frameworks of expert marketers who literally wrote the books. I don't spend more than 10 minutes on a full blog post, or more than 5 on a LinkedIn post. This blog post followed the same process.

A system that gets better on its own

Under all of this sits memory. The agents learn from what I accept, what I edit and what I throw away, so the whole system gets a little better every week. On top of that run the 21 scheduled agents, each firing daily or weekly, so I don't have to remember to ask.

None of these pieces is impressive on its own. Five minutes saved here, one minute there. But they happen hundreds of times a week, and they compound into many hours. That's the part people underestimate. The value isn't in any single agent. It's in the compounding.

This might all be irrelevant in six months

I'm under no illusion that this stack is permanent. The tools change every month. A new model will make some of my skills unnecessary. A new product will replace a tool I love. Six months from now, much of what I just described might be irrelevant.

I'm assuming that will happen, and I'm fine with it. The specific stack matters less than the habit behind it: keep asking what I'm still carrying by hand, hand it off, and rebuild when something better shows up. If I'm still running the exact same setup next spring, I've fallen behind.

The stakes are real. Companies that use AI this aggressively will beat the ones that don't. The ones already doing it are the ones we have to keep up with. And some of the best ideas for Emet come from watching what companies completely outside our space are doing with AI. You only see those things if you're in it yourself.

The question isn't what AI can do for you. It's what you're still carrying in your head that you don't need to.